[go: up one dir, main page]

CN120046007B - Intelligent diagnosis method and system for building electrical safety risk based on big data - Google Patents

Intelligent diagnosis method and system for building electrical safety risk based on big data

Info

Publication number
CN120046007B
CN120046007B CN202510518162.2A CN202510518162A CN120046007B CN 120046007 B CN120046007 B CN 120046007B CN 202510518162 A CN202510518162 A CN 202510518162A CN 120046007 B CN120046007 B CN 120046007B
Authority
CN
China
Prior art keywords
data
value
safety
objective function
intelligent diagnosis
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.)
Active
Application number
CN202510518162.2A
Other languages
Chinese (zh)
Other versions
CN120046007A (en
Inventor
周永明
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.)
Shenzhen Sanjiang Electric Co ltd
Original Assignee
Shenzhen Sanjiang Electric 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 Shenzhen Sanjiang Electric Co ltd filed Critical Shenzhen Sanjiang Electric Co ltd
Priority to CN202510518162.2A priority Critical patent/CN120046007B/en
Publication of CN120046007A publication Critical patent/CN120046007A/en
Application granted granted Critical
Publication of CN120046007B publication Critical patent/CN120046007B/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
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/10—Pre-processing; Data cleansing
    • 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
    • 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
    • G06F18/2433—Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/254—Fusion techniques of classification results, e.g. of results related to same input data
    • G06F18/256—Fusion techniques of classification results, e.g. of results related to same input data of results relating to different input data, e.g. multimodal recognition
    • 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
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • 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
    • 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
    • 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
    • G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
    • 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
    • G06N3/0985—Hyperparameter optimisation; Meta-learning; Learning-to-learn
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/26—Government or public services
    • G06Q50/265—Personal security, identity or safety
    • G—PHYSICS
    • G08—SIGNALLING
    • G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
    • G08B31/00—Predictive alarm systems characterised by extrapolation or other computation using updated historic data

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Business, Economics & Management (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Tourism & Hospitality (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Primary Health Care (AREA)
  • Computer Security & Cryptography (AREA)
  • Emergency Management (AREA)
  • General Business, Economics & Management (AREA)
  • Probability & Statistics with Applications (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Educational Administration (AREA)
  • Development Economics (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

The invention relates to the technical field of building electrical safety, and discloses a building electrical safety risk intelligent diagnosis method and system based on big data, wherein operation condition data of a building electrical system are collected in real time; the method comprises the steps of collecting operation condition data, cleaning the collected operation condition data to remove noise, carrying out normalization processing on the operation condition data after denoising to obtain preprocessed data, carrying out feature extraction on the preprocessed data, extracting time domain features and frequency domain features, obtaining energy distribution under different scales through wavelet transformation, capturing electrical signal anomalies to obtain multidimensional electrical feature data, inputting the multidimensional electrical feature data into an intelligent diagnosis model, diagnosing the electrical safety state of a building through the intelligent diagnosis model, outputting a safety diagnosis result, determining the electrical safety risk level of the building according to the safety diagnosis result, and sending corresponding early warning information based on the safety risk level.

Description

Intelligent diagnosis method and system for building electrical safety risk based on big data
Technical Field
The invention relates to the technical field of building electrical safety, in particular to an intelligent building electrical safety risk diagnosis method and system based on big data.
Background
In modern buildings, electrical systems are complex and bulky, covering many subsystems such as lighting, power, sockets, fire electricity, etc. Traditional electrical safety detection relies on manual regular inspection, and by means of experience and simple instrument measurement, the method is low in efficiency and high in subjectivity, potential risks such as line aging, overload, short circuit, ground fault and the like are difficult to capture comprehensively in real time, electrical fire or equipment damage can be caused instantaneously, and hidden dangers in the manual inspection interval period are difficult to discover timely.
Disclosure of Invention
The invention aims to solve the problems and designs an intelligent diagnosis method and system for building electrical safety risks based on big data.
The first aspect of the invention provides a building electrical security risk intelligent diagnosis method based on big data, which comprises the following steps:
Acquiring parameter original data of the building electrical system in real time through a plurality of sensors, docking with a building automation system, acquiring operation state information, acquiring electric energy quality index data through an electric energy quality monitor, and integrating the parameter original data, the operation state information and the electric energy quality index data to obtain operation condition data of the building electrical system;
Cleaning the collected operation condition data to remove noise, and carrying out normalization processing on the denoised operation condition data to obtain preprocessed data;
Extracting features of the preprocessed data, extracting time domain features and frequency domain features, obtaining energy distribution under different scales through wavelet transformation, capturing electrical signal anomalies, and obtaining multidimensional electrical feature data;
Inputting the multidimensional electrical characteristic data into an intelligent diagnosis model, diagnosing the electrical safety state of the building through the intelligent diagnosis model, and outputting a safety diagnosis result;
And determining the electrical safety risk level of the building according to the safety diagnosis result, and sending corresponding early warning information based on the safety risk level.
Optionally, in a first implementation manner of the first aspect of the present invention, the cleaning the collected operation condition data to remove noise, and normalizing the denoised operation condition data to obtain preprocessed data includes:
acquiring collected operation condition data, traversing the operation condition data to check whether a missing value exists in each data record, filling the detected missing value by adopting a linear interpolation method, and estimating the missing value by linear calculation according to the effective data points adjacent to each missing value in front of and behind the position of each missing value;
Detecting abnormal values of the filled data by a filtering algorithm based on statistical characteristics, calculating the mean value and standard deviation of each data type, calculating the absolute value of the difference value between each data point and the mean value, and judging the corresponding data point as the abnormal value if the absolute value of the difference value of a certain data point is larger than a threshold value, wherein the threshold value is three times the standard deviation;
Correcting the detected abnormal value by adopting smoothing processing, and removing noise from the data processed by the missing value and the abnormal value by adopting a median filter to obtain denoised data;
And carrying out normalization processing on the denoised data by using a Min-Max normalization method to obtain preprocessed data.
Optionally, in a second implementation manner of the first aspect of the present invention, the feature extracting is performed on the preprocessed data, a time domain feature and a frequency domain feature are extracted, energy distribution under different scales is obtained through wavelet transformation, and electrical signal anomalies are captured, so as to obtain multidimensional electrical feature data, including:
Acquiring the preprocessed data, and arranging the preprocessed data in time sequence to obtain a data sequence;
Calculating the mean value, standard deviation and effective value of the data sequence, traversing the data sequence to find the maximum value, the minimum value and the peak value, obtaining a peak factor based on the ratio of the peak value to the effective value, calculating the difference value of adjacent voltage values, calculating the standard deviation of the difference value to obtain a voltage fluctuation standard deviation, and integrating to obtain a time domain feature;
converting the time domain characteristics of the data sequence into frequency domain characteristics by applying fast Fourier transform, calculating the frequency position of each subharmonic according to the fundamental frequency for the frequency domain sequences, extracting the amplitude value of the corresponding frequency position, and counting the proportion of the amplitude value of each subharmonic to the amplitude value of the fundamental wave to obtain harmonic frequency spectrum distribution;
determining a low frequency range, extracting frequency domain components in the range, and analyzing the amplitude and the phase change of the low frequency components along with time, wherein the low frequency range is lower than 100Hz;
The energy distribution under different scales is obtained through wavelet transformation, and the electrical signal abnormality is captured to obtain multidimensional electrical characteristic data.
Optionally, in a third implementation manner of the first aspect of the present invention, the obtaining energy distribution under different scales by wavelet transformation, capturing electrical signal anomalies, and obtaining multidimensional electrical feature data includes:
performing N layers of wavelet decomposition on the data sequence by adopting Haar wavelet, and decomposing a signal into an approximate component and a detail component;
the energy of the approximate component and the detail component of each layer is calculated, the extracted time domain features, the frequency domain features and the energy distribution under different scales obtained by wavelet transformation are integrated to form a multi-dimensional feature vector which is used as final multi-dimensional electrical feature data.
Optionally, in a fourth implementation manner of the first aspect of the present invention, the multi-dimensional electrical characteristic data is input into an intelligent diagnosis model, and the building electrical safety state is diagnosed through the intelligent diagnosis model, and a safety diagnosis result is output, including:
constructing an intelligent diagnosis model based on a DNN network and an LSTM network, and inputting the multidimensional electrical characteristic data into the intelligent diagnosis model;
The multi-dimensional electrical characteristic data are input into an input layer of a DNN network, layer-by-layer calculation is carried out in a hidden layer of the DNN network, the output of the upper layer is taken as input for neurons of each layer, weighted summation is carried out, bias is added, nonlinear factors are introduced through a ReLU activation function, the output of the layer is obtained, and the output of an output layer of the DNN network is obtained after the calculation of a plurality of hidden layers;
Taking the output of the DNN network output layer as the input of the LSTM network, calculating the LSTM network according to the current input, the hiding state and the cell state at the last moment in each time step, capturing the time sequence relevance of the electrical parameters, and obtaining the final output of the LSTM network after the processing of all time steps;
and calculating the probability of the building electrical in different safety states through a Softmax function by the final output of the LSTM network, selecting the safety state with the highest probability as the safety state of the current building electrical according to the calculated probability, sorting the determined safety state and the corresponding probability value, and outputting a safety diagnosis result.
Optionally, in a fifth implementation manner of the first aspect of the present invention, the constructing a smart diagnostic model based on the DNN network and the LSTM network includes:
Acquiring multi-modal data of a historical multi-source sensor, carrying out feature screening on the collected multi-modal data by using a dynamic feature selection algorithm, evaluating the importance of each feature, acquiring the features with high correlation with building electrical safety diagnosis, and removing redundant features, wherein the multi-modal data is the previous running state of a building electrical system under different working conditions;
dynamically adjusting weights of data obtained after feature screening by adopting an attention mechanism, and performing noise reduction treatment by using a self-adaptive noise reduction mode combining wavelet transformation and an countermeasure generation network to obtain a training set;
Inputting the training set into a DNN network for training, and introducing a self-attention mechanism into the DNN network to pay attention to the interrelationship among different features;
Taking the output of the DNN network as the input of the LSTM network, introducing Peephole connection and Zoneout regularization into the LSTM network, wherein Peephole connection adds additional connection for a forgetting gate, an input gate and an output gate, and can receive the information of the cell state;
a differential gating mechanism is adopted to construct a dynamic feature fusion layer, the features output by the DNN network and the LSTM network are fused, and weights are dynamically distributed according to the importance of different features in the current diagnosis task;
And constructing a four-dimensional optimization objective function, and performing super-parameter optimization by adopting MOSHO algorithm to obtain an optimal solution so as to construct an intelligent diagnosis model.
Optionally, in a sixth implementation manner of the first aspect of the present invention, the constructing a four-dimensional optimization objective function, performing super-parametric optimization by adopting MOSHO algorithm to obtain an optimal solution, so as to construct an intelligent diagnosis model, includes:
Determining four indexes of the four-dimensional optimization objective function as diagnosis precision, reasoning speed, memory occupation and model stability, and distributing weights for the four indexes;
Randomly initializing a certain number of particles in a super-parameter space of a model to be optimized, wherein each particle represents a group of super-parameter combinations, and distributing an initial speed and position for each particle;
According to the initial hyper-parameter combination and the corresponding objective function value, carrying out initial training on the Bayesian model, applying the hyper-parameter combination represented by each particle in the current particle swarm to the intelligent diagnosis model for training and evaluation, and calculating the corresponding four-dimensional optimization objective function value;
comparing the objective function value of the current position of each particle with the objective function value of the historical optimal position for each particle, and if the objective function value of the current position is better, updating the historical optimal position of the particle;
Comparing the objective function values of the historic optimal positions of all particles, and selecting the position with the optimal objective function value as a global optimal position;
Introducing quantum behavior simulation to improve the updating rule of the particle swarm, predicting and obtaining a super-parameter combination with better objective function value in a super-parameter space by utilizing a trained Bayesian model, and adding the predicted super-parameter combination into the particle swarm as new particles;
adding the super-parameter combination of the newly added particles and the corresponding objective function values into the training data of the Bayesian model, and retraining the Bayesian model;
Checking whether the current iteration number reaches an upper limit, if so, stopping iteration, otherwise, returning to continue iteration optimization;
After iteration is finished, the super-parameter combination represented by the global optimal position is the optimal solution of the four-dimensional optimization objective function, and the intelligent diagnosis model is built by using the optimal super-parameter combination.
The invention provides a building electrical safety risk intelligent diagnosis system based on big data, which comprises a data acquisition module, a data preprocessing module, a feature extraction module, an intelligent diagnosis module and an early warning module, wherein the data acquisition module is used for acquiring parameter original data of a building electrical system in real time through various sensors, interfacing with a building automation system to acquire operation state information, communicating with an electric energy quality monitor to acquire electric energy quality index data, and integrating the parameter original data, the operation state information and the electric energy quality index data to obtain operation condition data of the building electrical system;
the data preprocessing module is used for cleaning the collected operation condition data to remove noise, and carrying out normalization processing on the denoised operation condition data to obtain preprocessed data;
The feature extraction module is used for extracting features of the preprocessed data, extracting time domain features and frequency domain features, obtaining energy distribution under different scales through wavelet transformation, capturing electrical signal anomalies and obtaining multidimensional electrical feature data;
The intelligent diagnosis module is used for inputting the multidimensional electrical characteristic data into an intelligent diagnosis model, diagnosing the electrical safety state of the building through the intelligent diagnosis model and outputting a safety diagnosis result;
And the early warning module is used for determining the electrical safety risk level of the building according to the safety diagnosis result and sending corresponding early warning information based on the safety risk level.
Optionally, in a first implementation manner of the second aspect of the present invention, the data preprocessing module includes a filling sub-module, a calculating sub-module, a correcting sub-module and a normalizing sub-module, where the filling sub-module is configured to obtain collected operation condition data, traverse the operation condition data to check whether a missing value exists in each data record, fill the detected missing value by using a linear interpolation method, and estimate the missing value by linear calculation according to valid data points adjacent to each missing value in front of and behind the location;
the computing sub-module is used for detecting abnormal values of the filled data based on a filtering algorithm with statistical characteristics, computing the average value and standard deviation of each data type, computing the absolute value of the difference value between each data point and the average value, and judging the corresponding data point as the abnormal value if the absolute value of the difference value of a certain data point is larger than a threshold value, wherein the threshold value is three times the standard deviation;
The correction submodule is used for correcting the detected abnormal value by adopting smoothing processing, removing noise from the data processed by the missing value and the abnormal value by adopting a median filter, and obtaining denoised data;
And the normalization processing sub-module is used for carrying out normalization processing on the denoised data by adopting a Min-Max normalization method to obtain preprocessed data.
Optionally, in a second implementation manner of the second aspect of the present invention, the feature extraction module includes an arrangement sub-module, a traversal sub-module, a conversion sub-module, an analysis sub-module, and a capturing sub-module, where the arrangement sub-module is configured to obtain the preprocessed data, and arrange the preprocessed data in time sequence to obtain a data sequence;
The traversing sub-module is used for calculating the mean value, standard deviation and effective value of the data sequence, traversing the data sequence to find the maximum value, the minimum value and the peak value, obtaining a peak factor based on the ratio of the peak value to the effective value, calculating the difference value of adjacent voltage values, calculating the standard deviation of the difference value to obtain the standard deviation of voltage fluctuation, and integrating to obtain the time domain feature;
the conversion sub-module is used for converting the time domain characteristics of the data sequence into frequency domain characteristics by applying fast Fourier transform, calculating the frequency position of each subharmonic according to the fundamental frequency for the frequency domain sequences, extracting the amplitude value of the corresponding frequency position, and counting the proportion of the amplitude value of each subharmonic to the amplitude value of the fundamental wave to obtain harmonic frequency spectrum distribution;
The analysis submodule is used for determining a low-frequency range to extract frequency domain components in the range and analyzing the amplitude and the phase change of the low-frequency components along with time, wherein the low-frequency range is lower than 100Hz;
and the capturing submodule is used for obtaining energy distribution under different scales through wavelet transformation and capturing electrical signal anomalies to obtain multidimensional electrical characteristic data.
According to the technical scheme, parameter raw data of a building electrical system are acquired in real time through various sensors and are in butt joint with a building automation system to obtain operation state information, an electric energy quality monitor is connected to acquire electric energy quality index data, the parameter raw data, the operation state information and the electric energy quality index data are integrated to obtain operation condition data of the building electrical system, the acquired operation condition data are cleaned to remove noise, the denoised operation condition data are subjected to normalization processing to obtain preprocessed data, the preprocessed data are subjected to feature extraction, time domain features and frequency domain features are extracted, energy distribution under different scales is obtained through wavelet transformation, electric signal anomalies are captured to obtain multidimensional electric feature data, the multidimensional electric feature data are input into an intelligent diagnosis model, the building electric safety state is diagnosed through the intelligent diagnosis model, a safety diagnosis result is output, a building electric safety risk level is determined according to the safety diagnosis result, corresponding early warning information is sent based on the safety risk level, and the building electric safety hidden danger is accurately and real-time identified to ensure the building electric safety.
Drawings
Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention.
FIG. 1 is a schematic diagram of a first embodiment of a building electrical security risk intelligent diagnosis method based on big data according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a second embodiment of a building electrical security risk intelligent diagnosis method based on big data according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a third embodiment of a building electrical security risk intelligent diagnosis method based on big data according to an embodiment of the present invention;
Fig. 4 is a schematic structural diagram of an intelligent diagnosis system for building electrical security risk based on big data according to an embodiment of the present invention.
Detailed Description
The terms "first," "second," "third," "fourth" and the like in the description and in the claims and in the above drawings, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments described herein may be implemented in other sequences than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, apparatus, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed or inherent to such process, method, article, or apparatus.
For ease of understanding, the following describes a specific flow of an embodiment of the present invention, please refer to fig. 1 for a schematic diagram of a first embodiment of a building electrical security risk intelligent diagnosis method based on big data, which specifically includes the following steps:
Step 101, acquiring parameter original data of a building electrical system in real time through various sensors, docking with a building automation system, acquiring operation state information, acquiring electric energy quality index data through an electric energy quality monitor, and integrating the parameter original data, the operation state information and the electric energy quality index data to obtain operation condition data of the building electrical system;
In the embodiment, the current transformer is focused on capturing the current in the line, and real-time values of the current are monitored at any time, such as conventional current during stable operation and current fluctuation caused by load abrupt change; the voltage transformer acquires voltage conditions, accurately measures the voltage value of each place, the temperature sensor is attached to the surface of the electrical equipment or a key circuit part, senses the temperature change of equipment and a circuit in real time, because the excessive temperature is often the front of an electrical fault, such as the overload of the circuit, poor contact and other problems, causes the rapid rise of the temperature, the leakage sensor closely monitors leakage current, a Building Automation System (BAS) controls the operation states of the electrical equipment in various aspects, the opening and closing states of the circuit breaker directly reflect the on-off conditions of the circuit, if the circuit breaker is frequently opened and closed, the problems of short circuit, overload and the like of the downstream circuit can be predicted, the frequent actions of the contactor can suggest that the controlled equipment is unstable or the control logic is disordered, the equipment which is continuously operated for a long time is more likely to have fatigue wear, overheat and other faults, the electric energy quality monitor comprehensively detects the electric energy quality, the harmonic content is one of key detection indexes, the harmonic content is more prominent along with the use of a large number of nonlinear loads (such as electronic equipment and frequency conversion devices and the like) in modern buildings, the harmonic content can interfere with the operation of the electrical equipment, even the three-phase power factors can be damaged, the three-phase power factors can be balanced, the power factors can be lost due to the fact that the overload of the other equipment is not normally monitored, and the power equipment is not normally balanced, and the power factors can be balanced, and the power factors are not normally balanced, and the power equipment can be damaged, affecting the power supply stability.
102, Cleaning the collected operation condition data to remove noise, and carrying out normalization processing on the denoised operation condition data to obtain preprocessed data;
In the embodiment, acquired operation condition data are acquired, the operation condition data are traversed to check whether missing values exist in each data record, a linear interpolation method is adopted to fill the detected missing values, the missing values are estimated according to the positions of the missing values and the adjacent effective data points, the abnormal values of the filled data are detected based on a filtering algorithm with statistical characteristics, the mean value and standard deviation of each data type are calculated, the absolute value of the difference value between each data point and the mean value is calculated, if the absolute value of the difference value of a certain data point is larger than a threshold value, the corresponding data point is judged to be the abnormal value, the threshold value is three times the standard deviation, the detected abnormal values are corrected by adopting smoothing processing, noise is removed from the data processed by the missing values and the abnormal values by adopting a median filter, the denoised data are normalized by adopting a Min-Max normalization method, and the preprocessed data are obtained.
Step 103, extracting characteristics of the preprocessed data, extracting time domain characteristics and frequency domain characteristics, obtaining energy distribution under different scales through wavelet transformation, capturing electrical signal anomalies, and obtaining multidimensional electrical characteristic data;
104, inputting the multidimensional electrical characteristic data into an intelligent diagnosis model, diagnosing the electrical safety state of the building through the intelligent diagnosis model, and outputting a safety diagnosis result;
And 105, determining the electrical safety risk level of the building according to the safety diagnosis result, and sending corresponding early warning information based on the safety risk level.
According to the embodiment of the invention, through an intelligent diagnosis model, whether the intelligent diagnosis model is used for carrying out gradual change of electric parameters caused by tiny electric leakage and long-term overload caused by line aging or complex hidden faults such as harmonic interference caused by nonlinear load, the maintenance time is greatly shortened, the power failure loss is reduced, hierarchical management is carried out on building electric facilities, conventional monitoring is continuously kept for low-risk areas, resource waste caused by excessive operation and maintenance is avoided, a detailed diagnosis flow is started in medium-risk areas, potential hazards are purposefully checked, accurate maintenance is carried out by reasonably allocating manpower and material resources, emergency measures such as power failure are immediately taken when the intelligent diagnosis model is used for judging high-risk areas, professional team repair is organized, the dynamic operation and maintenance strategy based on risks enables operation and maintenance resources to be optimally allocated, on the premise of guaranteeing electric safety, the operation and maintenance efficiency is maximally improved, the operation and maintenance cost is reduced, safety accidents such as electric fire, equipment damage and power interruption caused by various factors such as line faults, equipment aging and electric energy quality problems are effectively prevented, the life safety of personnel and property safety of a building are provided for a person in a building, and the intelligent diagnosis is carried out in real time, the lighting, the air conditioning, the electric system is well-condition is guaranteed, and the service system is in a good condition is guaranteed to be in the service condition of a good condition, and is satisfied, and the service is safe, and is in the service condition is safe, and has good service, and has a service condition is used in the building and a service condition is in the building.
Referring to fig. 2, a second embodiment of a building electrical security risk intelligent diagnosis method based on big data according to an embodiment of the present invention is shown, where the method includes:
Step 201, acquiring preprocessed data, and arranging the preprocessed data in time sequence to obtain a data sequence;
Step 202, calculating the mean value, standard deviation and effective value of a data sequence, traversing the data sequence to find the maximum value, the minimum value and the peak value, obtaining a peak factor based on the ratio of the peak value to the effective value, calculating the difference value of adjacent voltage values, calculating the standard deviation of the difference value to obtain a voltage fluctuation standard deviation, and integrating to obtain a time domain feature;
Step 203, converting the time domain features of the data sequence into frequency domain features by applying fast Fourier transform, calculating the frequency position of each subharmonic according to the fundamental frequency for the frequency domain sequences, extracting the amplitude value of the corresponding frequency position, and counting the proportion of the amplitude value of each subharmonic to the amplitude value of the fundamental wave to obtain harmonic frequency spectrum distribution;
step 204, determining a low frequency range to extract frequency domain components in the range, and analyzing the change of amplitude and phase of the low frequency components along with time;
In this embodiment, the low frequency range is below 100Hz.
Step 205, obtaining energy distribution under different scales through wavelet transformation, capturing electrical signal anomalies, and obtaining multidimensional electrical characteristic data.
In the embodiment, N layers of wavelet decomposition is carried out on the data sequence by adopting Haar wavelet to decompose the signal into approximate components and detail components, the energy of the approximate components and the detail components of each layer is calculated, and the extracted time domain features, frequency domain features and energy distribution under different scales obtained by wavelet transformation are integrated to form a multi-dimensional feature vector which is used as final multi-dimensional electrical feature data.
Referring to fig. 3, a third embodiment of a building electrical security risk intelligent diagnosis method based on big data according to an embodiment of the present invention is shown, where the method includes:
Step 301, constructing an intelligent diagnosis model based on a DNN network and an LSTM network, and inputting multidimensional electrical characteristic data into the intelligent diagnosis model;
In the embodiment, the multi-modal data of a historical multi-source sensor is obtained, a dynamic characteristic selection algorithm is used for carrying out characteristic screening on the collected multi-modal data, importance of each characteristic is evaluated, characteristics with high correlation with building electrical safety diagnosis are obtained, redundant characteristics are removed, the multi-modal data are operation states of a building electrical system under different working conditions in the past, a attention mechanism is used for dynamically adjusting weights of the data obtained after characteristic screening, a self-adaptive noise reduction mode combining wavelet transformation and an countermeasure generation network is used for carrying out noise reduction treatment to obtain a training set, the training set is input into a DNN network for training, a self-attention mechanism is introduced into the DNN network for paying attention to the correlation among different characteristics, the output of the DNN network is used as the input of an LSTM network, the LSTM network is introduced into Peephole connection and Zoneout regularization, peephole connection is formed by adding additional connection for a forgetting gate, an input gate and an output gate, information of a cell state can be received, a differential gating mechanism is used for constructing a dynamic characteristic fusion layer, the DNN network and the LSTM network output characteristics are combined, the characteristics are used for carrying out noise reduction treatment in a self-adaptive noise reduction mode of the countermeasure generation network according to different characteristics in the current diagnosis, a dynamic diagnosis performance optimization function is carried out by adopting a four-dimensional optimization function 35, and an optimal diagnosis parameter is optimized construction is carried out.
In the embodiment, four indexes of a four-dimensional optimization objective function are determined to be diagnosis precision, reasoning speed, memory occupation and model stability, weights are distributed to the four indexes, a certain number of particles are randomly initialized in a super-parameter space of a model to be optimized, each particle represents a group of super-parameter combinations and is distributed with initial speed and position, initial training is carried out on the Bayesian model according to the initial super-parameter combinations and corresponding objective function values, the super-parameter combinations represented by each particle in a current particle swarm are applied to an intelligent diagnosis model for training and evaluation, the corresponding four-dimensional optimization objective function values are calculated, the objective function value of the current position of each particle is compared with the objective function value of a historical optimal position, if the objective function value of the current position is better, the historical optimal position of the particle is updated, the objective function value of all particles is compared, the optimal position of the objective function value is selected to serve as a global optimal position, the updated rule of the particle swarm is improved by introducing quantum behavior simulation, the trained Bayesian model is predicted in the super-parameter space to obtain a super-parameter combination of a better objective, the predicted super-parameter combinations are applied to the intelligent diagnosis model for training and evaluation, the super-parameter combinations are added to the new super-parameter combinations to the new particle swarm model, if the super-parameter combinations are more optimal to the optimal position is more optimal, if the optimal position is reached, the iteration conditions are met, if the iteration conditions are met, the iteration conditions are met, and the iteration conditions are met are continued, and the iteration conditions are reached, and constructing an intelligent diagnosis model by using the optimal super-parameter combination.
Step 302, inputting multidimensional electrical characteristic data into an input layer of a DNN network, performing layer-by-layer calculation in a hidden layer of the DNN network, taking the output of the previous layer as input for neurons of each layer, performing weighted summation, adding bias, introducing a nonlinear factor through a ReLU activation function to obtain the output of the layer, and obtaining the output of an output layer of the DNN network after the calculation of a plurality of hidden layers;
step 303, taking the output of the DNN network output layer as the input of the LSTM network, calculating the LSTM network according to the current input, the hidden state at the last moment and the cell state at each time step, capturing the time sequence relevance of the electrical parameters, and obtaining the final output of the LSTM network after the processing of all time steps;
And 304, calculating the probability of the building electrical in different safety states through a Softmax function by the final output of the LSTM network, selecting the safety state with the highest probability as the safety state of the current building electrical according to the calculated probability, sorting the judged safety state and the corresponding probability value, and outputting a safety diagnosis result.
Referring to fig. 4, a schematic structural diagram of a building electrical safety risk intelligent diagnosis system based on big data provided by the embodiment of the invention includes a data acquisition module, a data preprocessing module, a feature extraction module, an intelligent diagnosis module and an early warning module, wherein the data acquisition module is used for acquiring parameter original data of a building electrical system in real time through various sensors, interfacing with a building automation system to acquire operation state information, communicating with an electric energy quality monitor to acquire electric energy quality index data, and integrating the parameter original data, the operation state information and the electric energy quality index data to obtain operation condition data of the building electrical system;
the data preprocessing module is used for cleaning the collected operation condition data to remove noise, and carrying out normalization processing on the denoised operation condition data to obtain preprocessed data;
The feature extraction module is used for extracting features of the preprocessed data, extracting time domain features and frequency domain features, obtaining energy distribution under different scales through wavelet transformation, capturing electrical signal anomalies and obtaining multidimensional electrical feature data;
the intelligent diagnosis module is used for inputting the multidimensional electrical characteristic data into an intelligent diagnosis model, diagnosing the electrical safety state of the building through the intelligent diagnosis model and outputting a safety diagnosis result;
And the early warning module is used for determining the electrical safety risk level of the building according to the safety diagnosis result and sending corresponding early warning information based on the safety risk level.
In this embodiment, the data preprocessing module includes a filling sub-module, a calculating sub-module, a correcting sub-module and a normalizing sub-module, where the filling sub-module is configured to obtain collected operation condition data, traverse the operation condition data to check whether there is a missing value in each data record, fill the detected missing value by using a linear interpolation method, and estimate the missing value by linear calculation according to valid data points adjacent to each missing value;
the computing sub-module is used for detecting abnormal values of the filled data based on a filtering algorithm with statistical characteristics, computing the average value and standard deviation of each data type, computing the absolute value of the difference value between each data point and the average value, and judging the corresponding data point as the abnormal value if the absolute value of the difference value of a certain data point is larger than a threshold value, wherein the threshold value is three times the standard deviation;
The correction submodule is used for correcting the detected abnormal value by adopting smoothing processing, removing noise from the data processed by the missing value and the abnormal value by adopting a median filter, and obtaining denoised data;
And the normalization processing sub-module is used for carrying out normalization processing on the denoised data by adopting a Min-Max normalization method to obtain preprocessed data.
In this embodiment, the feature extraction module includes an arrangement sub-module, a traversal sub-module, a conversion sub-module, an analysis sub-module, and a capture sub-module, where the arrangement sub-module is configured to obtain data after preprocessing, and arrange the preprocessed data in time sequence to obtain a data sequence;
The traversing sub-module is used for calculating the mean value, standard deviation and effective value of the data sequence, traversing the data sequence to find the maximum value, the minimum value and the peak value, obtaining a peak factor based on the ratio of the peak value to the effective value, calculating the difference value of adjacent voltage values, calculating the standard deviation of the difference value to obtain the standard deviation of the voltage fluctuation, and integrating to obtain the time domain feature;
the conversion sub-module is used for converting the time domain characteristics of the data sequence into frequency domain characteristics by applying fast Fourier transform, calculating the frequency position of each subharmonic according to the fundamental frequency for the frequency domain sequences, extracting the amplitude value of the corresponding frequency position, and counting the proportion of the amplitude value of each subharmonic to the amplitude value of the fundamental wave to obtain harmonic frequency spectrum distribution;
The analysis submodule is used for determining a low-frequency range to extract frequency domain components in the range and analyzing the amplitude and the phase change of the low-frequency components along with time, wherein the low-frequency range is lower than 100Hz;
and the capturing submodule is used for obtaining energy distribution under different scales through wavelet transformation and capturing electrical signal anomalies to obtain multidimensional electrical characteristic data.
The foregoing has shown and described the basic principles, principal features and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the above-described embodiments, and that the above-described embodiments and descriptions are only preferred embodiments of the present invention, and are not intended to limit the invention, and that various changes and modifications may be made therein without departing from the spirit and scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.

Claims (8)

1.基于大数据的建筑电气安全风险智能诊断方法，其特征在于，所述基于大数据的建筑电气安全风险智能诊断方法包括以下步骤：1. A method for intelligent diagnosis of building electrical safety risks based on big data, characterized in that the method comprises the following steps: 通过参数原始数据、运行状态信息和电能质量指标数据得到建筑电气系统的运行工况数据；Obtain the operating condition data of the building electrical system through the original parameter data, operating status information and power quality index data; 对采集的运行工况数据进行清洗和归一化处理，以得到预处理后的数据；Clean and normalize the collected operating condition data to obtain pre-processed data; 将预处理后的数据按时间顺序排列，得到数据序列；Arrange the preprocessed data in chronological order to obtain a data sequence; 获取数据序列的有效值和峰值，基于有效值与峰值之比得到峰值因数，利用相邻电压值的差值的标准差得到电压波动标准差，并基于峰值因数和电压波动标准差得到时域特征；Obtain the effective value and peak value of the data sequence, obtain the crest factor based on the ratio of the effective value to the peak value, obtain the voltage fluctuation standard deviation using the standard deviation of the difference between adjacent voltage values, and obtain the time domain characteristics based on the crest factor and the voltage fluctuation standard deviation; 将数据序列的时域特征转换为频域特征后，根据基波频率确定各次谐波的频率位置，并提取对应频率位置的幅值，统计各次谐波幅值占基波幅值的比例，得到谐波频谱分布；After converting the time domain features of the data sequence into frequency domain features, the frequency position of each harmonic is determined according to the fundamental frequency, and the amplitude of the corresponding frequency position is extracted. The proportion of the amplitude of each harmonic to the amplitude of the fundamental is counted to obtain the harmonic spectrum distribution; 通过小波变换对数据序列进行多尺度分解，得到不同尺度下的能量分布，并捕捉电气信号中的异常波动，将时域特征、频域特征以及不同尺度下的能量分布进行整合得到多维电气特征数据；The data sequence is decomposed into multiple scales through wavelet transform to obtain energy distribution at different scales, and abnormal fluctuations in electrical signals are captured. The time domain characteristics, frequency domain characteristics and energy distribution at different scales are integrated to obtain multi-dimensional electrical characteristic data. 将所述多维电气特征数据输入智能诊断模型中，通过智能诊断模型诊断建筑电气安全状态，输出安全诊断结果；Inputting the multi-dimensional electrical characteristic data into an intelligent diagnostic model, diagnosing the building electrical safety status through the intelligent diagnostic model, and outputting a safety diagnosis result; 根据所述安全诊断结果确定建筑电气安全风险等级，并基于所述安全风险等级发送对应的预警信息；Determine the building electrical safety risk level according to the safety diagnosis result, and send corresponding warning information based on the safety risk level; 所述智能诊断模型基于DNN网络和LSTM网络构建，具体包括：The intelligent diagnosis model is built based on the DNN network and the LSTM network, and specifically includes: 对收集到的不同工况下的运行状态数据进行特征筛选；Perform feature screening on the collected operating status data under different working conditions; 先对特征筛选后的运行状态数据使用结合小波变换与对抗生成网络的自适应降噪方式进行降噪处理，然后对降噪处理后的数据采用注意力机制进行动态调整权重，并利用处理后的数据得到训练集；First, the running status data after feature screening is denoised using an adaptive denoising method that combines wavelet transform and adversarial generative network. Then, the attention mechanism is used to dynamically adjust the weights of the denoised data, and the processed data is used to obtain the training set. 将训练集输入DNN网络中进行训练，在DNN网络引入自注意力机制；Input the training set into the DNN network for training, and introduce the self-attention mechanism into the DNN network; 将DNN网络的输出作为LSTM网络的输入，LSTM网络引入Peephole连接和Zoneout正则化，其中，Peephole连接为对遗忘门、输入门和输出门添加用于接收细胞状态的信息的连接；The output of the DNN network is used as the input of the LSTM network. The LSTM network introduces Peephole connection and Zoneout regularization. The Peephole connection is a connection added to the forget gate, input gate and output gate to receive information about the cell state. 采用可微分门控机制，融合DNN网络和LSTM网络输出的特征；Adopting a differentiable gating mechanism to fuse the features output by the DNN network and the LSTM network; 在待优化模型的超参数空间中随机初始化一定数量的粒子；Randomly initialize a certain number of particles in the hyperparameter space of the model to be optimized; 比较所有粒子的历史最优位置的目标函数值，选择目标函数值最优的位置作为全局最优位置；Compare the objective function values of the historical optimal positions of all particles and select the position with the best objective function value as the global optimal position; 全局最优位置所代表的超参数组合即为四维优化目标函数的最优解，使用最优超参数组合构建智能诊断模型。The hyperparameter combination represented by the global optimal position is the optimal solution to the four-dimensional optimization objective function. The optimal hyperparameter combination is used to construct an intelligent diagnostic model. 2.如权利要求1所述的基于大数据的建筑电气安全风险智能诊断方法，其特征在于，所述对采集的运行工况数据进行清洗和归一化处理，以得到预处理后的数据，包括：2. The method for intelligent diagnosis of building electrical safety risks based on big data according to claim 1, wherein the step of cleaning and normalizing the collected operating condition data to obtain pre-processed data comprises: 获取采集的运行工况数据，对所述运行工况数据进行遍历以检查每条数据记录中是否存在缺失值，采用线性插值法对检测到的缺失值进行填充，针对每个缺失值的所在位置，根据其前后相邻的有效数据点，通过线性计算来估计缺失值；Obtaining collected operating condition data, traversing the operating condition data to check whether there are missing values in each data record, filling the detected missing values using linear interpolation, and estimating the missing value through linear calculation based on the valid data points before and after the missing value; 基于统计特性的滤波算法检测填充后数据的异常值，先计算每个数据类型的均值和标准差，再计算每个数据点与均值的差值的绝对值，若某数据点的差值的绝对值大于阈值，则将对应的数据点判定为异常值，其中阈值为三倍的标准差；A statistical filtering algorithm is used to detect outliers in the padded data. The mean and standard deviation of each data type are first calculated, and then the absolute value of the difference between each data point and the mean is calculated. If the absolute value of the difference between a data point and the mean is greater than a threshold, the corresponding data point is considered an outlier. The threshold is three times the standard deviation. 采用平滑处理对检测到的异常值进行修正，并采用中值滤波器对经过缺失值和异常值处理后的数据去除噪声，得到去噪后的数据；Smoothing is used to correct the detected outliers, and a median filter is used to remove noise from the data after missing values and outliers are processed to obtain denoised data; 采用Min-Max归一化方法对去噪后的数据进行归一化处理，得到预处理后的数据。The Min-Max normalization method is used to normalize the denoised data to obtain the preprocessed data. 3.如权利要求1所述的基于大数据的建筑电气安全风险智能诊断方法，其特征在于，所述通过小波变换对数据序列进行多尺度分解，得到不同尺度下的能量分布，并捕捉电气信号中的异常波动，将时域特征、频域特征以及不同尺度下的能量分布进行整合得到多维电气特征数据，包括：3. The method for intelligent diagnosis of building electrical safety risks based on big data according to claim 1, wherein the method comprises performing multi-scale decomposition of the data sequence by wavelet transform to obtain energy distribution at different scales, capturing abnormal fluctuations in the electrical signal, and integrating time domain features, frequency domain features, and energy distribution at different scales to obtain multi-dimensional electrical characteristic data, including: 对所述数据序列采用Haar小波进行 N 层小波分解，将信号分解为近似分量和细节分量；Performing N-layer wavelet decomposition on the data sequence using Haar wavelet to decompose the signal into approximate components and detail components; 计算每一层的近似分量和细节分量的能量，将提取的时域特征、频域特征以及小波变换得到的不同尺度下的能量分布进行整合，形成一个多维的特征向量，作为最终的多维电气特征数据。The energy of the approximate component and detail component of each layer is calculated, and the extracted time domain features, frequency domain features and energy distribution at different scales obtained by wavelet transform are integrated to form a multidimensional feature vector as the final multidimensional electrical feature data. 4.如权利要求1所述的基于大数据的建筑电气安全风险智能诊断方法，其特征在于，将所述多维电气特征数据输入智能诊断模型中，通过智能诊断模型诊断建筑电气安全状态，输出安全诊断结果，包括：4. The method for intelligent diagnosis of building electrical safety risks based on big data according to claim 1, wherein the multi-dimensional electrical characteristic data is input into an intelligent diagnosis model, the building electrical safety status is diagnosed by the intelligent diagnosis model, and a safety diagnosis result is output, comprising: 基于DNN网络和LSTM网络构建智能诊断模型，将所述多维电气特征数据输入所述智能诊断模型中；Building an intelligent diagnosis model based on a DNN network and an LSTM network, and inputting the multi-dimensional electrical characteristic data into the intelligent diagnosis model; 多维电气特征数据输入到DNN网络的输入层，在DNN网络的隐藏层中进行逐层计算，对于每一层的神经元，将上一层的输出作为输入，进行加权求和，并加上偏置，通过ReLU激活函数引入非线性因素，得到该层的输出，经过多个隐藏层的计算后，得到DNN网络输出层的输出；Multi-dimensional electrical characteristic data is input to the input layer of the DNN network. The hidden layers of the DNN network are then calculated layer by layer. For each layer of neurons, the output of the previous layer is used as input, weighted summed, and biased. Nonlinear factors are introduced through the ReLU activation function to obtain the output of the layer. After calculations in multiple hidden layers, the output of the DNN network output layer is obtained. 将DNN网络输出层的输出作为LSTM网络的输入，在每个时间步，LSTM网络根据当前输入并结合上一时刻的隐藏状态和细胞状态进行计算，捕捉电气参数的时间序列关联性，经过所有时间步的处理后，得到LSTM网络的最终输出；The output of the DNN network output layer is used as the input of the LSTM network. At each time step, the LSTM network performs calculations based on the current input and combines the hidden state and cell state of the previous moment to capture the time series correlation of electrical parameters. After processing all time steps, the final output of the LSTM network is obtained. 将LSTM网络的最终输出通过Softmax函数计算建筑电气处于不同安全状态的概率，根据计算得到的概率，选择概率最大的安全状态作为当前建筑电气的安全状态，将判定的安全状态以及对应的概率值进行整理，输出安全诊断结果。The final output of the LSTM network is used through the Softmax function to calculate the probability of the building electrical system being in different safety states. Based on the calculated probability, the safety state with the highest probability is selected as the current safety state of the building electrical system. The determined safety states and corresponding probability values are sorted out to output the safety diagnosis results. 5.如权利要求4所述的基于大数据的建筑电气安全风险智能诊断方法，其特征在于，所述基于DNN网络和LSTM网络构建智能诊断模型，还包括：5. The method for intelligent diagnosis of building electrical safety risks based on big data according to claim 4, wherein the method for constructing an intelligent diagnosis model based on a DNN network and an LSTM network further comprises: 构建四维优化目标函数，采用MOSHO算法进行超参数优化，得到最优解，以构建智能诊断模型。A four-dimensional optimization objective function is constructed, and the MOSHO algorithm is used to optimize the hyperparameters to obtain the optimal solution to build an intelligent diagnosis model. 6.如权利要求5所述的基于大数据的建筑电气安全风险智能诊断方法，其特征在于，所述构建四维优化目标函数，采用MOSHO算法进行超参数优化，得到最优解，以构建智能诊断模型，包括：6. The method for intelligent diagnosis of building electrical safety risks based on big data according to claim 5, wherein the step of constructing a four-dimensional optimization objective function and using the MOSHO algorithm to perform hyperparameter optimization to obtain an optimal solution and construct an intelligent diagnosis model comprises: 确定四维优化目标函数的四个指标为诊断精度、推理速度、内存占用和模型稳定性，并分配为四个指标分配权重；Determine the four indicators of the four-dimensional optimization objective function as diagnostic accuracy, inference speed, memory usage, and model stability, and assign weights to the four indicators; 在待优化模型的超参数空间中随机初始化一定数量的粒子，每个粒子代表一组超参数组合，为每个粒子分配初始速度和位置；Randomly initialize a certain number of particles in the hyperparameter space of the model to be optimized. Each particle represents a set of hyperparameter combinations, and assign an initial velocity and position to each particle. 根据初始的超参数组合和对应的目标函数值，对贝叶斯优化算法进行初始化训练，将当前粒子群中的每个粒子所代表的超参数组合应用于智能诊断模型进行训练和评估，计算对应的四维优化目标函数值；Based on the initial hyperparameter combination and the corresponding objective function value, the Bayesian optimization algorithm is initialized and trained. The hyperparameter combination represented by each particle in the current particle swarm is applied to the intelligent diagnosis model for training and evaluation, and the corresponding four-dimensional optimization objective function value is calculated. 对于每个粒子，比较每个粒子当前位置的目标函数值与历史最优位置的目标函数值，如果当前位置的目标函数值更优，则更新该粒子的历史最优位置；For each particle, compare the objective function value of the current position of each particle with the objective function value of the historical optimal position. If the objective function value of the current position is better, update the historical optimal position of the particle. 比较所有粒子的历史最优位置的目标函数值，选择目标函数值最优的位置作为全局最优位置；Compare the objective function values of the historical optimal positions of all particles and select the position with the best objective function value as the global optimal position; 引入量子行为模拟机制，并利用训练好的初始化模型，在超参数空间中预测获得更优目标函数值的超参数组合，将预测的超参数组合作为新的粒子加入到粒子群中；A quantum behavior simulation mechanism is introduced, and the trained initialization model is used to predict the hyperparameter combination that obtains a better objective function value in the hyperparameter space. The predicted hyperparameter combination is added as a new particle to the particle swarm. 将新加入粒子群的超参数组合和对应的目标函数值加入到基于贝叶斯优化算法的训练数据中，重新进行训练；Add the newly added hyperparameter combination and the corresponding objective function value of the particle swarm to the training data based on the Bayesian optimization algorithm and retrain; 检查当前迭代次数是否达到上限，如果满足终止条件，则停止迭代；否则，返回继续迭代优化；Check whether the current number of iterations has reached the upper limit. If the termination condition is met, stop the iteration; otherwise, return to continue iterative optimization. 迭代结束后，全局最优位置所代表的超参数组合即为四维优化目标函数的最优解，使用最优超参数组合构建智能诊断模型。After the iteration, the hyperparameter combination represented by the global optimal position is the optimal solution to the four-dimensional optimization objective function, and the optimal hyperparameter combination is used to build an intelligent diagnostic model. 7.基于大数据的建筑电气安全风险智能诊断系统，其特征在于，所述基于大数据的建筑电气安全风险智能诊断系统包括数据采集模块、数据预处理模块、特征提取模块、智能诊断模块和预警模块，其中：7. A big data-based intelligent diagnostic system for building electrical safety risks, characterized in that the system comprises a data acquisition module, a data preprocessing module, a feature extraction module, an intelligent diagnostic module, and an early warning module, wherein: 数据采集模块，用于通过参数原始数据、运行状态信息和电能质量指标数据得到建筑电气系统的运行工况数据；The data acquisition module is used to obtain the operating condition data of the building electrical system through the original parameter data, operating status information and power quality index data; 数据预处理模块，用于对采集的运行工况数据进行清洗和归一化处理，以得到预处理后的数据；The data preprocessing module is used to clean and normalize the collected operating condition data to obtain preprocessed data; 特征提取模块，用于将预处理后的数据按时间顺序排列，得到数据序列；获取数据序列的有效值和峰值，基于有效值与峰值之比得到峰值因数，利用相邻电压值的差值的标准差得到电压波动标准差，并基于峰值因数和电压波动标准差得到时域特征；将数据序列的时域特征转换为频域特征后，根据基波频率确定各次谐波的频率位置，并提取对应频率位置的幅值，统计各次谐波幅值占基波幅值的比例，得到谐波频谱分布；通过小波变换对数据序列进行多尺度分解，得到不同尺度下的能量分布，并捕捉电气信号中的异常波动，将时域特征、频域特征以及不同尺度下的能量分布进行整合得到多维电气特征数据；The feature extraction module is used to arrange the pre-processed data in chronological order to obtain a data sequence; obtain the effective value and peak value of the data sequence, obtain the peak factor based on the ratio of the effective value to the peak value, obtain the standard deviation of the voltage fluctuation using the standard deviation of the difference between adjacent voltage values, and obtain the time domain characteristics based on the peak factor and the standard deviation of the voltage fluctuation; after converting the time domain characteristics of the data sequence into frequency domain characteristics, the frequency position of each harmonic is determined according to the fundamental frequency, and the amplitude of the corresponding frequency position is extracted, and the proportion of the amplitude of each harmonic to the amplitude of the fundamental wave is calculated to obtain the harmonic spectrum distribution; the data sequence is decomposed into multiple scales through wavelet transform to obtain the energy distribution at different scales, and abnormal fluctuations in the electrical signal are captured. The time domain characteristics, frequency domain characteristics and energy distribution at different scales are integrated to obtain multi-dimensional electrical characteristic data; 智能诊断模块，用于将所述多维电气特征数据输入智能诊断模型中，通过智能诊断模型诊断建筑电气安全状态，输出安全诊断结果：The intelligent diagnosis module is used to input the multi-dimensional electrical characteristic data into the intelligent diagnosis model, diagnose the building electrical safety status through the intelligent diagnosis model, and output the safety diagnosis results: 预警模块，用于根据所述安全诊断结果确定建筑电气安全风险等级，并基于所述安全风险等级发送对应的预警信息；An early warning module is used to determine the building electrical safety risk level according to the safety diagnosis result and send corresponding early warning information based on the safety risk level; 所述智能诊断模型基于DNN网络和LSTM网络构建，具体包括：The intelligent diagnosis model is built based on the DNN network and the LSTM network, and specifically includes: 对收集到的不同工况下的运行状态数据进行特征筛选；Perform feature screening on the collected operating status data under different working conditions; 先对特征筛选后的运行状态数据使用结合小波变换与对抗生成网络的自适应降噪方式进行降噪处理，然后对降噪处理后的数据采用注意力机制进行动态调整权重，并利用处理后的数据得到训练集；First, the running status data after feature screening is denoised using an adaptive denoising method that combines wavelet transform and adversarial generative network. Then, the attention mechanism is used to dynamically adjust the weights of the denoised data, and the processed data is used to obtain the training set. 将训练集输入DNN网络中进行训练，在DNN网络引入自注意力机制；Input the training set into the DNN network for training, and introduce the self-attention mechanism into the DNN network; 将DNN网络的输出作为LSTM网络的输入，LSTM网络引入Peephole连接和Zoneout正则化，其中，Peephole连接为对遗忘门、输入门和输出门添加用于接收细胞状态的信息的连接；The output of the DNN network is used as the input of the LSTM network. The LSTM network introduces Peephole connection and Zoneout regularization. The Peephole connection is a connection added to the forget gate, input gate and output gate to receive information about the cell state. 采用可微分门控机制，融合DNN网络和LSTM网络输出的特征；Adopting a differentiable gating mechanism to fuse the features output by the DNN network and the LSTM network; 在待优化模型的超参数空间中随机初始化一定数量的粒子；Randomly initialize a certain number of particles in the hyperparameter space of the model to be optimized; 比较所有粒子的历史最优位置的目标函数值，选择目标函数值最优的位置作为全局最优位置；Compare the objective function values of the historical optimal positions of all particles and select the position with the best objective function value as the global optimal position; 全局最优位置所代表的超参数组合即为四维优化目标函数的最优解，使用最优超参数组合构建智能诊断模型。The hyperparameter combination represented by the global optimal position is the optimal solution to the four-dimensional optimization objective function. The optimal hyperparameter combination is used to construct an intelligent diagnostic model. 8.如权利要求7所述的基于大数据的建筑电气安全风险智能诊断系统，其特征在于，所述数据预处理模块包括填充子模块、计算子模块、修正子模块和归一化处理子模块，其中：8. The big data-based intelligent diagnostic system for building electrical safety risks according to claim 7, wherein the data preprocessing module comprises a filling submodule, a calculation submodule, a correction submodule, and a normalization processing submodule, wherein: 填充子模块，用于获取采集的运行工况数据，对所述运行工况数据进行遍历以检查每条数据记录中是否存在缺失值，采用线性插值法对检测到的缺失值进行填充，针对每个缺失值的所在位置，根据其前后相邻的有效数据点，通过线性计算来估计缺失值；A filling submodule is used to obtain the collected operating condition data, traverse the operating condition data to check whether there are missing values in each data record, fill the detected missing values using linear interpolation, and estimate the missing value through linear calculation based on the valid data points before and after the location of each missing value; 计算子模块，用于基于统计特性的滤波算法检测填充后数据的异常值，先计算每个数据类型的均值和标准差，再计算每个数据点与均值的差值的绝对值，若某数据点的差值的绝对值大于阈值，则将对应的数据点判定为异常值，其中阈值为三倍的标准差；The calculation submodule is used to detect outliers in the filled data based on the statistical characteristics of the filtering algorithm. The mean and standard deviation of each data type are first calculated, and then the absolute value of the difference between each data point and the mean is calculated. If the absolute value of the difference of a data point is greater than the threshold, the corresponding data point is determined to be an outlier, where the threshold is three times the standard deviation. 修正子模块，用于采用平滑处理对检测到的异常值进行修正，并采用中值滤波器对经过缺失值和异常值处理后的数据去除噪声，得到去噪后的数据；The correction submodule is used to correct the detected outliers by smoothing, and to remove noise from the data after missing values and outliers are processed by using a median filter to obtain denoised data; 归一化处理子模块，用于采用Min-Max归一化方法对去噪后的数据进行归一化处理，得到预处理后的数据。The normalization processing submodule is used to normalize the denoised data using the Min-Max normalization method to obtain preprocessed data.
CN202510518162.2A 2025-04-24 2025-04-24 Intelligent diagnosis method and system for building electrical safety risk based on big data Active CN120046007B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202510518162.2A CN120046007B (en) 2025-04-24 2025-04-24 Intelligent diagnosis method and system for building electrical safety risk based on big data

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202510518162.2A CN120046007B (en) 2025-04-24 2025-04-24 Intelligent diagnosis method and system for building electrical safety risk based on big data

Publications (2)

Publication Number Publication Date
CN120046007A CN120046007A (en) 2025-05-27
CN120046007B true CN120046007B (en) 2025-07-29

Family

ID=95760268

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202510518162.2A Active CN120046007B (en) 2025-04-24 2025-04-24 Intelligent diagnosis method and system for building electrical safety risk based on big data

Country Status (1)

Country Link
CN (1) CN120046007B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN120277589B (en) * 2025-06-10 2025-09-02 山东德源电力科技股份有限公司 Semi-solid-state circuit breaker with intelligent collection function of full power data

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110443364A (en) * 2019-06-21 2019-11-12 深圳大学 A kind of deep neural network multitask hyperparameter optimization method and device
CN118884097A (en) * 2024-09-28 2024-11-01 广东南海电力设计院工程有限公司 A PT terminal fault detection method and monitoring system

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9778080B2 (en) * 2013-04-29 2017-10-03 Emerson Electric (Us) Holding Corporation (Chile) Limitada Selective decimation and analysis of oversampled data
CN119026900B (en) * 2024-07-22 2025-10-17 中钢集团马鞍山矿山研究总院股份有限公司 Explosion safety protection method based on unstable tunnel tunneling
CN119783676B (en) * 2025-03-11 2025-05-13 成都中医药大学 Method for identifying roles in acupoint compatibility based on named entity

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110443364A (en) * 2019-06-21 2019-11-12 深圳大学 A kind of deep neural network multitask hyperparameter optimization method and device
CN118884097A (en) * 2024-09-28 2024-11-01 广东南海电力设计院工程有限公司 A PT terminal fault detection method and monitoring system

Also Published As

Publication number Publication date
CN120046007A (en) 2025-05-27

Similar Documents

Publication Publication Date Title
Schlechtingen et al. Wind turbine condition monitoring based on SCADA data using normal behavior models. Part 1: System description
Liu et al. Technology development and commercial applications of industrial fault diagnosis system: a review
CN118428741B (en) An intelligent error-proof processing method suitable for special operations and emergency operations
Zhang et al. Fault diagnosis and prognosis using wavelet packet decomposition, Fourier transform and artificial neural network
Niu et al. Intelligent condition monitoring and prognostics system based on data-fusion strategy
Wang et al. Intelligent diagnosis method for rolling element bearing faults using possibility theory and neural network
Goodall et al. Concepts and techniques for railway condition monitoring
CN108604360A (en) Facility method for monitoring abnormality and its system
CN114519923B (en) Intelligent diagnosis and early warning method and system for power plant
CN109492790A (en) Wind turbines health control method based on neural network and data mining
CN120046007B (en) Intelligent diagnosis method and system for building electrical safety risk based on big data
CN111103137A (en) Wind turbine gearbox fault diagnosis method based on deep neural network
CN119024225A (en) Short-circuit fault judgment method based on the change of positive sequence voltage amplitude at load end
Samhouri et al. An intelligent machine condition monitoring system using time-based analysis: neuro-fuzzy versus neural network
CN120705643A (en) A method, device, equipment and storage medium for intelligent identification of early cable faults based on dynamic threshold
Aikhuele et al. Application of intuitionistic fuzzy topsis model for troubleshooting an offshore patrol boat engine
Wang Key techniques in intelligent predictive maintenance (IPdM)–a framework of intelligent faults diagnosis and prognosis system (IFDaPS)
CN112417612B (en) Wind power gear box degradation state tracking and failure aggregation risk assessment method
Du et al. A SOM based anomaly detection method for wind turbines health management through SCADA data
CN112380782A (en) Rotating equipment fault prediction method based on mixed indexes and neural network
Abbasi et al. Condition based maintenance of oil and gas equipment: A review
CN119556040A (en) An intelligent power equipment health monitoring system integrated with sensor network
Song et al. Anomaly detection of wind turbine generator based on temporal information
Sivakumar et al. Implementation of VLSI model as a tool in diagnostics of slowly varying process parameters which affect the performance of steam turbine
CN116165480A (en) Fault studying and judging method for secondary equipment of distribution network

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
GR01 Patent grant
GR01 Patent grant