CN114818859B - Heating power pipe network condition diagnosis method, device, terminal equipment and storage medium - Google Patents
Heating power pipe network condition diagnosis method, device, terminal equipment and storage mediumInfo
- Publication number
- CN114818859B CN114818859B CN202210280833.2A CN202210280833A CN114818859B CN 114818859 B CN114818859 B CN 114818859B CN 202210280833 A CN202210280833 A CN 202210280833A CN 114818859 B CN114818859 B CN 114818859B
- Authority
- CN
- China
- Prior art keywords
- acgan
- pipe network
- condition
- loss function
- svm
- 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
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
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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
- 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
- G06N3/08—Learning methods
-
- 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
- G06Q10/00—Administration; Management
- G06Q10/20—Administration of product repair or maintenance
-
- 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
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Tourism & Hospitality (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- General Business, Economics & Management (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- Primary Health Care (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
本发明公开了一种热力管网状况诊断方法、装置、终端设备和存储介质。包括：将含热力管网状况类别标签的随机噪声数据输入ACGAN生成器，得到伪样本，将已获得的热力管网图像真样本和伪样本输入ACGAN判别器进行判别真假训练和分类训练，并输入SVM模型进行分类训练，将ACGAN判别器和SVM模型的分类损失函数整合为统一损失函数，根据统一损失函数，更新所述ACGAN判别器参数，逐步降低SVM损失函数对统一损失函数的影响，在统一损失函数不受所述SVM损失函数影响后，使用ACGAN独立进行分类训练，根据训练好的ACGAN判别器，得到热力管网状况诊断模型，将采集的热力管网图像输入热力管网状况诊断模型，得到状况诊断结果。能够对热力管网状况的诊断更加准确，效率更高。
The present invention discloses a method, device, terminal equipment and storage medium for diagnosing the condition of a heat pipe network. The method includes: inputting random noise data containing a heat pipe network condition category label into an ACGAN generator to obtain a pseudo sample, inputting the obtained heat pipe network image true sample and pseudo sample into an ACGAN discriminator for true and false discrimination training and classification training, and inputting into an SVM model for classification training, integrating the classification loss functions of the ACGAN discriminator and the SVM model into a unified loss function, updating the ACGAN discriminator parameters according to the unified loss function, gradually reducing the influence of the SVM loss function on the unified loss function, and after the unified loss function is not affected by the SVM loss function, using ACGAN to independently perform classification training, obtaining a heat pipe network condition diagnosis model according to the trained ACGAN discriminator, and inputting the collected heat pipe network image into the heat pipe network condition diagnosis model to obtain a condition diagnosis result. The diagnosis of the heat pipe network condition can be more accurate and efficient.
Description
Technical Field
The invention relates to the technical field of heat pipe network fault diagnosis, in particular to a heat pipe network condition diagnosis method, a heat pipe network condition diagnosis device, terminal equipment and a storage medium.
Background
The prior fault diagnosis technology applied to the heating power pipe network mainly comprises a traditional machine learning method and a deep neural network method, wherein the traditional machine learning method comprises algorithms such as a support vector machine, a random forest, a multi-layer perceptron and the like, the neural network mainly comprises a VGGNet, googLeNet, resNet-class convolutional neural network, the learner is trained through normal samples and fault samples with balanced quantity, the learner can learn the characteristics of the fault samples and the normal samples, and after field data are obtained, the learner is used for correctly classifying the data, so that fault diagnosis is realized. The number of failure samples of the thermal power pipe network is far lower than that of normal samples, so that the traditional machine learning method and the deep neural network method are difficult to obtain enough failure training samples, and more accurate failure sample characteristics are difficult to extract. Generating the countermeasure network (GENERATIVE ADVERSARIAL Networks, abbreviated as GAN) provides a new idea for solving the imbalance of the dataset, the GAN can generate a large number of pseudo samples similar to the real samples, so that the dataset is expanded, ACGAN (auxliary CLASSIFIER GENERATIVE ADVERSARIAL Networks, assisted classification generation of the countermeasure network) is an improvement of the GAN, and the input data types can be accurately classified while the dataset is expanded.
However, ACGAN still has difficulty in early training of sample imbalance, and the parameters of the discriminant in ACGAN remain to be optimized in early training ACGAN, so that the classification effect on faults is poor. The conventional support vector machine SVM (Support Vector Machine) has a remarkable effect of classifying a dataset with a small sample size, but has a poor effect of classifying a dataset with a large sample size and has low efficiency.
Disclosure of Invention
In order to at least partially solve the technical problems existing in the prior art, the inventor makes the invention, and through the specific embodiment, a method, a device, a terminal device and a storage medium for diagnosing the condition of a heating power pipe network are provided.
In a first aspect, an embodiment of the present invention provides a training method for a condition diagnosis model of a heating power pipe network, including the following steps:
Inputting random noise data containing a heating pipe network condition type label into a ACGAN generator to obtain a pseudo sample containing the condition type label, inputting the obtained heating pipe network image true sample and the pseudo sample newly generated by the ACGAN generator and containing the condition type label into a ACGAN discriminator for discrimination and false training and classification training, inputting an SVM model for classification training, integrating a ACGAN discriminator and a classification loss function of the SVM model into a unified loss function, updating parameters of the ACGAN generator according to ACGAN discrimination of true and false, updating parameters of the ACGAN discriminator according to the unified loss function, and updating parameters of the SVM model according to the SVM model classification loss function;
after the steps are iterated for a plurality of times, reducing the influence of an SVM loss function on the unified loss function once every iteration for a certain number of times until the unified loss function is not influenced by the SVM loss function;
and after the unified loss function is not influenced by the SVM loss function, carrying out independent classification training by using ACGAN, and obtaining a heating power pipe network condition diagnosis model according to the trained ACGAN discriminator.
Optionally, before inputting the random noise data containing the heat pipe network condition type label into the ACGAN generator, the method includes the following steps:
Obtaining a real sample of a heat pipe network image;
setting a condition type label according to the condition type of the real sample of the heat pipe network image;
generating random noise data containing the condition category label;
Constructing an SVM model according to the condition category;
And constructing ACGAN a model.
Optionally, the obtaining a real sample of the heat pipe network image includes the following steps:
respectively collecting a plurality of heat distribution pipe network image samples under different condition categories, wherein the condition categories comprise normal pipe wall leakage, pipe bending deformation and pipe heat preservation layer damage;
At least one item of data enhancement processing is carried out on the collected heat pipe network image samples under different condition categories, so that the heat pipe network image samples after the data enhancement processing are obtained, wherein the data enhancement processing comprises image rotation, cutting, similar splicing, affine transformation, verification noise increase, illumination and shadow increase, contrast adjustment and saturation adjustment;
Carrying out random combination on the heat pipe network image samples subjected to data enhancement treatment to obtain heat pipe network image samples subjected to random combination;
And marking the collected heat pipe network image samples under different conditions, the heat pipe network image samples subjected to data enhancement processing and the heat pipe network image samples subjected to random combination as heat pipe network image true samples.
Optionally, the constructing an SVM model according to the condition category includes the following steps:
and constructing N-1 SVM sub-models according to the condition category number N, wherein N is a positive integer.
Optionally, the constructing ACGAN a model includes the following steps:
creating ACGAN a generator using a deconvolution neural network;
A ACGAN arbiter is built using a convolutional neural network.
Optionally, the generating random noise data including the condition category label includes the following steps:
A plurality of random noise data are collected, each random noise data is added with one condition category label, and each condition category label is added to the plurality of random noise data.
Optionally, the integrating the ACGAN discriminant and the classification loss function of the SVM model into a unified loss function includes the following steps:
determining a classification loss function L C of the ACGAN arbiter;
Determining a classification loss function L SVM of the SVM model;
According to the classification Loss functions L C and L SVM, the expression for determining the unified Loss function Loss is:
Loss=(1-λ)LSVM+λLC
Wherein λ is a confidence ratio of the classification result of the ACGAN discriminator, and an arbitrary number greater than 0 and less than 1 is set as an initial value of λ.
Optionally, the reducing the influence of the SVM loss function on the unified loss function once per iteration for a certain number of times until the unified loss function is not influenced by the SVM loss function includes the following steps:
Every iteration is performed for a certain number of times, lambda in the unified Loss function Loss expression is increased by a certain increment until lambda is increased to 1.
Optionally, after the unified loss function is not affected by the SVM loss function, performing classification training independently by using ACGAN, and obtaining a heat pipe network condition diagnosis model according to a trained ACGAN discriminator, including the following steps:
After lambda is increased to 1, inputting the obtained real sample of the heat pipe network image and the pseudo sample which is newly generated by the ACGAN generator and contains the condition type label into a ACGAN discriminator for classification training, and updating ACGAN discriminator parameters according to a unified loss function;
and after the training termination condition is met, stopping updating ACGAN the parameters of the discriminator, and obtaining a heating power pipe network condition diagnosis model according to the trained ACGAN discriminator.
Optionally, the obtaining the heating power pipe network condition diagnosis model according to the trained ACGAN discriminator includes the following steps:
And stripping the trained ACGAN discriminator from the ACGAN model, canceling the output of the trained ACGAN discriminator on discrimination of true and false, and only reserving classification output to obtain the heating power pipe network condition diagnosis model.
In a second aspect, an embodiment of the present invention provides a method for diagnosing a condition of a heating power pipe network, including the steps of:
And inputting the acquired heat pipe network image into a heat pipe network condition diagnosis model obtained by the method to obtain a condition diagnosis result output by the heat pipe network condition diagnosis model.
In a third aspect, an embodiment of the present invention provides a training device for a diagnosis model of a heating power pipe network, including:
the auxiliary classification module is used for inputting random noise data containing the condition type labels of the heating power pipe network into the ACGAN generator to obtain pseudo samples containing the condition type labels, inputting the obtained true samples of the heating power pipe network images and the pseudo samples newly generated by the ACGAN generator and containing the condition type labels into the ACGAN discriminator for discrimination training and classification training, and inputting the SVM model for classification training;
the loss function integration module is used for integrating the classification loss functions of the ACGAN discriminator and the SVM model into a unified loss function;
The parameter updating module is used for updating parameters of the ACGAN generator according to ACGAN to judge whether the model is true or false, updating parameters of the ACGAN discriminator according to the unified loss function, and updating parameters of the SVM model according to the SVM model classification loss function;
and the independent classification module is used for independently performing classification training by ACGAN after the unified loss function is not influenced by the SVM loss function, and obtaining a heating power pipe network condition diagnosis model according to the trained ACGAN discriminator.
Optionally, the method further comprises:
The classification preparation module is used for obtaining a real sample of the heat pipe network image, setting a condition type label according to the condition type of the real sample of the heat pipe network image, generating random noise data containing the condition type label, constructing an SVM model according to the condition type, and constructing a ACGAN model.
Optionally, the loss function integration module includes:
a discriminator loss function determining unit, configured to determine a classification loss function L C of the ACGAN discriminators;
an SVM penalty function determination unit configured to determine a classification penalty function L SVM of the SVM model;
the Loss function integrating unit is configured to determine, according to the classification Loss functions L C and L SVM, that the expression of the unified Loss function Loss is:
Loss=(1-λ)LSVM+λLC
Wherein λ is a confidence ratio of the classification result of the ACGAN discriminator, and an arbitrary number greater than 0 and less than 1 is set as an initial value of λ.
Optionally, the classification preparation module includes:
The real sample obtaining unit is used for respectively collecting a plurality of heat pipe network image samples under different condition categories, wherein the condition categories comprise normal heat pipe network image samples, pipe wall leakage, pipe bending deformation and pipe heat preservation damage, carrying out at least one item of data enhancement processing on the collected heat pipe network image samples under different condition categories to obtain heat pipe network image samples after the data enhancement processing, wherein the data enhancement processing comprises image rotation, cutting, similar splicing, affine transformation, increasing check noise, increasing illumination and shadow, adjusting contrast and adjusting saturation;
The condition type label setting unit is used for setting a condition type label according to the condition type of the real sample of the heat pipe network image;
An SVM model construction unit for constructing N-1 SVM sub-models according to the condition category number N, wherein N is a positive integer;
ACGAN a model building unit for building ACGAN generator using deconvolution neural network, building ACGAN discriminator using convolution neural network;
A noise data obtaining unit, configured to collect a plurality of random noise data, where each random noise data is added with one status type tag, and each status type tag is added to the plurality of random noise data.
In a fourth aspect, an embodiment of the present invention provides a heat pipe network condition diagnosis apparatus, including:
the image acquisition module is used for acquiring a heating power pipe network image;
the condition diagnosis module is used for inputting the acquired heat pipe network image into the heat pipe network condition diagnosis model obtained by the method, and obtaining the condition diagnosis result output by the heat pipe network condition diagnosis model.
Based on the same inventive concept, the embodiment of the invention also provides a terminal device, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the heating power pipe network condition diagnosis model training method when executing the computer program.
Based on the same inventive concept, the embodiment of the invention also provides a computer storage medium, wherein the computer storage medium stores computer executable instructions, and the computer executable instructions realize the heating power pipe network condition diagnosis model training method when executed.
Based on the same inventive concept, the embodiment of the invention also provides a terminal device, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the heating power pipe network condition diagnosis method when executing the computer program.
Based on the same inventive concept, the embodiment of the invention also provides a computer storage medium, wherein the computer storage medium stores computer executable instructions, and the computer executable instructions realize the heating power pipe network condition diagnosis method when executed.
The technical scheme provided by the embodiment of the invention has the beneficial effects that at least:
The embodiment of the invention provides a heating power pipe network condition diagnosis model training method, which is characterized in that a large number of simulation samples with the same sample characteristics as real fault samples are supplemented through a ACGAN generator, the problem that the number of the fault samples is unbalanced with that of normal samples is solved, the training effect of a ACGAN discriminator is improved, the characteristics of the fault samples are extracted more accurately, a condition diagnosis model with more accurate condition diagnosis is obtained, the condition diagnosis accuracy is improved, ACGAN is combined with an SVM, the characteristics of accurate and efficient classification effect of the SVM model when the sample size is smaller are utilized, the classification training of ACGAN is assisted by the classification result of the SVM in the early stage of ACGAN training, the defect that the classification effect of the ACGAN discriminator is to be optimized when the early stage parameter is increased is overcome, the classification effect of the ACGAN discriminator is gradually optimized, the classification effect of the SVM is poorer when the sample size is larger, the efficiency is very low, and therefore the ACGAN discriminator starts to independently train after updating multiple parameters, the advantages of the two are combined, and the diagnosis model of the heating power pipe network condition diagnosis is more accurate and the efficiency is higher.
The embodiment of the invention provides a heating power pipe network condition diagnosis method, which is characterized in that a large number of pseudo samples are obtained for training by simulating real samples of a heating power pipe network image by using a heating power pipe network condition diagnosis model, the more accurate characteristics of fault samples are extracted, and the advantages of accurate classification and high efficiency of an SVM model are combined in early training, so that the diagnosis of the heating power pipe network condition by using the heating power pipe network condition diagnosis model is more accurate and has higher efficiency.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims thereof as well as the appended drawings.
The technical scheme of the invention is further described in detail through the drawings and the embodiments.
Drawings
The accompanying drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate the invention and together with the embodiments of the invention, serve to explain the invention. In the drawings:
FIG. 1 is a flowchart of a training method for a condition diagnosis model of a heating power pipe network in an embodiment of the invention;
FIG. 2 is a flowchart of a method for diagnosing a condition of a heat distribution network according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a thermal pipe network condition diagnostic model training and diagnostic process in accordance with an embodiment of the present invention;
FIG. 4 is a block diagram of a training device for diagnosing the condition of a heating power pipe network according to an embodiment of the invention;
FIG. 5 is a block diagram of a heat pipe network condition diagnosing apparatus according to an embodiment of the present invention;
Fig. 6 is a schematic structural diagram of a terminal device in an embodiment of the present invention.
Detailed Description
Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
In order to solve the problems existing in the prior art, the embodiment of the invention provides a method, a device, terminal equipment and a storage medium for diagnosing the condition of a heating power pipe network.
Example 1
The first embodiment of the invention provides a training method for a heating power pipe network condition diagnosis model, which has a flow shown in a figure 1 and comprises the following steps:
Step S101, inputting random noise data containing a heating power pipe network condition type label into a ACGAN generator to obtain a pseudo sample containing the condition type label, inputting the obtained heating power pipe network image true sample and the pseudo sample newly generated by the ACGAN generator and containing the condition type label into a ACGAN discriminator for discrimination and true and false training and classifying training, inputting an SVM model for classifying training, integrating a classification loss function of the ACGAN discriminator and the SVM model into a unified loss function, updating parameters of the ACGAN generator according to the ACGAN discrimination and true and false condition, updating parameters of the ACGAN discriminator according to the unified loss function, and updating parameters of the SVM model according to the SVM model classification loss function;
Optionally, before inputting the random noise data containing the heat pipe network condition type label into the ACGAN generator, the method comprises the following steps:
Obtaining a real sample of a heat pipe network image;
setting a condition type label according to the condition type of the real sample of the heat pipe network image;
generating random noise data containing the condition category label;
Constructing an SVM model according to the condition category;
And constructing ACGAN a model.
Optionally, obtaining a real sample of the heat pipe network image includes the following steps:
respectively collecting a plurality of heat distribution pipe network image samples under different condition categories, wherein the condition categories comprise normal pipe wall leakage, pipe bending deformation and pipe heat preservation layer damage;
At least one item of data enhancement processing is carried out on the collected heat pipe network image samples under different condition categories, so that the heat pipe network image samples after the data enhancement processing are obtained, wherein the data enhancement processing comprises image rotation, cutting, similar splicing, affine transformation, verification noise increase, illumination and shadow increase, contrast adjustment and saturation adjustment;
Carrying out random combination on the heat pipe network image samples subjected to data enhancement treatment to obtain heat pipe network image samples subjected to random combination;
And marking the collected heat pipe network image samples under different conditions, the heat pipe network image samples subjected to data enhancement processing and the heat pipe network image samples subjected to random combination as heat pipe network image true samples.
Optionally, constructing an SVM model according to the condition category includes the following steps:
And constructing N-1 SVM sub-models according to the condition category number N, wherein N is a positive integer.
Optionally, constructing ACGAN a model, including the steps of:
the deconvolution neural network is used to build ACGAN the generator and the convolution neural network is used to build ACGAN the arbiter.
Optionally, generating random noise data including the condition category label includes the steps of:
A plurality of random noise data are collected, each random noise data is added with one condition category label, and each condition category label is added to the plurality of random noise data.
Optionally, integrating the ACGAN discriminant and the classification loss function of the SVM model into a unified loss function, comprising the steps of:
determining a classification loss function L C of the ACGAN arbiter;
Determining a classification loss function L SVM of the SVM model;
According to the classification Loss functions L C and L SVM, the expression for determining the unified Loss function Loss is:
Loss=(1-λ)LSVM+λLC
Wherein λ is a confidence ratio of the classification result of the ACGAN discriminator, and an arbitrary number greater than 0 and less than 1 is set as an initial value of λ. The value of lambda is between 0 and 1, including 0 and 1.
Step S102, after iterating the step S101 for a plurality of times, reducing the influence of an SVM loss function on the unified loss function once every iteration for a certain number of times until the unified loss function is not influenced by the SVM loss function;
Optionally, reducing the influence of the SVM loss function on the unified loss function once every certain number of iterations until the unified loss function is not influenced by the SVM loss function, including the steps of:
Every iteration is performed for a certain number of times, lambda in the unified Loss function Loss expression is increased by a certain increment until lambda is increased to 1.
And step 103, after the unified loss function is not influenced by the SVM loss function, performing independent classification training by using ACGAN, and obtaining a heating power pipe network condition diagnosis model according to a trained ACGAN discriminator.
Optionally, after the unified loss function is not affected by the SVM loss function, performing classification training independently by using ACGAN, and obtaining a heat pipe network condition diagnosis model according to a trained ACGAN discriminator, including the following steps:
After lambda is increased to 1, inputting the obtained real sample of the heat pipe network image and the pseudo sample which is newly generated by the ACGAN generator and contains the condition type label into a ACGAN discriminator for classification training, and updating ACGAN discriminator parameters according to a unified loss function;
and after the training termination condition is met, stopping updating ACGAN the parameters of the discriminator, and obtaining a heating power pipe network condition diagnosis model according to the trained ACGAN discriminator.
Optionally, according to the trained ACGAN discriminator, a heating power pipe network condition diagnosis model is obtained, which comprises the following steps:
And stripping the trained ACGAN discriminator from the ACGAN model, canceling the output of the trained ACGAN discriminator on discrimination of true and false, and only reserving classification output to obtain the heating power pipe network condition diagnosis model.
For example, by mounting the high-definition camera and the infrared thermal imager on the robot dog, the robot dog performs data acquisition by traveling in the heat pipe network tunnel, and 20 normal samples and 20 fault samples of various types such as pipe wall leakage, pipe bending deformation, pipe insulation layer damage and the like are acquired in total.
The collected few samples are subjected to data enhancement processing, the adopted data enhancement method comprises but is not limited to image rotation, cutting, same-class sample splicing, affine transformation, verification noise increase, illumination and shadow increase, contrast adjustment, saturation adjustment and the like, in addition, the plurality of enhancement methods can be randomly combined to better realize the diversity of data, each class of data is increased to 200 through enhancement processing, and the 200 samples are used as real samples of the heat pipe network image.
And generating one-hot codes according to the number of the condition type labels. one-hot encoding is a process that converts category variables into a form that is readily available to machine learning algorithms. Normal samples and various fault samples are of 4 types, the corresponding one-hot codes are thus respectively [1, 0], [0,1, 0], [0, 1].
And taking the first 4 bits of random noise covered by the above one-hot codes respectively as a noise condition type label, wherein the first 4 bits are randomly sampled 110 times in normal distribution with a mean value of 1 and a standard deviation of 0 as initial noise data.
ACGAN generator G takes as input random noise containing a condition category label.
As a representative of conventional machine learning, support vector machines often exhibit excellent characteristics in small sample classification tasks, and therefore, can assist ACGAN in training by training an SVM classifier with small samples in the early stages of insufficient training samples. The SVM realizes the classification of the sample by mapping the data to a high-dimensional space and searching a maximum margin hyperplane in the high-dimensional space. The method comprises the steps of firstly adopting a first-level classification SVM to divide the samples into two main classes, wherein the first main class comprises normal samples and first class fault samples, and the second main class comprises second class and third class fault samples. The classified two types of samples are respectively reclassified by adopting a two-stage classification SVM, so that three SVMs are required to be trained.
The generating countermeasure network (GENERATIVE ADVERSARIAL Networks, GAN) is composed of a generator G and a arbiter D, and the darashi equalization is achieved through the game training of both. The generator takes random noise as input, the training aim is to learn the data distribution rule of the real sample X real so as to generate a false sample X fake which is similar to the real sample and can be used for false spurious, the discriminator takes the real sample and the false sample generated by the generator as input, and the training aim is to accurately judge whether the input is the real sample or the false sample. Ideally, the trained generator can generate a pseudo sample which is identical to the real sample, the discriminator cannot discriminate the authenticity of the sample, and the output discrimination probability is always 0.5. The objective function of GAN is:
LGAN＝arg minGmaxD{E(lnD(Xreal))+E(ln(1-D(Xfake)))}
Wherein X fake = G (z), z is random noise. Because GAN takes random noise as input without any constraint, the training process is divergent, and the mechanisms of generator and arbiter game training increase the risk of network collapse in training. In addition, GAN has significant effects in the field of data enhancement, but is difficult to be qualified for classification tasks.
ACGAN is an improved network for GAN, ACGAN is also composed of a generator G and a discriminator D, the generator takes random noise z as input, but class labels c are added compared with GAN to guide the generator to generate pseudo samples of different classes, the discriminator takes true samples of different classes and the pseudo samples generated by the generator as input, the training target is improved compared with GAN, and the discriminator of ACGAN is required to judge whether input data is true or not and judge the class to which the input data belongs.
The training of the generator aims at generating pseudo samples of different categories according to the sample labels, namely when the generated data passes through the discriminator, the output probability of the discriminator is expected to be as large as possible and is as close to 1 as possible, so the objective function of the generator can be expressed as:
LG-S=max E(ln(D(G(z))))
The training objective of the discriminator is to accurately discriminate the authenticity of the data, i.e. when the input is real data, the expected discrimination probability is as large as possible, and when the input generator generates pseudo data, the probability of the expected output is as small as possible, so the objective function of the discriminator can be expressed as:
LD-S=max E(ln(D(x)))+min E(ln(D(G(z))))
another training goal of the arbiter is to accurately classify the sample, i.e. when the real data and the dummy data pass through the arbiter, the accuracy of the arbiter classification is high, and the arbiter classification loss can be expressed as:
LC=max E(ln(D(x)))+max E(ln(D(G(z))))
To keep the objective functions consistent, the above objective functions are integrated into the correct data input source log-likelihood function L S and the correct class log-likelihood function L C:
LS＝E(ln(P(S＝real|Xreal)))+E(ln(P(S＝fake|Xfake)))
=E(ln(DS(x)))+E(ln(DS(G(z)))
LC＝E(ln(P(C＝c|Xreal)))+E(ln(P(C＝c|Xfake)))
=E(ln(DC(x)))+E(ln(DC(G(z)))
wherein S represents discrimination true or false, and C represents classification.
The objective function of generator G is maximize L C-LS and the objective function of arbiter D is maximize L C+LS.
ACGAN guides the generator and the arbiter to train games in a mode of maximum minimization, so that the generator can generate various sufficiently lifelike pseudo samples, and the arbiter can accurately judge data sources and accurately divide sample types.
The invention adopts deconvolution neural network as ACGAN generator. Deconvolution is also called transpose convolution, and compared with the process that the number of feature images is gradually increased and the specification is gradually reduced by general convolution operation, deconvolution operation can be regarded as the inverse process of convolution operation, namely the feature images are larger in specification and smaller in number by deconvolution operation. The generator adopts a large number of deconvolution operations, and under the condition that the input sample specification is smaller, information is continuously integrated according to the instruction of the loss function, so that the grower can generate fine and vivid pseudo samples. To prevent gradient explosions after the deconvolution operation, a batch normalization operation BatchNormalization was added.
The invention adopts a convolutional neural network as a ACGAN discriminator. Convolutional neural networks are a type of feed-forward neural network, which is a supervised model trained end-to-end. The convolution layer is mainly used for feature extraction, and the convolution kernel adopts a form of local connection and weight sharing, so that model parameters are greatly reduced, and network complexity and risk of overfitting are also reduced. After feature extraction is completed, the full-connection layer is adopted to re-fit the image features, so that loss of image feature information is reduced, the classifier receives an output value of the image feature information, and data classification is completed according to actual requirements. According to different classification tasks, two groups of full-connection layers with different parameters are adopted in parallel, one group is used for integrating data source information and outputting the probability of true data, and the other group is used for integrating data category information and outputting the probability of the category to which the data belongs. To prevent gradient explosion after each convolution operation, a batch normalization operation BatchNormalization is added, while to prevent excessive dependence of the model on individual neurons, a Dropout layer is added in the model after each round of activation of the function.
ACGAN is a single alternate iterative training, i.e., generator and arbiter alternate training.
The two targets of the discriminator D are that, firstly, efforts are made to discriminate whether the input is real data or data generated by the generator, and secondly, various data are classified, namely, whether the input is normal data or some kind of fault is discriminated.
A generation network is obtained by randomly configuring the weights of all nodes of the generator, a false sample data set is obtained by inputting a random signal and a class label, and the generation network G is at a disadvantage because the model is not optimized, so that the generated sample does not learn any rule of a real sample, and the sample set is easily identified as false data by a discrimination network. The training of the arbiter D is then completed using the real samples and the generated pseudo samples. The labels of the true and false sample sets are defined artificially, all class labels of the true sample set are set to be 1, all class labels of the false sample set are set to be 0, class labels are respectively set for all samples, the true and false data set is sent to a discrimination network D for training, and the training target of the discriminator is not only to distinguish true samples from false samples, but also to discriminate data classes.
The function of the generator G is to generate as realistic as possible various types of samples.
When the generating network is trained, the purpose of training can be achieved by combining the judging network, so that the training of the generating network is actually the training of the generating-judging network in series connection. The labels of the pseudo samples are changed to 1, namely the pseudo samples are considered to be true samples when the network training is generated, so that the aim of confusing the discriminator is achieved, and the generated pseudo samples can be gradually approximated to the true samples. In addition, when training the generator, the parameters of the discriminator need to be fixed, namely the weight of the generator is not updated, and the discriminator D is only responsible for transmitting errors, so that the generator is guided to complete parameter updating.
And updating the ACGAN generator parameters according to ACGAN judgment of the true and false conditions, generating new pseudo samples for the previous random noise z according to the current new generation network after updating the generation parameters, and enabling the generated pseudo samples to be more similar to the true samples.
Meanwhile, the same dataset is adopted for multi-classification training of the SVM during training ACGAN, the classification effect of the SVM on the small sample dataset is outstanding, and in the early stage of ACGAN training, the D parameter of the discriminator is required to be optimized, so that the classification effect on faults is poor, at the moment, the classification result of the SVM is combined with the classification result of ACGAN, the classification confidence is enhanced, and the combined classification result is fed back to ACGAN as unified loss for guiding parameter updating. At this time, ACGAN classification loss functions are:
Loss=(1-λ)LSVM+λLC
Wherein λ is a confidence duty ratio of classification result of the discriminator in ACGAN, and any number greater than 0 and less than 1 is set as an initial value of λ. The value of lambda is between 0 and 1, including 0 and 1.
And updating the ACGAN discriminator parameters according to the unified loss function, and updating the SVM model parameters according to the SVM model classification loss function.
The random noise data containing the state type labels of the heating power pipe network is input into a ACGAN generator after updating parameters to obtain more realistic pseudo samples containing the state type labels, the obtained real samples of the heating power pipe network images and the more realistic pseudo samples containing the state type labels newly generated by the ACGAN generator are input into a ACGAN discriminator after updating the parameters to carry out discrimination true and false training and classification training, and the SVM model after inputting the updated parameters is input to carry out classification training.
After the steps are iterated for a plurality of times, the set formed by the pseudo sample generated by the trained ACGAN generator and the real sample of the heating power pipe network image is expanded to be 1 time of the original set, and the training is carried out on the basis of the training of the previous SVM and ACGAN by using a new data set. With the continuous optimization of ACGAN parameters, λ is increased once every certain iteration batch, for example, every certain iteration batch, the value of λ is increased by 0.1, the confidence duty ratio of the classification result of ACGAN is increased, meanwhile, the pseudo sample newly output by the generator is trained, the process is continued until λ is 1, the coefficient of L SVM in the unified loss function is 0, at this time, the parameters of the ACGAN discriminant are updated only according to the classification loss function L C of the ACGAN discriminant, the SVM model exits training, and ACGAN starts to train alone.
The obtained real sample of the heat pipe network image and the pseudo sample with the condition type label newly generated by the ACGAN generator are input into a ACGAN discriminator for classification training, parameters of the ACGAN discriminator are updated according to a classification loss function L C of the ACGAN discriminator, and the obtained real sample of the heat pipe network image and the pseudo sample with the condition type label newly generated by the ACGAN generator are input into a ACGAN discriminator after updating the parameters for classification training. The independent training process is repeated continuously, and the network parameter updating is stopped after a certain number of iterations is reached, at this time, the data generated by the generator G are quite true, and the discriminator cannot discriminate the true or false of the input data, but can accurately divide the condition types.
After training, the discriminator D in ACGAN is stripped, meanwhile, the output of D cancels the judgment of the true and false of the data, only the classified output is reserved, and after the image is acquired in real time, the image is input into the current discriminator D, namely a thermodynamic pipe network condition diagnosis model, so as to carry out condition diagnosis.
According to the method, a large number of simulation samples with the same sample characteristics as the real fault samples are supplemented through the ACGAN generator, the problem that the number of the fault samples is unbalanced with that of normal samples is solved, the training effect of the ACGAN discriminator is improved, the characteristics of the fault samples are extracted more accurately, a condition diagnosis model with more accurate condition diagnosis is obtained, the condition diagnosis accuracy is improved, ACGAN is combined with the SVM, the characteristics that the SVM model has accurate and efficient classification effect when the sample size is smaller are utilized, classification training of ACGAN is assisted through the classification result of the SVM in the early stage of ACGAN training, the defect that the ACGAN discriminator has the classification effect when the early stage parameters are to be optimized is overcome, the classification effect is gradually optimized along with the increase of the sample size, the classification effect of the ACGAN discriminator is poorer when the sample size of the SVM is larger, the classification effect is very low, therefore the ACGAN discriminator starts independent training after the update of the plurality of parameters, the condition diagnosis model obtained through training is more accurate and the diagnosis of the condition diagnosis model is higher in efficiency.
Example two
The second embodiment of the invention provides a method for diagnosing the condition of a heating power pipe network, which has a flow shown in fig. 2 and comprises the following steps:
Step S201, collecting a heat pipe network image;
step S202, inputting the acquired heat pipe network image into a heat pipe network condition diagnosis model obtained by the method, and obtaining a condition diagnosis result output by the heat pipe network condition diagnosis model.
The training and diagnosing process is shown in fig. 3, in which, G represents ACGAN generator to generate pseudo data, namely pseudo sample, D represents ACGAN discriminator, ACGAN discriminator carries out true and false discrimination and classification on the pseudo data and the real data, the true and false discrimination is to feed the discrimination result back to the generator to help the generator to generate more true pseudo data, the SVM model assists ACGAN discriminator to classify in early training stage, and the late ACGAN discriminator carries out classification training independently, and the trained ACGAN discriminator is set in the online fault diagnosing system to diagnose the real-time thermal network condition and output the diagnosis classification result.
In the method of the embodiment, the used thermal pipe network condition diagnosis model is used for obtaining a large number of pseudo samples for training by simulating real samples of the thermal pipe network image, so that the more accurate characteristics of the fault samples are extracted, and the advantages of accurate classification and high efficiency of the SVM model are combined in the early training, so that the thermal pipe network condition diagnosis model is used for diagnosing the thermal pipe network condition more accurately and more efficiently.
Example III
The third embodiment of the invention provides a training device for a diagnosis model of a heating power pipe network condition, the structure of which is shown in fig. 4, comprising:
The auxiliary classification module 101 is configured to input random noise data containing a condition type label of the heat pipe network into the ACGAN generator to obtain a pseudo sample containing the condition type label, input the obtained true sample of the heat pipe network image and the pseudo sample containing the condition type label newly generated by the ACGAN generator into the ACGAN discriminator for discrimination training and classification training, and input the pseudo sample into the SVM model for classification training;
the loss function integration module 102 is configured to integrate the classification loss functions of the ACGAN discriminators and the SVM model into a unified loss function;
The parameter updating module 103 is used for updating the parameters of the ACGAN generator according to ACGAN to judge whether the model is true or false, updating the parameters of the ACGAN discriminator according to the unified loss function, and updating the parameters of the SVM model according to the SVM model classification loss function;
And the independent classification module 104 is used for independently performing classification training by ACGAN after the unified loss function is not influenced by the SVM loss function, and obtaining a heating power pipe network condition diagnosis model according to the trained ACGAN discriminator.
Optionally, the method further comprises:
the classification preparation module 100 is used for obtaining a real sample of the heat pipe network image, setting a condition type label according to the condition type of the real sample of the heat pipe network image, generating random noise data containing the condition type label, constructing an SVM model according to the condition type, and constructing a ACGAN model.
Optionally, the loss function integration module includes:
a discriminator loss function determining unit, configured to determine a classification loss function L C of the ACGAN discriminators;
an SVM penalty function determination unit configured to determine a classification penalty function L SVM of the SVM model;
the Loss function integrating unit is configured to determine, according to the classification Loss functions L C and L SVM, that the expression of the unified Loss function Loss is:
Loss=(1-λ)LSVM+λLC
Wherein λ is a confidence ratio of the classification result of the ACGAN discriminator, and an arbitrary number greater than 0 and less than 1 is set as an initial value of λ.
Optionally, the classification preparation module includes:
The real sample obtaining unit is used for respectively collecting a plurality of heat pipe network image samples under different condition categories, wherein the condition categories comprise normal heat pipe network image samples, pipe wall leakage, pipe bending deformation and pipe heat preservation damage, carrying out at least one item of data enhancement processing on the collected heat pipe network image samples under different condition categories to obtain heat pipe network image samples after the data enhancement processing, wherein the data enhancement processing comprises image rotation, cutting, similar splicing, affine transformation, increasing check noise, increasing illumination and shadow, adjusting contrast and adjusting saturation;
The condition type label setting unit is used for setting a condition type label according to the condition type of the real sample of the heat pipe network image;
An SVM model construction unit for constructing N-1 SVM sub-models according to the condition category number N, wherein N is a positive integer;
ACGAN a model building unit for building ACGAN generator using deconvolution neural network, building ACGAN discriminator using convolution neural network;
A noise data obtaining unit, configured to collect a plurality of random noise data, where each random noise data is added with one status type tag, and each status type tag is added to the plurality of random noise data.
In the device, a large number of simulation samples with the same sample characteristics as the real fault samples are supplemented through a ACGAN generator, the problem that the number of the fault samples is unbalanced with that of normal samples is solved, the training effect of a ACGAN discriminator is improved, the characteristics of the fault samples are extracted more accurately, a condition diagnosis model with more accurate condition diagnosis is obtained, the condition diagnosis accuracy is improved, ACGAN is combined with SVM, the characteristics of the SVM model that the sample size is smaller and the classification effect is accurate and efficient are utilized, classification training of ACGAN is assisted through the classification result of the SVM in the early stage of ACGAN training, the defect that the ACGAN discriminator has the classification effect to be optimized in the early stage of parameter is overcome, the classification effect is gradually optimized along with the increase of the sample size, the classification effect of the ACGAN discriminator is poorer and the efficiency is lower when the sample size of the SVM is larger, the ACGAN discriminator starts independent training after the update of the plurality of parameters, and the condition diagnosis model obtained by training is more accurate and the diagnosis of the condition diagnosis condition of the thermodynamic pipe network is higher in efficiency.
Example IV
A fourth embodiment of the present invention provides a heat pipe network condition diagnosis device, whose structure is shown in fig. 5, including:
the image acquisition module 201 is used for acquiring a heat pipe network image;
the condition diagnosis module 202 is configured to input the collected heat pipe network image into the heat pipe network condition diagnosis model obtained by using the foregoing method, and obtain a condition diagnosis result output by the heat pipe network condition diagnosis model.
In the device of the embodiment, the used heat pipe network condition diagnosis model is used for obtaining a large number of pseudo samples for training by simulating the real samples of the heat pipe network image, so that the more accurate characteristics of the fault samples are extracted, and the advantages of accurate classification and high efficiency of the SVM model are combined in the early training, so that the diagnosis of the heat pipe network condition by using the heat pipe network condition diagnosis model is more accurate and has higher efficiency.
Based on the same inventive concept, the embodiment of the invention also provides a terminal device, the structure of which is shown in fig. 5, comprising a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the aforementioned heating power pipe network condition diagnosis model training method when executing the computer program.
Based on the same inventive concept, the embodiment of the invention also provides a computer storage medium, wherein the computer storage medium stores computer executable instructions, and the computer executable instructions realize the heating power pipe network condition diagnosis model training method when executed.
Based on the same inventive concept, the embodiment of the invention also provides a terminal device, the structure of which is shown in fig. 5, comprising a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the aforementioned heating power pipe network condition diagnosis method when executing the computer program.
Based on the same inventive concept, the embodiment of the invention also provides a computer storage medium, wherein the computer storage medium stores computer executable instructions, and the computer executable instructions realize the heating power pipe network condition diagnosis method when executed.
The specific manner in which the various modules perform the operations in the apparatus of the above embodiments have been described in detail in connection with the embodiments of the methods, and will not be described in detail herein.
Claims (18)
1. The heating power pipe network condition diagnosis model training method is characterized by comprising the following steps of:
Inputting random noise data containing a heating pipe network condition type label into a ACGAN generator to obtain a pseudo sample containing the condition type label, inputting the obtained heating pipe network image true sample and the pseudo sample newly generated by the ACGAN generator and containing the condition type label into a ACGAN discriminator for discrimination and false training and classification training, inputting an SVM model for classification training, integrating a ACGAN discriminator and a classification loss function of the SVM model into a unified loss function, updating parameters of the ACGAN generator according to ACGAN discrimination of true and false, updating parameters of the ACGAN discriminator according to the unified loss function, and updating parameters of the SVM model according to the SVM model classification loss function;
after the steps are iterated for a plurality of times, reducing the influence of an SVM loss function on the unified loss function once every iteration for a certain number of times until the unified loss function is not influenced by the SVM loss function;
After the unified loss function is not influenced by the SVM loss function, carrying out classification training independently by using ACGAN, and obtaining a heating power pipe network condition diagnosis model according to a trained ACGAN discriminator;
the integration of the ACGAN discriminant and the classification loss function of the SVM model into a unified loss function comprises the following steps:
determining a classification loss function L C of the ACGAN arbiter;
Determining a classification loss function L SVM of the SVM model;
According to the classification Loss functions L C and L SVM, the expression for determining the unified Loss function Loss is:
Loss=(1-λ)LSVM+λLC
Wherein λ is a confidence ratio of the classification result of the ACGAN discriminator, and an arbitrary number greater than 0 and less than 1 is set as an initial value of λ.
2. The method of claim 1, wherein before inputting ACGAN the random noise data containing the heat pipe network condition category label into the generator, comprising the steps of:
Obtaining a real sample of a heat pipe network image;
setting a condition type label according to the condition type of the real sample of the heat pipe network image;
generating random noise data containing the condition category label;
Constructing an SVM model according to the condition category;
And constructing ACGAN a model.
3. The method of claim 2, wherein obtaining a real sample of the heat pipe network image comprises the steps of:
respectively collecting a plurality of heat distribution pipe network image samples under different condition categories, wherein the condition categories comprise normal pipe wall leakage, pipe bending deformation and pipe heat preservation layer damage;
At least one item of data enhancement processing is carried out on the collected heat pipe network image samples under different condition categories, so that the heat pipe network image samples after the data enhancement processing are obtained, wherein the data enhancement processing comprises image rotation, cutting, similar splicing, affine transformation, verification noise increase, illumination and shadow increase, contrast adjustment and saturation adjustment;
Carrying out random combination on the heat pipe network image samples subjected to data enhancement treatment to obtain heat pipe network image samples subjected to random combination;
And marking the collected heat pipe network image samples under different conditions, the heat pipe network image samples subjected to data enhancement processing and the heat pipe network image samples subjected to random combination as heat pipe network image true samples.
4. The method of claim 2, wherein said constructing an SVM model from said condition categories comprises the steps of:
and constructing N-1 SVM sub-models according to the condition category number N, wherein N is a positive integer.
5. The method of claim 2, wherein the constructing ACGAN model comprises the steps of:
creating ACGAN a generator using a deconvolution neural network;
A ACGAN arbiter is built using a convolutional neural network.
6. The method of claim 2, wherein said generating random noise data comprising said condition category label comprises the steps of:
A plurality of random noise data are collected, each random noise data is added with one condition category label, and each condition category label is added to the plurality of random noise data.
7. The method of claim 1, wherein said reducing the effect of an SVM loss function on said unified loss function for each iteration a number of times until said unified loss function is unaffected by said SVM loss function comprises the steps of:
Every iteration is performed for a certain number of times, lambda in the unified Loss function Loss expression is increased by a certain increment until lambda is increased to 1.
8. The method of claim 1, wherein after the unified loss function is not affected by the SVM loss function, using ACGAN to perform independent classification training, and obtaining a heat pipe network condition diagnosis model according to the trained ACGAN discriminator, including the following steps:
After lambda is increased to 1, inputting the obtained real sample of the heat pipe network image and the pseudo sample which is newly generated by the ACGAN generator and contains the condition type label into a ACGAN discriminator for classification training, and updating ACGAN discriminator parameters according to a unified loss function;
and after the training termination condition is met, stopping updating ACGAN the parameters of the discriminator, and obtaining a heating power pipe network condition diagnosis model according to the trained ACGAN discriminator.
9. The method of claim 8, wherein the obtaining a heating network condition diagnostic model based on the trained ACGAN discriminators comprises the steps of:
And stripping the trained ACGAN discriminator from the ACGAN model, canceling the output of the trained ACGAN discriminator on discrimination of true and false, and only reserving classification output to obtain the heating power pipe network condition diagnosis model.
10. The method for diagnosing the condition of the heating power pipe network is characterized by comprising the following steps of:
Inputting the acquired heat pipe network image into a heat pipe network condition diagnosis model obtained by the method of any one of claims 1-9, and obtaining a condition diagnosis result output by the heat pipe network condition diagnosis model.
11. A heating power pipe network condition diagnosis model training device, characterized by comprising:
the auxiliary classification module is used for inputting random noise data containing the condition type labels of the heating power pipe network into the ACGAN generator to obtain pseudo samples containing the condition type labels, inputting the obtained true samples of the heating power pipe network images and the pseudo samples newly generated by the ACGAN generator and containing the condition type labels into the ACGAN discriminator for discrimination training and classification training, and inputting the SVM model for classification training;
the loss function integration module is used for integrating the classification loss functions of the ACGAN discriminator and the SVM model into a unified loss function;
The parameter updating module is used for updating parameters of the ACGAN generator according to ACGAN to judge whether the model is true or false, updating parameters of the ACGAN discriminator according to the unified loss function, and updating parameters of the SVM model according to the SVM model classification loss function;
The independent classification module is used for independently performing classification training by ACGAN after the unified loss function is not influenced by the SVM loss function, and obtaining a heating power pipe network condition diagnosis model according to a trained ACGAN discriminator;
the loss function integration module comprises:
a discriminator loss function determining unit, configured to determine a classification loss function L C of the ACGAN discriminators;
an SVM penalty function determination unit configured to determine a classification penalty function L SVM of the SVM model;
the Loss function integrating unit is configured to determine, according to the classification Loss functions L C and L SVM, that the expression of the unified Loss function Loss is:
Loss=(1-λ)LSVM+λLC
Wherein λ is a confidence ratio of the classification result of the ACGAN discriminator, and an arbitrary number greater than 0 and less than 1 is set as an initial value of λ.
12. The apparatus as recited in claim 11, further comprising:
The classification preparation module is used for obtaining a real sample of the heat pipe network image, setting a condition type label according to the condition type of the real sample of the heat pipe network image, generating random noise data containing the condition type label, constructing an SVM model according to the condition type, and constructing a ACGAN model.
13. The apparatus of claim 12, wherein the classification preparation module comprises:
The real sample obtaining unit is used for respectively collecting a plurality of heat pipe network image samples under different condition categories, wherein the condition categories comprise normal heat pipe network image samples, pipe wall leakage, pipe bending deformation and pipe heat preservation damage, carrying out at least one item of data enhancement processing on the collected heat pipe network image samples under different condition categories to obtain heat pipe network image samples after the data enhancement processing, wherein the data enhancement processing comprises image rotation, cutting, similar splicing, affine transformation, increasing check noise, increasing illumination and shadow, adjusting contrast and adjusting saturation;
The condition type label setting unit is used for setting a condition type label according to the condition type of the real sample of the heat pipe network image;
An SVM model construction unit for constructing N-1 SVM sub-models according to the condition category number N, wherein N is a positive integer;
ACGAN a model building unit for building ACGAN generator using deconvolution neural network, building ACGAN discriminator using convolution neural network;
A noise data obtaining unit, configured to collect a plurality of random noise data, where each random noise data is added with one status type tag, and each status type tag is added to the plurality of random noise data.
14. A heating power pipe network condition diagnosing apparatus, comprising:
the image acquisition module is used for acquiring a heating power pipe network image;
The condition diagnosis module is used for inputting the acquired heat pipe network image into the heat pipe network condition diagnosis model obtained by the method of any one of claims 1-9, and obtaining the condition diagnosis result output by the heat pipe network condition diagnosis model.
15. The terminal equipment is characterized by comprising a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the heating power pipe network condition diagnosis model training method according to any one of claims 1-9 when executing the computer program.
16. A computer storage medium, wherein computer executable instructions are stored in the computer storage medium, and when the computer executable instructions are executed, the method for training the thermal pipe network condition diagnosis model according to any one of claims 1-9 is realized.
17. A terminal device comprising a memory, a processor and a computer program stored on the memory and operable on the processor, the processor implementing the method for diagnosing a condition of a heat pipe network according to claim 10 when executing the computer program.
18. A computer storage medium having stored therein computer executable instructions that when executed implement the heat pipe network condition diagnostic method of claim 10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210280833.2A CN114818859B (en) | 2022-03-21 | 2022-03-21 | Heating power pipe network condition diagnosis method, device, terminal equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210280833.2A CN114818859B (en) | 2022-03-21 | 2022-03-21 | Heating power pipe network condition diagnosis method, device, terminal equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114818859A CN114818859A (en) | 2022-07-29 |
| CN114818859B true CN114818859B (en) | 2025-07-18 |
Family
ID=82530830
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210280833.2A Active CN114818859B (en) | 2022-03-21 | 2022-03-21 | Heating power pipe network condition diagnosis method, device, terminal equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114818859B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115936450B (en) * | 2023-03-15 | 2023-06-16 | 南京华助智能科技有限公司 | Loss diagnosis method and system for pipeline of steam heating system |
| CN116895016B (en) * | 2023-06-27 | 2024-12-06 | 中国人民解放军海军航空大学 | A method for generating and classifying ship targets in SAR images |
| CN119502003B (en) * | 2025-01-15 | 2025-04-15 | 浙江英集动力科技有限公司 | A steam heating network trap adjustment method based on an embodied intelligent robot dog |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110717579B (en) * | 2019-11-13 | 2023-05-19 | 上海海事大学 | Gear box data model training and using method |
| CN113869208B (en) * | 2021-09-28 | 2024-06-07 | 徐州卓越声振测控科技有限公司 | Rolling bearing fault diagnosis method based on SA-ACWGAN-GP |
-
2022
- 2022-03-21 CN CN202210280833.2A patent/CN114818859B/en active Active
Non-Patent Citations (1)
| Title |
|---|
| Real-Time High-Performance Laser Welding Defect Detection by Combining ACGAN-Based Data Enhancement and Multi-Model Fusion;kui Fan;《sensors》;20211102;第1-13页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114818859A (en) | 2022-07-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Shao et al. | Generative adversarial networks for data augmentation in machine fault diagnosis | |
| Zhang et al. | Machinery fault diagnosis with imbalanced data using deep generative adversarial networks | |
| CN114818859B (en) | Heating power pipe network condition diagnosis method, device, terminal equipment and storage medium | |
| CN115577248B (en) | Fault diagnosis system and method for wind generating set | |
| WO2025152448A1 (en) | Adaptive deep transfer fault diagnosis method and system, apparatus and medium | |
| CN113541985B (en) | Internet of things fault diagnosis method, model training method and related devices | |
| CN109813542A (en) | Fault Diagnosis Method of Air Handling Unit Based on Generative Adversarial Network | |
| CN111612051A (en) | A Weakly Supervised Object Detection Method Based on Graph Convolutional Neural Networks | |
| CN110045335A (en) | Based on the Radar Target Track recognition methods and device for generating confrontation network | |
| CN114067155A (en) | Image classification method, device, product and storage medium based on meta learning | |
| CN111192158A (en) | Transformer substation daily load curve similarity matching method based on deep learning | |
| CN115456082A (en) | A transfer learning framework for wind turbine fault detection | |
| You et al. | Ensemble learning based multi-fault diagnosis of air conditioning system | |
| CN116541507A (en) | Visual question-answering method and system based on dynamic semantic graph neural network | |
| CN114492149A (en) | Power distribution network production operation scene credible simulation method based on countermeasure game | |
| CN109101984B (en) | Image identification method and device based on convolutional neural network | |
| CN117368705A (en) | Digital integrated circuit timing monitoring method based on graph convolutional neural network | |
| CN113627538B (en) | Method for training asymmetric generation of image generated by countermeasure network and electronic device | |
| CN119128453B (en) | Federal domain generalized fault diagnosis method and system | |
| CN119166503A (en) | A method and system for generating software test data | |
| CN109583406B (en) | Facial Expression Recognition Method Based on Feature Attention Mechanism | |
| CN115952493B (en) | A black box model reverse attack method, attack device and storage medium | |
| Hu et al. | An improved metalearning framework to optimize bearing fault diagnosis under data imbalance | |
| CN115049852B (en) | Bearing fault diagnosis method and device, storage medium and electronic equipment | |
| CN118552718A (en) | A few-shot target detection method based on feature distribution correction |
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 |