[go: up one dir, main page]

CN118885818A - Soil monitoring data analysis method, device and equipment for mine ecological restoration - Google Patents

Soil monitoring data analysis method, device and equipment for mine ecological restoration Download PDF

Info

Publication number
CN118885818A
CN118885818A CN202411366744.5A CN202411366744A CN118885818A CN 118885818 A CN118885818 A CN 118885818A CN 202411366744 A CN202411366744 A CN 202411366744A CN 118885818 A CN118885818 A CN 118885818A
Authority
CN
China
Prior art keywords
training
data
generator
model
soil monitoring
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202411366744.5A
Other languages
Chinese (zh)
Other versions
CN118885818B (en
Inventor
王光凯
孙玉林
韩刚
刘波
何东林
魏新力
白云
郭瑞
于乐
李文悦
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shandong Zhengyuan Yeda Environmental Technology Co ltd
Original Assignee
Shandong Zhengyuan Yeda Environmental Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shandong Zhengyuan Yeda Environmental Technology Co ltd filed Critical Shandong Zhengyuan Yeda Environmental Technology Co ltd
Priority to CN202411366744.5A priority Critical patent/CN118885818B/en
Publication of CN118885818A publication Critical patent/CN118885818A/en
Application granted granted Critical
Publication of CN118885818B publication Critical patent/CN118885818B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
    • G01N33/24—Earth materials
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/211—Selection of the most significant subset of features
    • G06F18/2111—Selection of the most significant subset of features by using evolutionary computational techniques, e.g. genetic algorithms
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2136—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on sparsity criteria, e.g. with an overcomplete basis
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/2431—Multiple classes
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/24323—Tree-organised classifiers
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/29—Graphical models, e.g. Bayesian networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N10/00—Quantum computing, i.e. information processing based on quantum-mechanical phenomena
    • G06N10/60—Quantum algorithms, e.g. based on quantum optimisation, quantum Fourier or Hadamard transforms
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G06N3/0455—Auto-encoder networks; Encoder-decoder networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0475—Generative networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/094—Adversarial learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/098—Distributed learning, e.g. federated learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L63/00—Network architectures or network communication protocols for network security
    • H04L63/04—Network architectures or network communication protocols for network security for providing a confidential data exchange among entities communicating through data packet networks
    • H04L63/0428—Network architectures or network communication protocols for network security for providing a confidential data exchange among entities communicating through data packet networks wherein the data content is protected, e.g. by encrypting or encapsulating the payload
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04Q—SELECTING
    • H04Q9/00—Arrangements in telecontrol or telemetry systems for selectively calling a substation from a main station, in which substation desired apparatus is selected for applying a control signal thereto or for obtaining measured values therefrom
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04Q—SELECTING
    • H04Q2209/00—Arrangements in telecontrol or telemetry systems
    • H04Q2209/80—Arrangements in the sub-station, i.e. sensing device
    • H04Q2209/84—Measuring functions

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Computer Security & Cryptography (AREA)
  • Chemical & Material Sciences (AREA)
  • Physiology (AREA)
  • Analytical Chemistry (AREA)
  • Signal Processing (AREA)
  • Computer Hardware Design (AREA)
  • Pathology (AREA)
  • Immunology (AREA)
  • Biochemistry (AREA)
  • Medicinal Chemistry (AREA)
  • Food Science & Technology (AREA)
  • Remote Sensing (AREA)
  • Geology (AREA)
  • General Life Sciences & Earth Sciences (AREA)
  • Computational Mathematics (AREA)
  • Condensed Matter Physics & Semiconductors (AREA)
  • Mathematical Analysis (AREA)

Abstract

The invention provides a soil monitoring data analysis method, a device and equipment for mine ecological restoration, which relate to the technical field of data processing and comprise the following steps: acquiring soil monitoring data; inputting the soil monitoring data into a pre-trained soil monitoring data analysis model, and outputting a soil monitoring data analysis result; the training mode of the pre-trained soil monitoring data analysis model comprises the following steps: the method comprises the steps that a plurality of local models are respectively trained based on a training data set and then summarized to a global model of a central server for parameter updating, and the plurality of local models are subjected to global iteration based on updated parameters until training converges so as to complete training; the training data set acquisition step comprises the step of carrying out data expansion by utilizing a generating countermeasure network based on the decomposition of the variable decibel-leaf sparse matrix. The method reduces the cost of data transmission and storage, avoids centralized and transmission of data, effectively reduces the risk of data leakage, and ensures comprehensive utilization of the data.

Description

Soil monitoring data analysis method, device and equipment for mine ecological restoration
Technical Field
The invention relates to the technical field of data processing, in particular to a soil monitoring data analysis method, device and equipment for mine ecological restoration.
Background
Mining activities often have serious environmental impact, resulting in soil degradation, pollution and ecological system destruction. In order to effectively perform ecological restoration, the affected soil needs to be monitored and evaluated in order to formulate an appropriate restoration strategy. However, conventional soil monitoring methods face challenges such as data collection difficulties, data privacy and security issues, and inefficient data processing. In addition, due to the complexity and diversity of soil conditions, efficient data analysis methods are needed to process and parse large amounts of soil monitoring data to ensure the scientificity and effectiveness of the remediation measures.
In the related art, the soil monitoring data of multiple sites usually need to be centralized, which not only increases the cost of data transmission and storage, but also brings risks of data privacy and security, lacks effective data privacy protection measures, and limits the comprehensive utilization of the data.
Disclosure of Invention
In view of the above, the invention aims to provide a soil monitoring data analysis method, a device and equipment for mine ecological restoration, which solve the problems of data privacy and data safety when soil monitoring is carried out at multiple places, reduce the cost of data transmission and storage, avoid the central concentration and transmission of data by respectively carrying out model training and parameter updating on each local model, effectively reduce the risk of data leakage and ensure the comprehensive utilization of the data.
In a first aspect, an embodiment of the present invention provides a method for analyzing soil monitoring data for ecological restoration of a mine, where the method includes: acquiring soil monitoring data; inputting the soil monitoring data into a pre-trained soil monitoring data analysis model, and outputting a soil monitoring data analysis result; the training mode of the pre-trained soil monitoring data analysis model is as follows: training a plurality of local models based on the training data set to obtain respective model parameters; based on the model parameters of each of the local models, carrying out parameter updating by using a global model of a central server to obtain target model parameters; transmitting the target model parameters to a plurality of local models for global iteration until training converges, and obtaining a pre-trained soil monitoring data analysis model; the training data set is obtained based on labeling, data expansion and feature extraction of a plurality of collected soil sample data; the data expansion method comprises the following steps: sample generation is carried out on the countermeasure network by utilizing generation based on the variable decibel leaf sparse matrix decomposition so as to realize data expansion; the generating an countermeasure network based on the variational Bayesian sparse matrix decomposition comprises: a generator and a arbiter; the method for generating samples by using the generation countermeasure network based on the variational Bayesian sparse matrix decomposition comprises the following steps: initializing network parameters of a setting generator and a discriminator; performing corresponding adjustment by using the self-attention layer; optimizing sparse matrix parameters in a generator by a variable decibel leaf method, and setting the sparse matrix of the generator asThe update rule of the sparse matrix is as follows: ; wherein, For the updated sparse matrix of the generator,Is a sparsity adjustment parameter; Is the learning rate of the sparse matrix; Is a loss function that generates an antagonism network; is to generate a loss function of the countermeasure network Is a gradient of (2); A generator function; after the sparse matrix learning is completed, performing countermeasure training of the generator and the discriminator; in the training process of the generator, the sample generation process of the generator is enhanced through the updated parameters of the self-attention layer, and optimized through the updated sparse matrix of the generator, and the sample generation process is expressed as the following formula: ; wherein, Representing the bias of the generator; is distributed from a priori Random noise extracted from the sample; The function is activated for Sigmoid.
In a preferred embodiment of the present invention, the training manner of the local model is: labeling the collected multiple soil sample data to obtain a first training data set; performing data expansion on the first training data set to obtain a second training data set; extracting features of the second training data set to obtain a third training data set; the local model is trained based on the third training dataset.
In a preferred embodiment of the present invention, labeling the collected plurality of soil sample data to obtain the first training data set includes: labeling a plurality of soil sample data based on preset data categories to obtain a first training data set; wherein, the data category includes: is suitable for repairing, needs further improvement and serious pollution.
In a preferred embodiment of the present invention, the performing data expansion on the first training data set to obtain a second training data set includes: sample generation is carried out on the basis of the first training data set by utilizing a generating countermeasure network based on the decomposition of the variable decibel-based sparse matrix so as to realize data expansion to obtain a second training data set; the loss functions of the generator and the arbiter are expressed by the following formulas:；
Wherein, Is represented as inputAndGenerating a loss function against the networkUnder the constraint of (1) the generator minimizes the loss function and the arbiter maximizes the loss function; representing compliance with a particular distribution; Representing a sample of the real data, Representing the distribution from a prioriRandom noise extracted from the sample; Representing a priori distribution of the real data samples; representing a priori distribution of random noise; Representing the desire; the function of the arbiter is represented as such, Representing the generator function.
In a preferred embodiment of the present invention, the feature extraction of the second training data set to obtain a third training data set includes: performing feature extraction based on the second training data set by using a self-evolution genetic algorithm to obtain a third training data set; the training mode based on the self-evolution genetic algorithm comprises the following steps: randomly generating a group of neural network populations; wherein, each neural network has a weight and a structural parameter initialized randomly; performing adaptability evaluation on each neural network, and selecting the neural network which meets the preset standard based on the result of the adaptability evaluation; performing crossover and mutation operations on the selected neural network to generate a new neural network population; dynamically adjusting parameters of an activation function of a new neural network based on the feature distribution of the current data to realize feature extraction to obtain a third training data set; wherein the operation of mutation is achieved by adding small random perturbations, expressed by the following formula:， ; wherein, Is the weight of the neural network after the mutation,Is the bias of the neural network after the mutation,Is the variation rate of the product, and the product is a modified product,Is the variance of the disturbance.
In a preferred embodiment of the present invention, the feature extraction of the second training data set to obtain a third training data set includes: performing feature extraction by using a self-encoder algorithm based on depth feature remodeling based on the second training data set to obtain a third training data set; the training mode of the self-encoder algorithm based on depth feature remodeling comprises the following steps: initializing parameters of a self-encoder; the self-encoder includes an encoder and a decoder; in the process of encoder training, input data is compressed into low-dimensional characteristic representation through forward propagation, and a neural network abstracts and transforms the input data layer by layer to extract corresponding characteristic information; during the decoder training process, the decoder receives the output of the encoder, reconstructing the original input; co-optimizing using error feedback of the encoder and decoder; wherein the forward propagation process is expressed as the following equation: ; wherein, Is an activation function of the self-encoder; is an input high-dimensional data vector; Is the encoded low-dimensional feature vector; The normal distribution which represents the mean value of 0 and the covariance matrix of the unit matrix is used for generating noise; Is the noise intensity; wherein the activation function of the self-encoder adapts to the characteristic change of the training data in a dynamic adjustment mode, and the input of the activation function is set Is thatThe activation function calculation method of the self-encoder is expressed as the following expression:， ; wherein, Is a coefficient dynamically adjusted according to the data variance; Is a parameter for adjusting sensitivity; is the variance of the input data; is the threshold of variance; is a function of the activation of the ReLU, Is a Sigmoid activation function.
In a preferred embodiment of the present invention, training the local model based on the third training data set comprises: training the local model by utilizing a quantum state depth neural decision tree algorithm based on conflict driving based on a third training data set; the training mode of the quantum state depth neural decision tree algorithm based on conflict driving comprises the following steps: initializing all decision nodes of a neural decision tree into a quantum state; gradually adding decision nodes from the root node of the neural decision tree; evaluating and obtaining a plurality of segmentation conditions based on each decision node, and selecting the segmentation conditions for maximizing the information gain; at each decision node of the neural decision tree, the quantum state of the decision node is adjusted according to the path of the data passing through the decision node; in the construction process of the neural decision tree, detecting classification conflicts, and solving the classification conflicts by adjusting quantum states of relevant decision nodes; pruning is carried out on the built neural decision tree, and training of the local model is completed after pruning is completed.
In a preferred embodiment of the present invention, the objective function of the pre-trained soil monitoring data analysis model is represented by the following equation:， ; wherein, Analyzing an objective function of the model for pre-trained soil monitoring data,As a parameter of the model, it is possible to provide,Is a model parameterFor the firstA loss prediction of the data of the individual samples,In order to train the loss of the training,Is the firstSample characteristics of the individual sample data,First, theA tag for each sample data; the objective function of the local model is represented by the following equation: ; wherein, Is the firstThe number of samples of the individual local model,Is the firstThe objective function of the individual local model is,Is the firstData distribution of individual nodes.
In a second aspect, an embodiment of the present invention further provides a soil monitoring data analysis device for mine ecological restoration, including: the soil monitoring data acquisition module is used for acquiring soil monitoring data; the soil monitoring data analysis module is used for inputting the soil monitoring data into a pre-trained soil monitoring data analysis model and outputting a soil monitoring data analysis result; the training mode of the pre-trained soil monitoring data analysis model is as follows: training a plurality of local models based on the training data set to obtain respective model parameters; based on the model parameters of each of the local models, carrying out parameter updating by using a global model of a central server to obtain target model parameters; transmitting the target model parameters to a plurality of local models for global iteration until training converges, and obtaining a pre-trained soil monitoring data analysis model; the training data set is obtained based on labeling, data expansion and feature extraction of a plurality of collected soil sample data; the data expansion method comprises the following steps: sample generation is carried out on the countermeasure network by utilizing generation based on the variable decibel leaf sparse matrix decomposition so as to realize data expansion; the generating an countermeasure network based on the variational Bayesian sparse matrix decomposition comprises: a generator and a arbiter; the method for generating samples by using the generation countermeasure network based on the variational Bayesian sparse matrix decomposition comprises the following steps: initializing network parameters of a setting generator and a discriminator; performing corresponding adjustment by using the self-attention layer; optimizing sparse matrix parameters in a generator by a variable decibel leaf method, and setting the sparse matrix of the generator asThe update rule of the sparse matrix is as follows: ; wherein, For the updated sparse matrix of the generator,Is a sparsity adjustment parameter; Is the learning rate of the sparse matrix; Is a loss function that generates an antagonism network; is to generate a loss function of the countermeasure network Is a gradient of (2); A generator function; after the sparse matrix learning is completed, performing countermeasure training of the generator and the discriminator; in the training process of the generator, the sample generation process of the generator is enhanced through the updated parameters of the self-attention layer, and optimized through the updated sparse matrix of the generator, and the sample generation process is expressed as the following formula: ; wherein, Representing the bias of the generator; is distributed from a priori Random noise extracted from the sample; The function is activated for Sigmoid.
In a third aspect, an embodiment of the present invention further provides an electronic device, including a processor and a memory, where the memory stores computer executable instructions executable by the processor, and the processor executes the computer executable instructions to implement the soil monitoring data analysis method for mine ecological restoration according to the first aspect.
The embodiment of the invention has the following beneficial effects:
The embodiment of the invention provides a soil monitoring data analysis method, a device and equipment for mine ecological restoration, which are used for inputting soil monitoring data into a pre-trained soil monitoring data analysis model and outputting a soil monitoring data analysis result by acquiring the soil monitoring data. The training mode of the pre-trained soil monitoring data analysis model is as follows: training a plurality of local models based on the training data set to obtain respective model parameters; based on the model parameters of each of the local models, carrying out parameter updating by using a global model of a central server to obtain target model parameters; transmitting target model parameters to a plurality of local models for global iteration until training is converged, obtaining a pre-trained soil monitoring data analysis model, and obtaining a training data set based on labeling, data expansion and feature extraction of a plurality of collected soil sample data; the data expansion method comprises the following steps: sample generation is carried out on the countermeasure network by utilizing generation based on the variable decibel leaf sparse matrix decomposition so as to realize data expansion. In the mode, the problems of data privacy and data safety during soil monitoring at multiple sites are solved, the cost of data transmission and storage is reduced, the central concentration and transmission of data are avoided by respectively carrying out model training and parameter updating on each local model, the risk of data leakage is effectively reduced, and the comprehensive utilization of the data is ensured.
Additional features and advantages of the disclosure will be set forth in the description which follows, or in part will be obvious from the description, or may be learned by practice of the techniques of the disclosure.
The foregoing objects, features and advantages of the disclosure will be more readily apparent from the following detailed description of the preferred embodiments taken in conjunction with the accompanying drawings.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are needed in the description of the embodiments or the prior art will be briefly described, and it is obvious that the drawings in the description below are some embodiments of the present invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.
Fig. 1 is a flowchart of a soil monitoring data analysis method for mine ecological restoration provided by an embodiment of the invention;
FIG. 2 is a model training architecture for distributed federal learning provided in an embodiment of the present invention;
FIG. 3 is a flowchart of another analysis method of soil monitoring data for mine ecological restoration provided by the embodiment of the invention;
Fig. 4 is a schematic structural diagram of a soil monitoring data analysis device for mine ecological restoration provided by the embodiment of the invention;
fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Mining activities often have serious environmental impact, resulting in soil degradation, pollution and ecological system destruction. In order to effectively perform ecological restoration, the affected soil needs to be monitored and evaluated in order to formulate an appropriate restoration strategy. However, conventional soil monitoring methods face challenges such as data collection difficulties, data privacy and security issues, and inefficient data processing. In addition, due to the complexity and diversity of soil conditions, efficient data analysis methods are needed to process and parse large amounts of soil monitoring data to ensure the scientificity and effectiveness of the remediation measures.
In the related art, the soil monitoring data of multiple sites usually need to be centralized, which not only increases the cost of data transmission and storage, but also brings risks of data privacy and security, lacks effective data privacy protection measures, and limits the comprehensive utilization of the data.
Based on the above, the soil monitoring data analysis method, the device and the equipment for mine ecological restoration provided by the embodiment of the invention can be used for inputting the soil monitoring data into a pre-trained soil monitoring data analysis model by acquiring the soil monitoring data and outputting a soil monitoring data analysis result. The training mode of the pre-trained soil monitoring data analysis model is as follows: training a plurality of local models based on the training data set to obtain respective model parameters; based on the model parameters of each of the local models, carrying out parameter updating by using a global model of a central server to obtain target model parameters; transmitting target model parameters to a plurality of local models for global iteration until training is converged, obtaining a pre-trained soil monitoring data analysis model, and obtaining a training data set based on labeling, data expansion and feature extraction of a plurality of collected soil sample data; the data expansion method comprises the following steps: sample generation is carried out on the countermeasure network by utilizing generation based on the variable decibel leaf sparse matrix decomposition so as to realize data expansion. In the mode, the problems of data privacy and data safety during soil monitoring at multiple sites are solved, the cost of data transmission and storage is reduced, the central concentration and transmission of data are avoided by respectively carrying out model training and parameter updating on each local model, the risk of data leakage is effectively reduced, and the comprehensive utilization of the data is ensured.
In order to facilitate understanding of the embodiment, the soil monitoring data analysis method for mine ecological restoration disclosed by the embodiment of the invention is first described in detail.
Example 1
The embodiment of the invention provides a soil monitoring data analysis method for mine ecological restoration, and fig. 1 is a flow chart of the soil monitoring data analysis method for mine ecological restoration. As shown in fig. 1, the soil monitoring data analysis method for mine ecological restoration can include the following steps:
Step S101, acquiring soil monitoring data.
Among other things, soil monitoring data includes, but is not limited to: multidimensional information such as chemical composition, physical property, biological activity and the like of soil, for example, pH value of soil, organic matter content of soil, moisture content of soil, heavy metal content of soil, nitrogen, phosphorus and potassium content of soil, density of soil, temperature of soil, conductivity of soil, total bacterial amount of soil, particle size of soil and the like.
And step S102, inputting the soil monitoring data into a pre-trained soil monitoring data analysis model, and outputting a soil monitoring data analysis result.
The analysis result of the soil monitoring data can be one of the following conditions: is suitable for repairing, needs further improvement and serious pollution.
The training mode of the pre-trained soil monitoring data analysis model is as follows: training a plurality of local models based on the training data set to obtain respective model parameters; based on the model parameters of each of the local models, carrying out parameter updating by using a global model of a central server to obtain target model parameters; and sending the target model parameters to a plurality of local models for global iteration until training converges, and obtaining a pre-trained soil monitoring data analysis model.
The training data set is obtained based on labeling, data expansion and feature extraction of a plurality of collected soil sample data; the data expansion method comprises the following steps: the method for generating samples by using the generation of the correlation network based on the variational Bayesian sparse matrix decomposition to realize data expansion will be specifically described in the following embodiments, and will not be described herein.
Wherein the local model may comprise: a model for data annotation, a model for data augmentation, and a model for feature extraction.
The global model can be tested by inputting test data, global iteration is carried out by presetting iteration times, training is completed after convergence to obtain a final global model, and if the preset iteration times are not converged, the local model is adjusted to continue the steps.
The soil monitoring data analysis model is obtained through federal learning training, and because a large amount of effective training data is needed for characteristic extraction analysis learning of the soil monitoring data analysis task, the distributed model training and data mining of the multi-node combined model are realized through a distributed federal learning model training architecture, the problem of data resource island of a traditional centralized training mode is broken, and more effective data is obtained through aggregation. Meanwhile, the distributed federal learning architecture prevents the leakage of local data of each node, the data is used locally, the model is trained locally, and the data safety is effectively ensured.
In the federal learning framework, each node does not need to share the training data of its local model, but rather, trains the local model and sends the updated model to the centralized learning unit for summarization, and for convenience of understanding, fig. 2 is a model training framework of distributed federal learning provided by an embodiment of the present invention.
Specifically, for each global iteration, let the number of nodes beThe total number of samples owned isAnd (1)The number of samples of each node isThe federally learned objective function may be defined as the following formulas (1-1), (1-2):
（1-1）
（1-2）
Wherein, Analyzing an objective function of the model for pre-trained soil monitoring data,As a parameter of the model, it is possible to provide,Is a model parameterFor the firstA loss prediction of the data of the individual samples,In order to train the loss of the training,Is the firstSample characteristics of the individual sample data,First, theLabels of the individual sample data. Preferably, training lossAnd calculating by adopting a cross entropy loss function.
Further, for the firstFor a node, the objective function of the node may be defined by the following equation (2):
（2）
Wherein, Is the firstThe number of samples of the individual nodes,Is the firstThe objective function of the individual nodes is that,Is the firstData distribution of individual nodes.
In one embodiment, in the firstAt the time of iteration, the firstThe parameter updating mode of each node is taken as an example, and the first is setThe parameter gradient of each node isThen at the firstThe manner in which the model updates the parameters at the time of iteration can be expressed as the following equation (3):
（3）
Wherein, Is the firstThe model parameters of the number of iterations,Is the firstThe model parameters of the number of iterations,For the learning rate of the current parameter update,Is the firstNumber of samples of individual nodes.
Further, the parameter update method of the global model of the central server may be expressed as the following expression (4):
（4）
Wherein, Is the firstParameters of the global model of the central server for the next iteration.
Further, the iterative operation is repeatedly performed, that is, the global model representing the central server and the model training of each node are completed.
In one embodiment, the preset stop iteration condition is that a preset maximum number of iterations is reached, preferably the preset maximum number of iterations is set to 5000.
According to the soil monitoring data analysis method for mine ecological restoration, which is provided by the embodiment of the invention, the soil monitoring data can be input into a pre-trained soil monitoring data analysis model by acquiring the soil monitoring data, and the soil monitoring data analysis result is output. The training mode of the pre-trained soil monitoring data analysis model is as follows: training a plurality of local models based on the training data set to obtain respective model parameters; based on the model parameters of each of the local models, carrying out parameter updating by using a global model of a central server to obtain target model parameters; transmitting target model parameters to a plurality of local models for global iteration until training is converged, obtaining a pre-trained soil monitoring data analysis model, and obtaining a training data set based on labeling, data expansion and feature extraction of a plurality of collected soil sample data; the data expansion method comprises the following steps: sample generation is carried out on the countermeasure network by utilizing generation based on the variable decibel leaf sparse matrix decomposition so as to realize data expansion. In the mode, the problems of data privacy and data safety during soil monitoring at multiple sites are solved, the cost of data transmission and storage is reduced, the central concentration and transmission of data are avoided by respectively carrying out model training and parameter updating on each local model, the risk of data leakage is effectively reduced, and the comprehensive utilization of the data is ensured.
Example 2
The embodiment of the invention also provides another soil monitoring data analysis method for mine ecological restoration; the method is realized on the basis of the method of the embodiment; the method focuses on describing a specific implementation of the training mode of the local model.
Fig. 3 is a flowchart of another method for analyzing soil monitoring data for ecological restoration of a mine, where the training data set is obtained by labeling, expanding and extracting features of a plurality of collected soil sample data, and as shown in fig. 3, the training mode of the local model may include the following steps:
step S201, marking the collected multiple soil sample data to obtain a first training data set.
Specifically, labeling the collected plurality of soil sample data to obtain a first training data set may include: labeling a plurality of soil sample data based on preset data categories to obtain a first training data set; wherein, the data category includes: is suitable for repairing, needs further improvement and serious pollution.
The data is stored in a vector form and is in a JSON format.
Wherein, the attribute of the data may include: a1 represents pH value, reflecting the pH value of soil; a2 represents the organic matter content and is expressed in percentage; a3 represents the moisture content, expressed in percent; a4 represents the content of heavy metals such as lead, cadmium and the like, and the unit is mg/kg; a5 represents the content of nitrogen, phosphorus and potassium and is expressed in mg/kg; a6 represents soil density in g/cm; a7 represents soil temperature in degrees centigrade; a8 represents the conductivity of soil, and the unit is mS/cm; a9 represents the total amount of bacteria in CFU/g; a10 represents the soil particle size in microns.
Illustratively, a specific value of one of the pieces of soil sample data may be: a1 A2=5%, a3=20%, a4=20 mg/kg, a5=150 mg/kg, a6=1.3 g/cm, a7=22 ℃, a8=0.5 mS/cm, a9=10000 CFU/g, a10=200 μm.
It should be noted that, the present embodiment is only for illustrating one data format and kind of the present invention, in practical application, the data attributes are generally more than 10 attributes, and the number of attributes of the data may reach tens or hundreds, which is not limited to the above examples.
The labeling of the collected plurality of soil sample data may specifically be manual labeling, or may be labeling by pre-training a labeling model, which is not limited herein. The types of annotations may include: is suitable for repairing, needs further improvement and serious pollution.
Step S202, data expansion is performed on the first training data set to obtain a second training data set.
Specifically, performing data expansion on the first training data set to obtain a second training data set may include: and generating samples based on the first training data set by utilizing a generating countermeasure network based on the variable decibel-based sparse matrix decomposition so as to realize data expansion and obtain a second training data set.
The generation countermeasure network based on the variable decibel leaf sparse matrix decomposition comprises two parts: the generator is responsible for generating new data similar to the real data, and the discriminator is used for trying to distinguish the generated data from the real data. The self-attention layer is adopted in the generator and the discriminator, so that the model can better understand the characteristic dependency relationship in the data, and the model has more adaptability to complex interactions among soil attributes.
Specifically, the training process for generating the countermeasure network algorithm based on the variational Bayesian sparse matrix decomposition can comprise the following steps:
Step A1, initializing and setting network parameters of a generator and a discriminator, and setting the generator as The discriminator isThe weight of the generator isThe weight of the discriminator isThe initialization process adopts a random initialization method based on normal distribution to ensure the diversity of network parameters and preliminary nonlinear processing capacity, and is expressed as the following formulas (5-1) and (5-2):
（5-1）
（5-2）
Wherein, Representing an initialization standard deviation; representing compliance with a particular distribution; a normal distribution is indicated and the distribution is determined, Representing the mean value as 0 and the standard deviation as the identity matrixIs a normal distribution of (c). Preferably, the method comprises the steps of,Set to 0.01.
Step A2, performing specific adjustment by using the self-attention layer to enable the self-attention layer to be more suitable for key dependency relationship of soil data characteristics, further improving the correlation and accuracy of data generation, and specifically setting parameters of the self-attention layer as followsThe update rule of the parameter of the self-attention layer is the following expression (6):
（6）
Wherein, For the updated parameters of the self-attention layer,、AndA query matrix, a key matrix, and a value matrix, respectively, and these matrices are feature transformations learned from the previous layer of the generator; is a transpose of the key matrix; Representing the dimensions of the key matrix; as a Softmax function.
Further, the calculation modes of the query matrix, the key matrix and the value matrix are expressed as the following formulas (7-1), (7-2) and (7-3):
（7-1）
（7-2）
（7-3）
Wherein, 、AndA weight matrix of queries, keys, and values, respectively;、 And Is the bias vector of the query, key and value, respectively; Is the feature matrix of the input.
Step A3, optimizing sparse matrix parameters in the generator by a variable decibel leaf method, ensuring that the generated data can capture key few characteristics while keeping the data sparsity, and specifically, setting the sparse matrix of the generator asThe update rule of the sparse matrix is the following equation (8):
（8）
Wherein, For the updated sparse matrix of the generator,Is a sparsity adjustment parameter; Is the learning rate of the sparse matrix; Is a loss function that generates an antagonism network; is to generate a loss function of the countermeasure network Is a gradient of (2); A generator function. Preferably, the method comprises the steps of, The setting is made to be 0.3,Set to 0.7.
Further, generating a loss function for the antagonism networkIs realized by a chain law, expressed as the following expression (9):
（9）
Wherein, Is the sign of the partial derivative.
And step A4, after the sparse matrix learning is completed, performing countermeasure training of the generator and the discriminator, wherein the generator and the discriminator improve the mutual performance through continuous countermeasure, the generator learns and manufactures data which are more and more difficult to identify by the discriminator, and the discriminator strives to improve the capability of distinguishing true and false data.
During training of the generator, the sample generation process of the generator is enhanced by the updated parameters of the self-attention layer and optimized by the updated sparse matrix of the generator, expressed as the following equation (10):
（10）
Wherein, Representing the bias of the generator; is distributed from a priori Random noise extracted from the sample; The function is activated for Sigmoid.
Further, the generator and the arbiter are optimized by a loss function corresponding to the following expression (11):
（11）
Wherein, Is represented as inputAndGenerating a loss function against the networkUnder the constraint of (1) the generator minimizes the loss function and the arbiter maximizes the loss function; representing compliance with a particular distribution; Representing a sample of the real data, Is distributed from a prioriRandom noise extracted from the sample; Representing a priori distribution of the real data samples; representing a priori distribution of random noise; Representing the desire; the function of the arbiter is represented as such, Representing the generator function.
And step A5, repeating the steps until a preset iteration stopping condition is met, namely, finishing model training. In one embodiment, the preset stop iteration condition is that a preset maximum number of iterations is reached, preferably the preset maximum number of iterations is set to 1000.
After the data expansion model is trained, the number of samples is increased by using the trained data expansion model. In one embodiment, the original acquired samples are 800, 200 samples are obtained by expanding the data expansion model, and the expanded data set contains 1000 samples.
Step S203, extracting features from the second training data set to obtain a third training data set.
It should be noted that, the local model needs to perform feature extraction on a sample, and the embodiment of the application provides two types of feature extraction models, and when the feature dimension is lower than 30, a fully connected neural network of 6 layers is selected for feature extraction.
In the prior art, some schemes use neural networks to perform feature extraction, and in some neural network structures, problems of gradient disappearance, gradient explosion or local optimal solution may be encountered, which affect training stability and performance of the model. In the embodiment of the application, the training method of the local model adopts self-evolution genetic algorithm-based optimization to train a 6-layer fully-connected neural network, adopts self-adaptive genetic operation on the basis of the traditional genetic algorithm, and dynamically adjusts genetic parameters according to the performance of the network on a feature extraction task, thereby accelerating the convergence speed and improving the quality of the solution so as to maximize the information transmission and nonlinear expression capability in the feature extraction process.
Specifically, performing feature extraction on the second training data set to obtain a third training data set may include: and extracting features based on the second training data set by using a self-evolution genetic algorithm to obtain a third training data set.
The training mode based on the self-evolution genetic algorithm can comprise the following steps:
Step B1, initializing population according to genetic algorithm, randomly generating a group of neural network population, wherein each network has randomly initialized weight and structural parameters, and setting each neural network Is given by the weight ofEach neural networkIs biased toThe initialization method is expressed as the following formulas (12-1), (12-2):
（12-1）
（12-2）
Wherein, Standard deviation representing the initialization weight; a normal distribution with a mean value of 0 and a standard deviation of 1 is represented; And Respectively the firstWeights and biases for the individual neural networks. Preferably, the method comprises the steps of,Set to 0.01.
Step B2, carrying out adaptability evaluation on each neural network, calculating a fitness score by taking the characteristic extraction capacity of each neural network on a soil monitoring data set as a standard, and defining a fitness function as the following formula (13):
（13）
Wherein, Is a neural network; is a fitness function; Is the true output And network predictionThe mean square error between the two is determined by the real label corresponding to the input sample, and the network prediction is determined by classifying the characteristics of the neural network output by a preset Softmax function; Is a true characteristic value; is a neural network Is provided.
And B3, executing a selection operation, selecting a neural network with better performance to enter the next generation according to the fitness score, wherein the selection operation selects the neural network according to the fitness by using a roulette method, and the selection operation is expressed as the following formula (14):
（14）
Wherein, Is a neural networkProbability of being selected; Calculating a function for the selection probability; Is the first in the population A neural network.
Step B4, performing crossover and mutation operations on the selected neural networks to generate a new network population, specifically, the crossover operations are completed by randomly selecting two neural networks and exchanging their partial weights and biases, which are expressed as the following formulas (15-1), (15-2):
（15-1）
（15-2）
Wherein, Is a cross-over function of the two,Is the weight of the first neural network,Is the weight of the second neural network,Is the bias of the first neural network,Is the bias of the second neural network,Is the weight of the neural network after the crossover,Is the neural network bias after crossover.
Also, the mutation operation is realized by adding small random disturbance, expressed as the following formulas (16-1), (16-2):
（16-1）
（16-2）
Wherein, Is the weight of the neural network after the mutation,Is the bias of the neural network after the mutation,Is the variation rate of the product, and the product is a modified product,Is the variance of the disturbance. Preferably, the method comprises the steps of,Set to 0.01.
In one embodiment, the weights and biases of the parent neural network are disconnected at some point in real time, exchanging parts to generate children, expressed as the following equation (17):
（17）
Wherein, Is a randomly selected crossover point; is a random number in the range of 0,1, and is used to determine the direction of intersection.
Step B5, dynamically adjusting parameters of an activation function in the network according to the characteristic distribution of the current data so as to adapt to the specific characteristics of the data, and setting the activation function of the neural network asThe parameters are dynamically adjusted according to the network performance, and the adjustment mode is expressed as the following formula (18):
（18）
Wherein, The learning rate of the parameters of the activation function is adjusted; the parameters of the activation function are adjusted; is the partial derivative of the mean square error with respect to the activation function parameters. Preferably, the method comprises the steps of, Set to 0.01.
And step B6, repeating the steps until a preset iteration stopping condition is met, namely, finishing model training. In one embodiment, the preset stop iteration condition is that a preset maximum number of iterations is reached, preferably the preset maximum number of iterations is set to 1000.
When the feature dimension is equal to or higher than 30, the self-encoder is selected to extract the features. The present invention employs a self-encoder algorithm based on depth feature remodeling, comprising an encoder portion and a decoder portion. The encoder is responsible for compressing the high-dimensional input data into a low-dimensional representation from which the decoder attempts to reconstruct the original input data. The invention adopts a layering characteristic extraction mechanism to improve the characteristic processing capacity of the self-encoder, so that the self-encoder model can capture more complex nonlinear relations and more abstract data representations.
Specifically, performing feature extraction on the second training data set to obtain a third training data set may include: and extracting features by using a self-encoder algorithm based on depth feature remodeling based on the second training data set to obtain a third training data set.
The training mode of the self-encoder algorithm based on depth feature remodeling can comprise the following steps:
Step C1, initializing parameters of the self-encoder, and setting the weight of the self-encoder as Offset from encoder toThe manner of initialization is expressed as the following formulas (19-1), (19-2):
（19-1）
（19-2）
Wherein, Standard deviation initialized for weights; a normal distribution with a mean value of 0 and a standard deviation of 1 is represented; Representing the self-encoder first Layer to the firstWeights of the layers; Representing the self-encoder first Layer bias. Preferably, the method comprises the steps of,Set to 0.1.
And step C2, in the training stage of the encoder, the input data is compressed into a low-dimensional characteristic representation through forward propagation, and the neural network abstracts and transforms the input data layer by layer so as to extract key characteristic information. Let the input high-dimensional data beThe data features converted into the low-dimensional feature space areThe process of data forward propagation is expressed as the following equation (20):
（20）
Wherein, Is an activation function of the self-encoder; is an input high-dimensional data vector; Is the encoded low-dimensional feature vector; The normal distribution which represents the mean value of 0 and the covariance matrix of the unit matrix is used for generating noise; Is the noise intensity.
Further, the invention adopts the self-adaptive noise injection mechanism to enhance the generalization capability and the noise interference resistance capability of the model, and dynamically adjusts the noise intensity according to the performance of the model in the training process, thereby optimizing the training effect and improving the data representation quality, and particularly, the noise intensityAccording to the current reconstruction error self-adaptive adjustment of the model, the adjustment mode is expressed as the following formula (21):
（21）
Wherein, Is the baseline value of noise intensity; is an adjusting factor, and controls the attenuation speed of noise along with the reduction of errors; is the reconstruction error of the current model. Preferably, the method comprises the steps of, The setting is made to be 1.5,Set to 0.2.
In one embodiment, the activation function of the self-encoder adapts to the characteristic change of the training data by means of dynamic adjustment, and the input of the activation function is setIs thatThe activation function calculation method of the self-encoder is expressed by the following formulas (22-1), (22-2):
（22-1）
（22-2）
Wherein, Is a coefficient dynamically adjusted according to the data variance; Is a parameter for adjusting sensitivity; is the variance of the input data; is the threshold of variance; is a function of the activation of the ReLU, Is a Sigmoid activation function. Preferably, the method comprises the steps of,The setting is made to be 0.3,Is set to be 1, and is set to be 1,Set to 0.5.
Step C3, in the training phase of the decoder, the decoder receives the output of the encoder and tries to reconstruct the original input, the invention adjusts the weight and bias of the decoder by minimizing the reconstruction error so that the output is as close as possible to the original input, and the original data reconstructed from the low-dimensional feature space is set asThe calculation method is expressed by the following expression (23):
（23）
Wherein, AndThe weight and bias of the decoder respectively,Is a reconstruction function of the decoder. Preferably, the method comprises the steps of,Set as a transpose of the encoder weights.
Step C4, utilizing error feedback of the encoder and the decoder to carry out cooperative optimization, dynamically balancing coding precision and reconstruction quality by adjusting weight of a loss function, optimizing performance of the whole network, and setting the loss function of the self-encoder asThe calculation method is expressed by the following expression (24):
（24）
Wherein, Is the weight of the reconstruction error term; is the L2 norm; Is the weight of the regularization term to avoid overfitting; representing a reconstruction error, and representing the square of Euclidean distance between original data and reconstruction data; Is the square of the L2 norm of the weight as a regularization term.
And step C5, repeating the steps until a preset iteration stopping condition is met, namely, the model training is completed. In one embodiment, the preset stop iteration condition is that a preset maximum number of iterations is reached, preferably the preset maximum number of iterations is set to 8000.
It should be further noted that, if the feature extraction model of each local node is trained, the feature extraction model after the training is optionally used to perform the relevant decision task.
Step S204, training the local model based on the third training data set.
The classifier model is trained on the feature extracted data. The embodiment of the application adopts a quantum state depth neural decision tree algorithm based on conflict driving as a classification algorithm, and enhances the decision capability of the neural decision tree by utilizing the principle of quantum computation, so that the neural decision tree can provide more accurate classification when facing complex soil monitoring data.
Specifically, training the local model based on the third training data set may include: the local model is trained based on the third training data set by utilizing a quantum state depth neural decision tree algorithm based on conflict driving.
The training mode of the quantum state depth neural decision tree algorithm based on conflict driving can comprise the following steps:
Step D1, initializing all decision nodes of a neural decision tree into quantum states, and setting the quantum state of each decision node as the quantum state It is initialized to the superimposed state, expressed by the following equation (25):
（25）
Wherein, AndRepresenting the fundamental state of the qubit.
Step D2, starting from the root node of the neural decision tree, gradually adding decision nodes, at each decision point, evaluating all possible segmentation conditions by the algorithm, and selecting conditions capable of maximizing information gain, for the quantum state of each nodeThe quantum measurement operation is used to determine the splitting strategy of the node, and is expressed as the following equation (26):
（26）
Wherein, Representing a quantum measurement operation; And The representation corresponds to the firstThe ground state of the individual features is referred to as,Characterized as the firstThe right-hand vector of the individual features,Characterized as the firstLeft vectors of the individual features; the current feature sequence number; representing the right-hand vector of the vector, Representing the left vector.
And D3, at each node of the neural decision tree, adjusting the quantum state of the node according to the path of the data passing through the node, wherein the state is adjusted by adopting a quantum gate to reflect the decision process from the father node to the current node, and the decision process is expressed as the following formula (27):
（27）
Wherein, The quantum state of each decision node after quantum gate adjustment is obtained; Adjusting a function for the quantum state; The Hamiltonian quantity represents the total energy of the quantum system and is in direct proportion to the information gain of the nodes; Is an element multiplication operation; is the iteration number interval; is an imaginary unit; And The revolving doors are respectively arranged around the Y axis and the Z axis and are used for adjusting the quantum state according to input data; And Is a rotation angle calculated dynamically based on the data characteristics.
In one embodiment, the rotation angle is calculated by the following formulas (28-1), (28-2):
（28-1）
（28-2）
Wherein, As a hyperbolic tangent function; Is the error of the reconstruction and, Is the standard deviation of the data, used to adjust the rotation amplitude; Is a parameter of the learning rate and, AndThe actual input and the predicted output, respectively.
Step D4, detecting classification conflicts in the tree construction process, and solving the conflicts by adjusting the quantum states of the related nodes, wherein if the classification conflicts exist, namely samples in the same class are misclassified in different leaf nodes, conflict resolution is carried out, each adjustment is ensured to be carried out in the direction of reducing the overall classification errors, and the quantum state adjustment mode is expressed as the following formula (29):
（29）
Wherein, Quantum state of each decision node after adjustment for classification conflict; is a quantum gate operation for resolving conflicts; Pauli-X gate operation for flipping quantum states; is a control NOT gate for realizing the first Sum of quantum bitsConditional logic between individual qubits; The Hadamard gate is used for creating a superposition state and increasing the information entropy of decision nodes; is an identity operation, meaning that the first is unchanged States of the individual qubits.
Step D5, pruning the built neural decision tree to remove nodes with small contribution to classification in order to prevent overfitting, pruning by calculating the loss function of the neural decision tree, and setting the loss function of the neural decision tree asThe method is based on classification errors and purity of quantum states to calculate, so that the model can be correctly classified and the stability of the quantum states can be maintained, and the calculation mode is expressed as the following formula (30):
（30）
Wherein, Is the L2 norm; Is a real label; Is a label for model prediction; And Is a factor that balances classification errors and quantum state purity; Is a dirac symbol in quantum coding; Representing quantum states In the ground stateIs a projection amplitude of (2); Is a measure of the purity of the quantum state, used to evaluate the uniformity and stability of the quantum state, Is the entropy weight; Is the quantum entropy. Preferably, the method comprises the steps of, The setting is made to be 0.3,Set to 0.7.
Furthermore, the invention adopts a dynamic quantum entropy weight to adjust the processing capability of a mechanical enhancement model on uncertainty information, optimizes information gain calculation of decision tree nodes by adjusting quantum entropy weight, further improves classification accuracy and robustness, and particularly adopts an adaptive entropy weight adjuster based on data flow change to dynamically adjust the weight of quantum entropy according to the diversity and change rate of data flow passing through the decision tree nodes, and defines the quantum entropy asThe calculation method is expressed by the following expression (31):
（31）
Wherein, Is the density matrix of the current node; And (3) representing trace operation for calculating the trace of the matrix.
Further, the adjustment of the entropy weight is obtained according to the change rate and diversity of the data, and the calculation mode is expressed as the following formula (32):
（32）
Wherein, Is an entropy weight adjustment factor for balancing the sensitivity of weight adjustment; representing passing nodes Calculating the average difference between the characteristics of the current batch data and the previous batch data; the data diversity is calculated as the variance of the current data; Is a small constant. Preferably, the method comprises the steps of, Set to 0.001.
Based on this, in pruning, if the value of the loss function of the neural decision tree decreases after pruning the node, it is confirmed that the pruning operation is performed.
And D6, after model pruning is completed, the model training is completed. After model training is completed, a new soil monitoring data sample is processed by utilizing the feature extraction model which is already trained to obtain feature extracted soil monitoring data, and further, the feature extracted soil monitoring data is input into a classifier which is already trained, and a final classification result is output.
According to the soil monitoring data analysis method for mine ecological restoration, provided by the embodiment of the invention, a distributed federal learning model training architecture is adopted, so that the problems of data privacy and data safety in soil monitoring at multiple sites are solved, and the central concentration and transmission of data are avoided and the risk of data leakage is effectively reduced by locally carrying out model training and parameter updating at each node; the data expansion is carried out by adopting a generating countermeasure network based on the decomposition of the variable decibel leaf sparse matrix, so that the problems of insufficient original data quantity and unbalanced samples are solved, high-quality and diversified synthetic data are generated in the countermeasure process of the generator and the discriminator, and the training effect of the model is enhanced; the self-evolution genetic algorithm and the self-encoder algorithm are adopted to perform feature extraction and dimension reduction, so that the understanding capability and the processing efficiency of the soil data features are improved, and particularly when the feature dimension is high, the self-encoder effectively compresses and reconstructs data through the encoding and decoding processes, so that the calculation complexity is reduced while key information is kept; the quantum state deep neural decision tree based on conflict driving is adopted for classification, so that the accuracy and the efficiency of classification are improved, and the capability of the decision tree for processing complex data is enhanced by utilizing the quantum computing principle when the soil monitoring data of multiple categories and complex decision boundaries are faced.
Example 3
Corresponding to the above method embodiment, the embodiment of the present invention provides a soil monitoring data analysis device for mine ecological restoration, and fig. 4 is a schematic structural diagram of the soil monitoring data analysis device for mine ecological restoration provided by the embodiment of the present invention, as shown in fig. 4, the soil monitoring data analysis device for mine ecological restoration may include:
The soil monitoring data acquisition module 301 is configured to acquire soil monitoring data.
The soil monitoring data analysis module 302 is configured to input soil monitoring data into a pre-trained soil monitoring data analysis model, and output a soil monitoring data analysis result; the training mode of the pre-trained soil monitoring data analysis model is as follows: training a plurality of local models based on the training data set to obtain respective model parameters; based on the model parameters of each of the local models, carrying out parameter updating by using a global model of a central server to obtain target model parameters; transmitting the target model parameters to a plurality of local models for global iteration until training converges, and obtaining a pre-trained soil monitoring data analysis model;
The training data set is obtained based on labeling, data expansion and feature extraction of a plurality of collected soil sample data; the data expansion method comprises the following steps: sample generation is carried out on the countermeasure network by utilizing generation based on the variable decibel leaf sparse matrix decomposition so as to realize data expansion; the generating an countermeasure network based on the variable decibel leaf sparse matrix decomposition comprises: a generator and a arbiter;
The method for generating the sample by using the generation countermeasure network based on the variable decibel leaf sparse matrix decomposition comprises the following steps: initializing network parameters of the generator and the discriminator; performing corresponding adjustment by using the self-attention layer; optimizing sparse matrix parameters in the generator by a variable decibel leaf method, and setting the sparse matrix of the generator as The updating rule of the sparse matrix is as follows: ; wherein, For the updated sparse matrix of the generator,Is a sparsity adjustment parameter; Is the learning rate of the sparse matrix; Is a loss function that generates an antagonism network; is to generate a loss function of the countermeasure network Is a gradient of (2); A generator function; after the sparse matrix learning is completed, performing countermeasure training of the generator and the discriminator; in the training process of the generator, the sample generation process of the generator is enhanced through the updated parameters of the self-attention layer, and optimized through the updated sparse matrix of the generator, and the sample generation process of the generator is expressed as the following formula: ; wherein, Representing the bias of the generator; is distributed from a priori Random noise extracted from the sample; The function is activated for Sigmoid.
According to the soil monitoring data analysis device for mine ecological restoration, provided by the embodiment of the invention, the soil monitoring data can be input into a pre-trained soil monitoring data analysis model by acquiring the soil monitoring data, and the soil monitoring data analysis result is output. The training mode of the pre-trained soil monitoring data analysis model is as follows: training a plurality of local models based on the training data set to obtain respective model parameters; based on the model parameters of each of the local models, carrying out parameter updating by using a global model of a central server to obtain target model parameters; transmitting target model parameters to a plurality of local models for global iteration until training is converged, obtaining a pre-trained soil monitoring data analysis model, and obtaining a training data set based on labeling, data expansion and feature extraction of a plurality of collected soil sample data; the data expansion method comprises the following steps: sample generation is carried out on the countermeasure network by utilizing generation based on the variable decibel leaf sparse matrix decomposition so as to realize data expansion. In the mode, the problems of data privacy and data safety during soil monitoring at multiple sites are solved, the cost of data transmission and storage is reduced, the central concentration and transmission of data are avoided by respectively carrying out model training and parameter updating on each local model, the risk of data leakage is effectively reduced, and the comprehensive utilization of the data is ensured.
In some embodiments, the training data set is obtained based on labeling, data expansion and feature extraction of the collected multiple soil sample data, and the soil monitoring data analysis module is further configured to label the collected multiple soil sample data to obtain a first training data set; performing data expansion on the first training data set to obtain a second training data set; extracting features of the second training data set to obtain a third training data set; the local model is trained based on the third training dataset.
In some embodiments, the soil monitoring data analysis module is further configured to label the plurality of soil sample data based on a preset data category to obtain a first training data set; wherein, the data category includes: is suitable for repairing, needs further improvement and serious pollution.
In some embodiments, the soil monitoring data analysis module is further configured to perform sample generation based on the first training data set by using a generating countermeasure network based on a variational decibel-based sparse matrix decomposition, so as to achieve data expansion to obtain a second training data set; the generating an countermeasure network based on the variational Bayesian sparse matrix decomposition comprises: a generator and a arbiter; the loss functions of the generator and the arbiter are expressed by the following formulas: ; wherein, Is represented as inputAndGenerating a loss function against the networkUnder the constraint of (1) the generator minimizes the loss function and the arbiter maximizes the loss function; representing compliance with a particular distribution; Representing a sample of the real data, Representing the distribution from a prioriRandom noise extracted from the sample; Representing a priori distribution of the real data samples; representing a priori distribution of random noise; Representing the desire; the function of the arbiter is represented as such, Representing the generator function.
In some embodiments, the soil monitoring data analysis module is further configured to obtain a third training data set based on the second training data set by performing feature extraction based on a self-evolutionary genetic algorithm; the training mode based on the self-evolution genetic algorithm comprises the following steps: randomly generating a group of neural network populations; wherein, each neural network has a weight and a structural parameter initialized randomly; performing adaptability evaluation on each neural network, and selecting the neural network which meets the preset standard based on the result of the adaptability evaluation; performing crossover and mutation operations on the selected neural network to generate a new neural network population; dynamically adjusting parameters of an activation function of a new neural network based on the feature distribution of the current data to realize feature extraction to obtain a third training data set; wherein the operation of mutation is achieved by adding small random perturbations, expressed by the following formula:， ; wherein, Is the weight of the neural network after the mutation,Is the bias of the neural network after the mutation,Is the variation rate of the product, and the product is a modified product,Is the variance of the disturbance.
In some embodiments, the soil monitoring data analysis module is further configured to perform feature extraction based on the second training data set using a self-encoder algorithm based on depth feature remodeling to obtain a third training data set; the training mode of the self-encoder algorithm based on depth feature remodeling comprises the following steps: initializing parameters of a self-encoder; the self-encoder includes an encoder and a decoder; in the process of encoder training, input data is compressed into low-dimensional characteristic representation through forward propagation, and a neural network abstracts and transforms the input data layer by layer to extract corresponding characteristic information; during the decoder training process, the decoder receives the output of the encoder, reconstructing the original input; co-optimizing using error feedback of the encoder and decoder;
wherein the forward propagation process is expressed as the following equation: ; wherein, Is an activation function of the self-encoder; is an input high-dimensional data vector; Is the encoded low-dimensional feature vector; The normal distribution which represents the mean value of 0 and the covariance matrix of the unit matrix is used for generating noise; Is the noise intensity; wherein the activation function of the self-encoder adapts to the characteristic change of the training data in a dynamic adjustment mode, and inputs the activation function Is thatThe activation function calculation mode of the self-encoder is expressed as the following formula:， ; wherein, Is a coefficient dynamically adjusted according to the data variance; Is a parameter for adjusting sensitivity; is the variance of the input data; is the threshold of variance; is a function of the activation of the ReLU, Is a Sigmoid activation function.
In some embodiments, the soil monitoring data analysis module is further configured to train the local model using a quantum state deep neural decision tree algorithm based on collision driving based on the third training data set; the training mode of the quantum state depth neural decision tree algorithm based on conflict driving comprises the following steps: initializing all decision nodes of a neural decision tree into a quantum state; gradually adding decision nodes from the root node of the neural decision tree; evaluating and obtaining a plurality of segmentation conditions based on each decision node, and selecting the segmentation conditions for maximizing the information gain; at each decision node of the neural decision tree, the quantum state of the decision node is adjusted according to the path of the data passing through the decision node; in the construction process of the neural decision tree, detecting classification conflicts, and solving the classification conflicts by adjusting quantum states of relevant decision nodes; pruning is carried out on the built neural decision tree, and training of the local model is completed after pruning is completed.
In some embodiments, the objective function of the pre-trained soil monitoring data analysis model is represented by the following equation:， ; wherein, Analyzing an objective function of the model for pre-trained soil monitoring data,As a parameter of the model, it is possible to provide,Is a model parameterFor the firstA loss prediction of the data of the individual samples,In order to train the loss of the training,Is the firstSample characteristics of the individual sample data,First, theA tag for each sample data; the objective function of the local model is represented by the following equation: ; wherein, Is the firstThe number of samples of the individual local model,Is the firstThe objective function of the individual local model is,Is the firstData distribution of individual nodes.
The device provided by the embodiment of the present invention has the same implementation principle and technical effects as those of the foregoing method embodiment, and for the sake of brevity, reference may be made to the corresponding content in the foregoing method embodiment where the device embodiment is not mentioned.
Example 4
The embodiment of the invention also provides electronic equipment for running the soil monitoring data analysis method facing the mine ecological restoration; referring to the schematic structural diagram of an electronic device shown in fig. 5, the electronic device includes a memory 400 and a processor 401, where the memory 400 is configured to store one or more computer instructions, and the one or more computer instructions are executed by the processor 401 to implement the above-mentioned soil monitoring data analysis method for mine ecological restoration.
Further, the electronic device shown in fig. 5 further includes a bus 402 and a communication interface 403, and the processor 401, the communication interface 403, and the memory 400 are connected by the bus 402.
The memory 400 may include a high-speed random access memory (RAM, random Access Memory), and may further include a non-volatile memory (non-volatile memory), such as at least one disk memory. The communication connection between the system network element and at least one other network element is implemented via at least one communication interface 403 (which may be wired or wireless), which may use the internet, a wide area network, a local network, a metropolitan area network, etc. Bus 402 may be an ISA bus, a PCI bus, an EISA bus, or the like. The buses may be divided into address buses, data buses, control buses, etc. For ease of illustration, only one bi-directional arrow is shown in FIG. 5, but not only one bus or type of bus.
The processor 401 may be an integrated circuit chip having signal processing capabilities. In implementation, the steps of the above method may be performed by integrated logic circuits of hardware in the processor 401 or by instructions in the form of software. The processor 401 may be a general-purpose processor, including a central processing unit (Central Processing Unit, abbreviated as CPU), a network processor (Network Processor, abbreviated as NP), and the like; but may also be a digital signal Processor (DIGITAL SIGNAL Processor, DSP), application Specific Integrated Circuit (ASIC), field-Programmable gate array (FPGA) or other Programmable logic device, discrete gate or transistor logic device, discrete hardware components. The disclosed methods, steps, and logic blocks in the embodiments of the present invention may be implemented or performed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of the method disclosed in connection with the embodiments of the present invention may be embodied directly in the execution of a hardware decoding processor, or in the execution of a combination of hardware and software modules in a decoding processor. The software modules may be located in a random access memory, flash memory, read only memory, programmable read only memory, or electrically erasable programmable memory, registers, etc. as well known in the art. The storage medium is located in the memory 400, and the processor 401 reads the information in the memory 400, and in combination with its hardware, performs the steps of the method of the previous embodiment.
The embodiment of the invention also provides a computer readable storage medium, which stores computer executable instructions that, when being called and executed by a processor, cause the processor to implement the above soil monitoring data analysis method for mine ecological restoration, and specific implementation can be seen in the method embodiment and will not be repeated here.
The computer program product for performing the soil monitoring data analysis method for mine ecological restoration provided by the embodiment of the invention comprises a computer readable storage medium storing non-volatile program codes executable by a processor, wherein the instructions included in the program codes can be used for executing the method described in the method embodiment, and specific implementation can be seen in the method embodiment and will not be repeated here.
It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, and are not repeated herein.
Finally, it should be noted that: the above examples are only specific embodiments of the present invention, and are not intended to limit the scope of the present invention, but it should be understood by those skilled in the art that the present invention is not limited thereto, and that the present invention is described in detail with reference to the foregoing examples: any person skilled in the art may modify or easily conceive of the technical solution described in the foregoing embodiments, or perform equivalent substitution of some of the technical features, while remaining within the technical scope of the present disclosure; such modifications, changes or substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention, and are intended to be included in the scope of the present invention. Therefore, the protection scope of the invention is subject to the protection scope of the claims.

Claims (10)

1. The soil monitoring data analysis method for mine ecological restoration is characterized by comprising the following steps of:
Acquiring soil monitoring data;
inputting the soil monitoring data into a pre-trained soil monitoring data analysis model, and outputting a soil monitoring data analysis result;
the training mode of the pre-trained soil monitoring data analysis model is as follows: training a plurality of local models based on the training data set to obtain respective model parameters; based on the model parameters of each of the local models, carrying out parameter updating by using a global model of a central server to obtain target model parameters; transmitting the target model parameters to a plurality of local models for global iteration until training converges, and obtaining a pre-trained soil monitoring data analysis model;
The training data set is obtained based on labeling, data expansion and feature extraction of a plurality of collected soil sample data; the data expansion method comprises the following steps: sample generation is carried out on the countermeasure network by utilizing generation based on the variable decibel leaf sparse matrix decomposition so as to realize data expansion; the generating an countermeasure network based on the variable decibel leaf sparse matrix decomposition comprises: a generator and a arbiter;
The method for generating the sample by using the generation countermeasure network based on the variable decibel leaf sparse matrix decomposition comprises the following steps: initializing network parameters of the generator and the discriminator; performing corresponding adjustment by using the self-attention layer; optimizing sparse matrix parameters in the generator by a variable decibel leaf method, and setting the sparse matrix of the generator as The updating rule of the sparse matrix is as follows: ; wherein, For the updated sparse matrix of the generator,Is a sparsity adjustment parameter; Is the learning rate of the sparse matrix; Is a loss function that generates an antagonism network; is to generate a loss function of the countermeasure network Is a gradient of (2); A generator function; after the sparse matrix learning is completed, performing countermeasure training of the generator and the discriminator; in the training process of the generator, the sample generation process of the generator is enhanced through the updated parameters of the self-attention layer, and optimized through the updated sparse matrix of the generator, and the sample generation process of the generator is expressed as the following formula: ; wherein, Representing the bias of the generator; is distributed from a priori Random noise extracted from the sample; The function is activated for Sigmoid.
2. The method of claim 1, wherein the training dataset is based on labeling, data expansion, and feature extraction of the plurality of collected soil sample data; the training mode of the local model is as follows:
Labeling the collected multiple soil sample data to obtain a first training data set;
performing data expansion on the first training data set to obtain a second training data set;
extracting features of the second training data set to obtain a third training data set;
training the local model based on the third training dataset.
3. The method of claim 2, wherein labeling the plurality of collected soil sample data to obtain a first training data set comprises:
labeling a plurality of soil sample data based on preset data types to obtain a first training data set;
wherein the data category comprises: is suitable for repairing, needs further improvement and serious pollution.
4. The method of claim 2, wherein the data augmenting the first training data set to obtain a second training data set comprises:
Sample generation is carried out on the basis of the first training data set by utilizing a generating countermeasure network based on the decomposition of the variable decibel-based sparse matrix so as to realize data expansion to obtain a second training data set;
the loss functions of the generator and the arbiter are represented by the following formulas: ; wherein, Is represented as inputAndGenerating a loss function against the networkUnder the constraint of (1) the generator minimizes the loss function and the arbiter maximizes the loss function; representing compliance with a particular distribution; Representing a sample of the real data, Representing the distribution from a prioriRandom noise extracted from the sample; Representing a priori distribution of the real data samples; representing a priori distribution of random noise; Representing the desire; the function of the arbiter is represented as such, Representing the generator function.
5. The method of claim 2, wherein the feature extraction of the second training data set to obtain a third training data set comprises:
Performing feature extraction based on the second training data set by using a self-evolution genetic algorithm to obtain a third training data set;
the training mode based on the self-evolution genetic algorithm comprises the following steps:
randomly generating a group of neural network populations; wherein, each neural network has a weight and a structural parameter initialized randomly;
Performing adaptability evaluation on each neural network, and selecting the neural network meeting the preset standard based on the result of the adaptability evaluation;
Performing crossover and mutation operations on the selected neural network to generate a new neural network population;
Dynamically adjusting parameters of an activation function of the new neural network based on the feature distribution of the current data to realize feature extraction to obtain a third training data set;
wherein the operation of mutation is achieved by adding small random perturbations, expressed by the following formula: ，；
Wherein, Is the weight of the neural network after the mutation,Is the bias of the neural network after the mutation,Is the variation rate of the product, and the product is a modified product,Is the variance of the disturbance.
6. The method of claim 2, wherein the feature extraction of the second training data set to obtain a third training data set comprises:
Performing feature extraction by using a self-encoder algorithm based on depth feature remodeling based on the second training data set to obtain a third training data set;
The training mode of the self-encoder algorithm based on depth feature remodeling comprises the following steps:
initializing parameters of a self-encoder; the self-encoder includes an encoder and a decoder;
In the process of training the encoder, input data is compressed into low-dimensional characteristic representation through forward propagation, and the neural network abstracts and transforms the input data layer by layer to extract corresponding characteristic information;
During the decoder training process, the decoder receives the output of the encoder, reconstructing the original input;
Co-optimizing using error feedback of the encoder and decoder;
wherein the forward propagation process is expressed as the following equation: ; wherein, Is an activation function of the self-encoder; is an input high-dimensional data vector; Is the encoded low-dimensional feature vector; The normal distribution which represents the mean value of 0 and the covariance matrix of the unit matrix is used for generating noise; is the noise intensity;
wherein the activation function of the self-encoder adapts to the characteristic change of the training data in a dynamic adjustment mode, and inputs the activation function Is thatThe activation function calculation mode of the self-encoder is expressed as the following formula:， ; wherein, Is a coefficient dynamically adjusted according to the data variance; Is a parameter for adjusting sensitivity; is the variance of the input data; is the threshold of variance; is a function of the activation of the ReLU, Is a Sigmoid activation function.
7. The method of claim 2, wherein the training the local model based on the third training data set comprises:
Training the local model by utilizing a quantum state depth neural decision tree algorithm based on conflict driving based on the third training data set;
The training mode of the quantum state depth neural decision tree algorithm based on conflict driving comprises the following steps:
initializing all decision nodes of a neural decision tree into a quantum state;
Gradually adding decision nodes from the root node of the neural decision tree;
evaluating a plurality of segmentation conditions based on each decision node, and selecting the segmentation conditions for maximizing the information gain;
at each decision node of the neural decision tree, adjusting the quantum state of the decision node according to the path of data passing through the decision node;
In the construction process of the neural decision tree, detecting classification conflicts, and solving the classification conflicts by adjusting quantum states of relevant decision nodes;
pruning is carried out on the built neural decision tree, and training of the local model is completed after pruning is completed.
8. The method of claim 1, wherein the objective function of the pre-trained soil monitoring data analysis model is represented by the following equation:，；
Wherein, Analyzing an objective function of the model for pre-trained soil monitoring data,As a parameter of the model, it is possible to provide,Is a model parameterFor the firstA loss prediction of the data of the individual samples,In order to train the loss of the training,Is the firstSample characteristics of the individual sample data,First, theA tag for each sample data;
the objective function of the local model is represented by the following equation: ；
Wherein, Is the firstThe number of samples of the individual local model,Is the firstThe objective function of the individual local model is,Is the firstData distribution of individual nodes.
9. Soil monitoring data analysis device towards ecological restoration of mine, characterized in that, the device includes:
the soil monitoring data acquisition module is used for acquiring soil monitoring data;
The soil monitoring data analysis module is used for inputting the soil monitoring data into a pre-trained soil monitoring data analysis model and outputting a soil monitoring data analysis result; the training mode of the pre-trained soil monitoring data analysis model is as follows: training a plurality of local models based on the training data set to obtain respective model parameters; based on the model parameters of each of the local models, carrying out parameter updating by using a global model of a central server to obtain target model parameters; transmitting the target model parameters to a plurality of local models for global iteration until training converges, and obtaining a pre-trained soil monitoring data analysis model;
The training data set is obtained based on labeling, data expansion and feature extraction of a plurality of collected soil sample data; the data expansion method comprises the following steps: sample generation is carried out on the countermeasure network by utilizing generation based on the variable decibel leaf sparse matrix decomposition so as to realize data expansion; the generating an countermeasure network based on the variable decibel leaf sparse matrix decomposition comprises: a generator and a arbiter;
The method for generating the sample by using the generation countermeasure network based on the variable decibel leaf sparse matrix decomposition comprises the following steps: initializing network parameters of the generator and the discriminator; performing corresponding adjustment by using the self-attention layer; optimizing sparse matrix parameters in the generator by a variable decibel leaf method, and setting the sparse matrix of the generator as The updating rule of the sparse matrix is as follows: ; wherein, For the updated sparse matrix of the generator,Is a sparsity adjustment parameter; Is the learning rate of the sparse matrix; Is a loss function that generates an antagonism network; is to generate a loss function of the countermeasure network Is a gradient of (2); A generator function; after the sparse matrix learning is completed, performing countermeasure training of the generator and the discriminator; in the training process of the generator, the sample generation process of the generator is enhanced through the updated parameters of the self-attention layer, and optimized through the updated sparse matrix of the generator, and the sample generation process of the generator is expressed as the following formula: ; wherein, Representing the bias of the generator; is distributed from a priori Random noise extracted from the sample; The function is activated for Sigmoid.
10. An electronic device comprising a processor and a memory, the memory storing computer executable instructions executable by the processor, the processor executing the computer executable instructions to implement the mine ecorepair-oriented soil monitoring data analysis method of any one of claims 1 to 8.
CN202411366744.5A 2024-09-29 2024-09-29 Soil monitoring data analysis method, device and equipment for mine ecological restoration Active CN118885818B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202411366744.5A CN118885818B (en) 2024-09-29 2024-09-29 Soil monitoring data analysis method, device and equipment for mine ecological restoration

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202411366744.5A CN118885818B (en) 2024-09-29 2024-09-29 Soil monitoring data analysis method, device and equipment for mine ecological restoration

Publications (2)

Publication Number Publication Date
CN118885818A true CN118885818A (en) 2024-11-01
CN118885818B CN118885818B (en) 2025-02-14

Family

ID=93234600

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202411366744.5A Active CN118885818B (en) 2024-09-29 2024-09-29 Soil monitoring data analysis method, device and equipment for mine ecological restoration

Country Status (1)

Country Link
CN (1) CN118885818B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN119249240A (en) * 2024-12-04 2025-01-03 四川准达信息技术股份有限公司 A big data analysis method for photovoltaic power generation
CN119831744A (en) * 2024-12-23 2025-04-15 中国工商银行股份有限公司 Data processing method, data processing device, computer readable storage medium and electronic equipment
CN119886976A (en) * 2025-03-28 2025-04-25 南充市农业科学院 Soil quality control method based on data analysis
CN120328286A (en) * 2025-04-11 2025-07-18 波士顿电梯（湖州）有限公司 Elevator operation status monitoring and early warning method and platform based on multi-source data fusion

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103971162A (en) * 2014-04-04 2014-08-06 华南理工大学 Method for improving BP (back propagation) neutral network and based on genetic algorithm
US20190303717A1 (en) * 2018-03-28 2019-10-03 Kla-Tencor Corporation Training a neural network for defect detection in low resolution images
CN110889496A (en) * 2019-12-11 2020-03-17 北京工业大学 Human brain effect connection identification method based on confrontation generation network
CN116108371A (en) * 2023-04-13 2023-05-12 西华大学 Cloud service abnormity diagnosis method and system based on cascade abnormity generation network
CN116992980A (en) * 2023-09-28 2023-11-03 中国人民解放军总医院 A prognosis prediction and early warning model training method, system and equipment based on hypernetwork and federated learning
CN117336071A (en) * 2023-10-17 2024-01-02 江苏新巢天诚智能技术有限公司 Security protection method and device for Internet of Things equipment based on distributed AI
CN117892251A (en) * 2024-03-18 2024-04-16 山东神力索具有限公司 Rigging forging process parameter monitoring and early warning method and device based on artificial intelligence
CN118569506A (en) * 2024-07-31 2024-08-30 山东正元冶达环境科技有限公司 Mine ecological environment restoration and management assessment method and device

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103971162A (en) * 2014-04-04 2014-08-06 华南理工大学 Method for improving BP (back propagation) neutral network and based on genetic algorithm
US20190303717A1 (en) * 2018-03-28 2019-10-03 Kla-Tencor Corporation Training a neural network for defect detection in low resolution images
CN110889496A (en) * 2019-12-11 2020-03-17 北京工业大学 Human brain effect connection identification method based on confrontation generation network
CN116108371A (en) * 2023-04-13 2023-05-12 西华大学 Cloud service abnormity diagnosis method and system based on cascade abnormity generation network
CN116992980A (en) * 2023-09-28 2023-11-03 中国人民解放军总医院 A prognosis prediction and early warning model training method, system and equipment based on hypernetwork and federated learning
CN117336071A (en) * 2023-10-17 2024-01-02 江苏新巢天诚智能技术有限公司 Security protection method and device for Internet of Things equipment based on distributed AI
CN117892251A (en) * 2024-03-18 2024-04-16 山东神力索具有限公司 Rigging forging process parameter monitoring and early warning method and device based on artificial intelligence
CN118569506A (en) * 2024-07-31 2024-08-30 山东正元冶达环境科技有限公司 Mine ecological environment restoration and management assessment method and device

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
JING TIAN ET AL: "Pattern-adaptive generative adversarial network with sparse data for traffic state estimation", 《PHYSICA A》, vol. 608, 20 October 2022 (2022-10-20), pages 1 - 20 *
熊鹏文 等: "基于变分贝叶斯高斯混合噪声模型的机器人跨模态生成对抗网络", 《中国科学：信息科学》, vol. 51, no. 1, 31 January 2021 (2021-01-31), pages 104 - 121 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN119249240A (en) * 2024-12-04 2025-01-03 四川准达信息技术股份有限公司 A big data analysis method for photovoltaic power generation
CN119249240B (en) * 2024-12-04 2025-03-18 四川准达信息技术股份有限公司 Big data analysis method for photovoltaic power generation
CN119831744A (en) * 2024-12-23 2025-04-15 中国工商银行股份有限公司 Data processing method, data processing device, computer readable storage medium and electronic equipment
CN119886976A (en) * 2025-03-28 2025-04-25 南充市农业科学院 Soil quality control method based on data analysis
CN120328286A (en) * 2025-04-11 2025-07-18 波士顿电梯（湖州）有限公司 Elevator operation status monitoring and early warning method and platform based on multi-source data fusion
CN120328286B (en) * 2025-04-11 2025-11-07 波士顿电梯（湖州）有限公司 A Method and Platform for Elevator Operation Status Monitoring and Early Warning Based on Multi-Source Data Fusion

Also Published As

Publication number Publication date
CN118885818B (en) 2025-02-14

Similar Documents

Publication Publication Date Title
CN118885818B (en) Soil monitoring data analysis method, device and equipment for mine ecological restoration
Khan et al. An unsupervised deep learning ensemble model for anomaly detection in static attributed social networks
Hui et al. Collaborative graph convolutional networks: Unsupervised learning meets semi-supervised learning
Jiang et al. Deep learning based multi-channel intelligent attack detection for data security
Zhang et al. Deep adversarial learning in intrusion detection: A data augmentation enhanced framework
Sulaiman et al. Credit Card Fraud Detection Using Improved Deep Learning Models.
JP2021060992A (en) Machine learning system and method
Lopes et al. Effective network intrusion detection via representation learning: A Denoising AutoEncoder approach
Shrivastava et al. GLAD: Learning sparse graph recovery
Wang et al. Autoencoder, low rank approximation and pseudoinverse learning algorithm
Wang et al. Prediction method of wind farm power generation capacity based on feature clustering and correlation analysis
CN118606794B (en) Water pollution control sampling data analysis method and device based on machine learning
CN118485155B (en) Quantum noise interference resisting method and system based on quantum random ban
CN112633426A (en) Method and device for processing data class imbalance, electronic equipment and storage medium
Fan et al. Deep adversarial canonical correlation analysis
He et al. VAEAT: Variational AutoeEncoder with adversarial training for multivariate time series anomaly detection
CN118569892A (en) A method for predicting second-hand housing prices based on deep learning
CN118779435A (en) Natural language knowledge extraction method and device based on large and small model collaboration
CN109409434B (en) Method for extracting classification rules of liver disease data based on random forest
Vieloszynski et al. Latentqgan: A hybrid qgan with classical convolutional autoencoder
Sun et al. A compound structure for wind speed forecasting using MKLSSVM with feature selection and parameter optimization
CN119293410A (en) Energy efficiency evaluation and optimization method and device for cleaning equipment based on artificial intelligence
Bebarta et al. Polynomial based functional link artificial recurrent neural network adaptive system for predicting Indian stocks
CN116108340B (en) CGAN-based weapon combat effectiveness evaluation method based on small sample data generation
CN112183733A (en) Gaussian Embedded Neural Network Model and Modeling Method for Time Series Forecasting

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
CB02 Change of applicant information

Country or region after: China

Address after: Lixia District, Shandong city of Ji'nan Province, 250014 mountain East Road No. 14

Applicant after: Shandong Zhengyuan Environmental Technology Co.,Ltd.

Address before: Lixia District, Shandong city of Ji'nan Province, 250014 mountain East Road No. 14

Applicant before: Shandong Zhengyuan Yeda Environmental Technology Co.,Ltd.

Country or region before: China

CB02 Change of applicant information
GR01 Patent grant
GR01 Patent grant