CN116842459A - Electric energy metering fault diagnosis method and diagnosis terminal based on small sample learning - Google Patents

Electric energy metering fault diagnosis method and diagnosis terminal based on small sample learning Download PDF

Info

Publication number
CN116842459A
CN116842459A CN202311119877.8A CN202311119877A CN116842459A CN 116842459 A CN116842459 A CN 116842459A CN 202311119877 A CN202311119877 A CN 202311119877A CN 116842459 A CN116842459 A CN 116842459A
Authority
CN
China
Prior art keywords
data
electrical parameters
sample
electric energy
energy metering
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202311119877.8A
Other languages
Chinese (zh)
Other versions
CN116842459B (en
Inventor
王永贵
李温静
刘柱
刘迪
黄吕超
邓思阳
李云鹏
张帅
杜月
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.)
State Grid Information and Telecommunication Co Ltd
Original Assignee
State Grid Information and Telecommunication 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 State Grid Information and Telecommunication Co Ltd filed Critical State Grid Information and Telecommunication Co Ltd
Priority to CN202311119877.8A priority Critical patent/CN116842459B/en
Publication of CN116842459A publication Critical patent/CN116842459A/en
Application granted granted Critical
Publication of CN116842459B publication Critical patent/CN116842459B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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; CALCULATING OR 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
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computational Linguistics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Evolutionary Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The application provides an electric energy metering fault diagnosis method and a diagnosis terminal based on small sample learning, which relate to the technical field of electric energy metering and are used for extracting and classifying the characteristics of electric parameters in a data set; selecting a normal data sample set as a training set sample, and calculating the pearson coefficients among the electrical parameters by using correlation analysis; establishing a pearson correlation coefficient matrix among the electrical parameters, and calculating the weight of each electrical parameter by using an entropy weight method; defining electrical parameters, constructing a model by adopting a twin network based on the repaired electrical data, and determining a final classification result by calculating the distance between a sample to be detected and a known label sample and finding the nearest neighbor type. The twin network adopts LSTM and CNN networks as feature extraction networks, two sub-networks with shared weights are used for receiving two input samples simultaneously, the output result is the similarity of the two samples, and meanwhile, improvement loss functions are considered to improve classification accuracy.

Description

Electric energy metering fault diagnosis method and diagnosis terminal based on small sample learning
Technical Field
The application relates to the technical field of electric energy metering, in particular to an electric energy metering fault diagnosis method and a diagnosis terminal based on small sample learning.
Background
The electric energy metering is one of core services of the power grid, and accurate and reliable electric energy metering data is an important basis for ensuring safe and economic operation of the power grid. The electric energy metering and collecting equipment is widely distributed, and due to the severe deployment environment and complex working condition of part of equipment, data loss or abnormality is easy to cause in the process of collection and transmission, the analysis and judgment of power grid business are seriously affected, and particularly, the analysis of the data loss or abnormal data aiming at the electric energy metering business can generate metering results with great deviation from actual conditions.
The loss of the related data of the electric energy measurement can be generally divided into three aspects, (1) the main body of the electric energy meter fails, and the service life of sensing equipment or an internal key chip can be reduced due to the complex equipment deployment environment when the equipment is used for a long time, so that the main body of the equipment cannot work normally; (2) When the transmission network fails, the electric energy meter and the edge equipment are connected in a wired or wireless way, and once the transmission network fails, the situations of packet loss, network congestion and the like are caused; (3) The environmental factors cause faults, and due to the interference of temperature, humidity, strong electromagnetic fields and the like, abnormal data of the electric energy meter can be caused, and the power distribution network can be damaged more seriously.
The data collected by all edge internet of things equipment in the power grid has strong correlation, the collected data has time sequence characteristics, and the data loss influences the fault diagnosis and positioning of the subsequent power grid service. Therefore, the multi-source data of different sensing devices converged to the edge equipment are needed to be utilized, and the data is preprocessed by creating a missing data compensation model, so that the problem of data missing is solved.
In the prior art, besides data loss, the abnormality of the electric energy metering data can be caused by the electric energy metering fault, and the rapid diagnosis of the electric energy metering fault is an effective means for safe operation of the power grid. Aiming at common power grid faults, the number of fault samples is required to be collected according to the power grid fault types, and a fault judgment model is established through processing collected data to realize accurate fault diagnosis. For the electric energy metering faults, because the data are difficult to collect effectively, the number of samples is small, the normal data and the abnormal data cannot be subjected to effective classification analysis, fault diagnosis is inaccurate, and the fault state and the fault type are difficult to diagnose effectively.
Disclosure of Invention
The application provides a small sample learning-based electric energy metering fault diagnosis method, which utilizes K nearest neighbor (K Nearest Neighbors, KNN) clustering algorithm technology, LSTM (local area network) and CNN (computer network) network technology and other models, and can be combined with sensor monitoring, data transmission and other technologies to repair and complement electric energy metering data loss or abnormality. And an electric energy metering fault diagnosis model based on small sample learning is constructed, and an L2 weight attenuation coefficient with a self-adaptive coefficient is introduced through a characteristic extraction network of a long-short-term memory artificial neural network, a convolution neural network and a pooling layer combination, so that high-accuracy electric energy metering fault diagnosis based on multidimensional data fusion is realized.
For the present application, small sample learning: also known as low-sample learning (LSL), is a machine learning method that trains datasets containing limited information. It is common practice in the field of machine learning applications to provide models that can receive as much data as possible. This is because in most machine learning applications, providing more data enables the model to predict better. However, small sample learning aims to build accurate machine learning models with less training data.
Clustering: the process of dividing a collection of physical or abstract objects into classes composed of similar objects is called clustering. Clusters generated by a cluster are a collection of data objects that are similar to objects in the same cluster, and are different from objects in other clusters. "the subjects are grouped together", and there are a number of classification problems in natural science and social science. Cluster analysis, also known as cluster analysis, is a statistical analysis method for studying (sample or index) classification problems. Cluster analysis originates in taxonomies, but clusters are not equal to classifications. Clustering differs from classification in that the class into which clustering requires partitioning is unknown. The clustering analysis content is very rich, and the clustering analysis method comprises a systematic clustering method, an ordered sample clustering method, a dynamic clustering method, a fuzzy clustering method, a graph theory clustering method, a clustering forecasting method and the like.
And (3) diagnosing electric energy metering faults: through analyzing the collected electric energy metering data, possible faults of the electric energy metering device are judged, wherein the possible faults comprise faults of an electric energy meter body, network transmission faults, environmental factors and the like. Common abnormal faults of electric energy metering include stop and go of an electric energy meter, abnormal voltage and current, abnormal power factor and the like.
In particular, the method defines the collected data set asI.e. the data set contains m electrical parameters; the fault diagnosis method comprises the following steps:
s1: for data set X i Extracting characteristics of the electrical parameters in the data processing system, classifying the abnormal data and the normal data to form an abnormal data sample set and a normal data sample set;
s2: selecting positiveThe constant data sample set is used as a training set sample, and the pearson coefficients among all the electrical parameters are calculated by utilizing relevance analysis; based on abnormal electrical parameters, screening out electrical parameters which are strongly related to the abnormal electrical parameters, establishing a pearson correlation coefficient matrix sigma among the electrical parameters, and calculating the weight of each electrical parameter by using an entropy weight method；
S3: definition of electrical parametersIs expressed as +.>For electrical parameters->Is a strongly associated attribute +.>In data set->Find and +.>Data of the same time sequence point, denoted +.>Configuration ofThe cluster center of the class is +.>；
S4: based on electrical parametersSelecting all m electrical parameters, and carrying out weighted summation on the m electrical parameters to obtain abnormal data +.>Is the optimum correction result of (a): />And further, repairing and filling of abnormal data are realized.
S5: based on the repaired electrical data, a twin network construction model is adopted, and the nearest neighbor class is found to determine the final classification result by calculating the distance between the sample to be detected and the known label sample. The twin network adopts LSTM and CNN networks as feature extraction networks, two sub-networks with shared weights are used for receiving two input samples simultaneously, the output result is the similarity of the two samples, and meanwhile, improvement loss functions are considered to improve classification accuracy.
It should be further noted that, in step S1, a data deletion repair model is further constructed by combining clustering and correlation analysis, and all input electrical parameters are clustered into different sets by using a KNN clustering algorithm, so that data deletion repair is realized by data correlation analysis.
It should be further noted that, in step S2, the pearson correlation coefficient is used to evaluate the correlation between two data, and when the pearson correlation coefficient is greater than a preset characteristic value, it is determined that the two data have a strong correlation.
It should be further noted that the pearson coefficients between the respective electrical parameters are expressed as:
(1-1)
wherein ,two n-dimensional data variables, +.>Respectively->Standard deviation of>Is the covariance between the two variables;
the covariance calculation formula is as follows:
(1-2)
wherein 、/>For n-dimensional variables-> and />Is a data average of (a).
In step S2, the entropy weight method determines the weight by indicating the size of the content of information, calculates the entropy between the variable j and other variablesExpressed as:
(1-3)
wherein S is the number of variables.
It should be further noted that the weights of the electrical parameters in step S2Expressed as:
(1-4)
wherein the weight isIndicating the size of the index information amount, +.>。
In the method, the KNN clustering algorithm performs weighting processing based on a gaussian function, and the gaussian function is as follows:
(1-15)
wherein a is the height of the peak of the Gaussian curve, and a >0; b is the x coordinate value of the peak center of the Gaussian curve; c is the standard deviation, and the width of the Gaussian curve is controlled;
weight function of KNN algorithmExpressed as:
(1-16)
k is the selected number of neighbors for the distance between the electrical parameter to be measured and the first neighbor point.
It should be further noted that, the method also performs linear proportion normalization processing on the collected electrical parameters, foriTime of day original electrical parametersNormalized data ∈>Can be expressed as:
(1-13)
in the formula ,is the maximum value of the corresponding data channel;
the method also defines a current inversion flagAnd average current +.>；
The concrete steps are as follows:
(1-11)
(1-12)。
it should be further noted that, the method is based on metric learning in small sample learning, is simple in calculation, convenient to operate, relatively suitable for classifying and judging fewer samples, and suitable for edge computing equipment. The twin network uses a sub-network shared by two weights to simultaneously receive two input samples, and the output result is the similarity of the two samples. The twin network adopts LSTM and CNN networks as feature extraction networks, and meanwhile, improvement loss functions are considered to improve classification accuracy.
The application also provides a diagnosis terminal, which comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, wherein the processor realizes the steps of the electric energy metering fault diagnosis method based on small sample learning when executing the program.
From the above technical scheme, the application has the following advantages:
the electric energy metering fault diagnosis method based on small sample learning solves the problem of low accuracy of electric energy metering business caused by factors such as electric energy metering data deficiency, abnormality and the like, and realizes high-accuracy electric energy metering fault diagnosis based on multidimensional data fusion.
Aiming at the problems of low accuracy of the electric energy metering service and the like caused by factors such as electric energy metering data loss, abnormality and the like, the application constructs an electric energy metering data loss repair model based on the combination of clustering and correlation analysis by utilizing the strong correlation of the electric energy metering related data, and supplements data loss; an electric energy metering fault diagnosis model based on small sample learning is constructed, and an L2 weight attenuation coefficient with a self-adaptive coefficient is introduced through a feature extraction network of a long-short-term memory artificial neural network, a convolution neural network and a pooling layer combination, so that high-accuracy electric energy metering fault diagnosis based on multidimensional data fusion is realized.
Drawings
In order to more clearly illustrate the technical solutions of the present application, the drawings that are needed in the description will be briefly introduced below, it being obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of an embodiment of a method for diagnosing a power metering fault;
FIG. 2 is a schematic diagram of a data loss repair model based on clustering and correlation analysis;
FIG. 3 is a model of fault diagnosis of electric energy metering based on a twin network;
FIG. 4 is a schematic view of LSTM structure;
FIG. 5 is a diagram of a twinning network;
FIG. 6 is a flow chart of an implementation of a twin network based power metering diagnostic model.
Detailed Description
The electric energy metering fault diagnosis method based on small sample learning provided by the application is based on an artificial intelligence algorithm, and the judgment of the fault type of the electric energy meter is realized through the acquisition and fusion of the multidimensional data of the electric energy meter. The electric energy metering fault diagnosis method based on small sample learning is based on the artificial intelligence technology to acquire and process the associated data.
Of course, the electric energy metering fault diagnosis method based on the small sample learning has the technology of both hardware level and software level. The basic technology of the electric energy metering fault diagnosis method generally comprises technologies such as a sensor, a special artificial intelligent chip, cloud computing, distributed storage, big data processing technology, an operation/interaction system, electromechanical integration and the like. The software technology of the electric energy metering fault diagnosis method mainly comprises a computer visual angle technology, a programming language, machine learning/deep learning and the like. Programming languages include, but are not limited to, object-oriented programming languages such as Java, smalltalk, C ++, and conventional procedural programming languages such as the "C" language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or power server. In the case of a remote computer, the remote computer may be connected to the user's computer through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computer (for example, through the Internet using an Internet service provider).
A flowchart of a preferred embodiment of the small sample learning-based power metering fault diagnosis method of the present application is shown in fig. 1. The electric energy metering fault diagnosis method based on small sample learning is applied to one or more diagnosis terminals, wherein the diagnosis terminals are devices capable of automatically performing numerical calculation and/or information processing according to preset or stored instructions, and the hardware comprises, but is not limited to, microprocessors, application-specific integrated circuits (SpecificIntegratedCircuit, ASIC), programmable gate arrays (Field-ProgrammableGate Array, FPGA), digital processors (DigitalSignalProcessor, DSP), embedded devices and the like.
The diagnostic terminal may be any electronic product that can interact with a user, such as a personal computer, tablet, smart phone, personal digital assistant (PersonalDigitalAssistant, PDA), interactive web TV (InternetProtocolTelevision, IPTV), etc.
The diagnostic terminal may also include a network device and/or a user device. Wherein the network device includes, but is not limited to, a single network server, a server group made up of multiple network servers, or a cloud based on cloud computing (CloudComputing) made up of a large number of hosts or network servers.
The network in which the diagnostic terminal is located includes, but is not limited to, the internet, a wide area network, a metropolitan area network, a local area network, a virtual private network (VirtualPrivateNetwork, VPN), and the like.
The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
Referring to fig. 1 to 6, a flowchart of a method for diagnosing a fault of electric energy measurement based on small sample learning in an embodiment is shown, where the embodiment is specific to a plurality of abnormal data classes of electric energy measurementEach dataclass can only provide a small number of annotation samples. For data set->，/>For the input multidimensional data vector, +.>For inputting multidimensional vector +.>The electric energy metering abnormality diagnosis research based on small sample learning is to model and optimize the electric energy metering abnormality diagnosis business by using a limited data set, so that the fault diagnosis accuracy of the model is improved. The small sample problem can be expressed as N-way K-shot, N represents category, K represents the corresponding sample number of each category, and model training is mainly carried out when the training set has only N x K samples。
The data set for a small number of markers can be represented asK is the class of the marked sample, +.>For an input dataset of samples marked K, and (2)>Is->And (3) inputting the type of the abnormal electric energy metering faults corresponding to the data, constructing an abnormal electric energy metering fault diagnosis model by using a small amount of sample data, and judging the type corresponding to the sample to be tested by importing test data into the model.
Aiming at the problems of few marked electric energy metering abnormality diagnosis samples, uneven sample distribution, data loss and the like, the embodiment aims at electric energy metering fault diagnosis, and realizes electric energy metering data quality improvement and fault diagnosis by creating a data loss repair model and an electric energy metering fault diagnosis model based on small sample learning and utilizing multidimensional data fusion among different edge internet-of-things agents such as intelligent fusion terminals, concentrators, energy routers and the like.
In an exemplary embodiment, for the problems of missing of electric energy metering data, fault diagnosis and the like, a data missing repair model utilizing multidimensional data fusion and an electric energy metering fault diagnosis model based on small sample learning can be constructed, and abnormal complementation and fault discrimination of power grid data are realized by converging different edge equipment side data.
It can be seen that the correlation between the related data of the electric energy metering of the power grid is strong, the correlation between the data with different dimensions is not negligible, the data deletion repair model is constructed by combining the clustering and the correlation analysis based on the characteristics of the correlation of the multidimensional data, all input data are gathered into different sets by using a K nearest neighbor (K Nearest Neighbors, KNN) clustering algorithm, and the data deletion repair is realized by the data correlation analysis, as shown in fig. 2.
According to the embodiment of the application, the main idea of the KNN algorithm is to classify by measuring the distances between different characteristic values, and when deciding the class of the sample, only the class of k 'neighbor' samples around the sample is referred, so that the method is suitable for processing the scene with more overlapping sample sets.
It should be noted that, the relevance analysis algorithm adopts the pearson relevance coefficient method to measure the relevance between the data of different edge equipment sides, and then calculates the weight relation between different clusters through the entropy weight method.
Pearson correlation coefficient of two variablesThis can be expressed as:
(1-1)
wherein ,two n-dimensional data variables, +.>Respectively->Standard deviation of>Is the covariance between the two variables. The covariance calculation formula is as follows:
(1-2)
wherein 、/>For n-dimensional variables-> and />Is a data average of (a). The pearson correlation coefficient is used to evaluate the correlation between two data, and when the pearson correlation coefficient is greater than a certain characteristic value, the two data are considered to have a strong correlation.
The entropy weight method of the embodiment determines the weight through the size of the index information content, can effectively utilize data and eliminates the influence of subjective factors. Calculating entropy between variable j and each variableThis can be expressed as:
(1-3)
wherein S is the number of variables.
Weights of the respective variablesThis can be expressed as:
(1-4)
wherein the weight isIndicating the size of the index information amount, +.>。
In the embodiment of the application, the measurement learning in the small sample learning is simple in calculation, convenient to operate, relatively suitable for classifying and judging less samples, and suitable for edge computing equipment. The twin network uses a sub-network shared by two weights to simultaneously receive two input samples, and the output result is the similarity of the two samples. To adapt to the characteristics of data such as current, voltage, power and the like of a time sequence, the twin network adopts LSTM and CNN networks as characteristic extraction networks, and meanwhile, improvement loss functions are considered to improve classification accuracy, as shown in figure 3.
The feature extraction network module adopts a network structure of LSTM, CNN and Pooling Layer (PL), because LSTM is often used for processing time series data, and is used for extracting time features of data samples, CNN network is used for processing space information, thereby realizing feature extraction of information with different dimensions.
According to the embodiment of the application, LSTM is a special cyclic neural network (Recurrent Neural Network, RNN), and a selective mechanism of gate control is introduced on the basis of the RNN, namely a forgetting gate, an input gate and an output gate, as shown in figure 4, so that information is selectively reserved or deleted, and long-term dependency can be well learned.
(1-5)
wherein ,output information for forgetting gate,/->、/>Weight and bias parameters for forgetting gate, +.>For inputting the output information of the gate, +.>、/>Weight and bias parameters for input gates, +.>、/>For the weight and bias function of the tanh layer, +.>For the cellular state->Is an activation function, the output range is 0-1, < >>For outputting the output information of the gate, +.>、/>For outputting the weight and bias parameters of the gate, +.>Representing the output of the previous cell, +.>For the output of the unit, +.>Representing the current input.
Model two samples [ ],/>) Obtaining a feature vector through a feature extraction network> and />The similarity measure can be expressed as:
(1-6)
the similarity measurement method is commonly used for cosine distance functions, euclidean distances and the like.
N-way K-shot is a small sample classification problem, if a single category is judged, a sigmoid activation function is used for judging the similarity of two categories, a result of 1 indicates that the category of the sample to be detected is consistent with the category of the comparison sample, and a result of 0 indicates that the category of the sample to be detected is inconsistent with the category of the comparison sample. For the N-class multi-classification problem, the vector is typically normalized using a Softmax activation function in the multi-classification problem, resulting in probability distributions for the multiple classes. After obtaining the similarity between the query sample x and the N categories, inputting the similarity into a Softmax function to obtain probabilities of belonging to the N categories, wherein the probabilities are specifically expressed as follows:
(1-7)
when two samples are input to belong to the same category, the similarity approaches 1, and when two sample data and different categories, the phase velocity approaches 0. The twin network adopts a comparison loss function to optimize a network model, a calculation loss function adopts a cross entropy loss function, and a self-adaptive coefficient is introduced simultaneously through an activation functionL2 weight decay coefficient term of->Preventing the model from being over-fitted during retraining to obtain the loss function +.>This can be expressed as:
(1-8)
(1-9)
before the diagnosis and analysis of the collected electrical parameters are carried out, data processing is carried out, specifically, terminal data are collected through different edge devices, a plurality of electrical parameters, environmental parameters and the like are collected, the three-phase electrical parameters comprise three-phase voltage, three-phase current, three-phase power, total power, three-phase power factor, total power factor, temperature and humidity, the total power factors are 16 electrical parameters, the data are ordered according to a time sequence, the data collection frequency of an ammeter is 15 minutes each time generally, 96 times per day are collected, and relevant data with continuous longer time are stored for analysis.
Common abnormal faults of electric quantity comprise stop and go of an electric energy meter, abnormal voltage and current, abnormal power factor and the like. The application selects 7 typical electric energy anomalies such as stop-and-go of an electric energy metering device, constant magnetic field interference, voltage unbalance, current loss, power factor anomaly, reverse connection of wiring, under-voltage of a battery of the metering device and the like for analysis, wherein the voltage unbalance and the current unbalance are indexes for reflecting the distribution condition of voltage and current of each phase, and since the voltage and the current are directional, the wiring error can reverse the numerical value, so that calculation errors are caused, and the voltage unbalance can be expressed as follows:
(1-10)
in the embodiment, the wiring error or the abnormal power factor of the electric energy meter can cause one or two phases of split-phase current to be inverted, so that a current inversion sign can be introducedAnd average current +.>The index is expressed as:
(1-11)
(1-12)
I a the current is a current of the phase a,I b the current is the current of the B phase,I c the current is a current of the C phase,maximum value of phase current>Phase current minimum.
To collect the electrical quantities according to the above, a sample set is formed using daily data, and the sample size of the resulting data set is 96×16, representing that 16 data channels are collected 96 times. Before processing the acquired data, taking the inconsistency of the data proportion into consideration, performing linear proportion normalization processing on the data, and regarding the original data at the moment iNormalized data ∈>Can be expressed as:
(1-13)
in the formula ,is the maximum value of the corresponding data channel.
In addition, if the time series of the acquired data of different channels are inconsistent, the data of different channels need to be aligned by interpolation.
In this embodiment, the electrical energy metering diagnosis data contains a sample set of missing or abnormal data, each electrical parameter is classified by using a KNN algorithm, and missing items on the electrical parameter time sequence are repaired according to data correlation between the various types.
And carrying out model training on the training set samples marked with the sample point categories by the KNN algorithm, searching K neighbors for the samples to be tested according to the test sample information in the test set after training is completed, and judging the categories of the samples to be tested according to the selected K neighbors. The core of the data missing repair model is a KNN algorithm, and a plurality of neighbors are selected to be the core problem of the whole model. The selection of the distance between the different sample points of the euclidean distance metric in this embodiment can be expressed as:
(1-14)
wherein x represents a sample to be tested, y represents a training set sample,the distance between the sample to be measured and the training sample is represented, and when the sample to be measured is similar to the training sample, the smaller the distance is, mb represents the dimension of the sample. The sample to be measured here may be an electrical parameter, i.e. an electrical parameter to be measured.
The value of K in KNN calculation influences the classification result, the larger the K is selected, the more inaccurate the classification is possible, and the smaller the K is selected in the same way, the classification error is caused. Therefore, the K value selection optimizing problem is the key of the KNN algorithm, adopts the cross-validation idea, and uses training set data to optimize the K value. And the sample data are divided into different training sets and verification sets in a crossing way, the divided training sets are used as training data, the accuracy of a verification set test model is used, and the average value of the accuracy of the test set is calculated as a final result. The cross verification is applied to different K values, the super parameter with the highest precision is selected as the K value created by the model, and the implementation flow is as follows:
(1) Dividing N training set data into N sub-sample sets；
(2) Using sub-sample setsAs verification data, the rest as training data;
(3) Calculating verification accuracy;
(4) Repeating the steps (2) and (3) by using different test sets until all equal parts are used as trained data;
(5) And averaging the accuracy as an estimate of the accuracy of the unknown data prediction.
According to the embodiment, a KNN algorithm is improved by introducing a weighting mode, and because the traditional KNN algorithm adopts a few ideas obeying majority to judge the category of the sample points to be classified, classification errors possibly occur, aiming at the problem, a weighting idea is introduced, higher weight is given to the nearest neighbor sample points to be classified, the weight is smaller as the distance is far, and the weighting processing is carried out on the basis of a Gaussian function, wherein the Gaussian function is as follows:
(1-15)
wherein a is the height of the peak of the Gaussian curve, and a>0; b is the x coordinate value of the peak center of the Gaussian curve; c is the standard deviation, controlling the width of the gaussian. Weight function of KNN algorithmExpressed as:
(1-16)
k is the selected number of neighbors for the distance between the sample point to be measured and the first neighbor point.
Further, as a refinement and extension of the specific implementation manner of the foregoing embodiment, for fully describing the specific implementation process in this embodiment, the method includes: collected data setI.e. the data set contains m electrical parameters;
extracting the characteristics of electrical parameters in the data set by utilizing an improved KNN algorithm, and classifying the abnormal data and the normal data to form an abnormal data sample set and a normal data sample set;
selecting a normal data sample set from the training set samples, calculating the Pelson coefficient between each electrical parameter by utilizing correlation analysis, screening out the electrical parameter strongly related to the abnormal electrical parameter type, establishing a Pelson correlation coefficient matrix sigma between each electrical parameter, and calculating the weight of each electrical parameter by utilizing an entropy weight method。
Electrical parametersIs expressed as +.>For electrical parameters->Is a strongly associated attribute +.>Corresponding data set->Find and +.>The data at the same time series point is expressed as +.>The cluster center of the class where the value is +.>。
For electrical parametersAll m strongly related electrical parameters are selected, and the abnormal data is obtained through weighted summationIs the optimum correction result of (a): />。
The power metering fault diagnostic model is also trained during the diagnostic process based on the modified twin network. The twin network is a small sample learning method based on similarity measurement, has a simple structure and is easy to train, the twin network is a conjoined neural network, the conjoined neural network is realized through sharing weight, and the twin network is firstly used for verifying whether the signature on a check is consistent with a reserved signature of a bank or not and then is used for comparing the similarity of the two inputs.
The twin network of the present embodiment enters the neural network 1 and the neural network 2 for the two input samples 1 and 2, respectively, and evaluates the similarity of the two neural networks by calculating the loss function, as shown in fig. 6. Because of the weight sharing, it is generally used to deal with the problem that the difference between two inputs is not very large, such as comparing two pictures, two sentences, and the similarity of two words, as shown in fig. 5.
In the embodiment, the idea of sharing a weight twin network is adopted, the characteristic extraction is composed of LSTM, CNN and pooling layers by combining the data characteristics of the electric energy metering anomaly diagnosis, the time information extraction of electric energy metering data is carried out by utilizing two LSTM layers in the characteristic extraction module, the space information extraction is carried out by utilizing the convolutional CNN network, and the LSTM utilizes the cycle characteristics of the LSTM, so that the gradient vanishing phenomenon in training can be relieved.
In the training process of the model provided by the embodiment, except for the learning rate, other default parameters are used as weight initialization, regularization is adopted in the loss function mainly for preventing the model from being over-fitted, a mode of combining mini-batch and an adaptive moment estimation algorithm (Adaptive Moment Estimation, adam) is adopted for model training and optimization, and the Adam optimization algorithm has the characteristics of considering both first moment and second moment and the like. Model learning is dynamically monitored during training and the learning rate and training dataset size are adjusted by the performance of the model during training, and furthermore, the LSTM and CNN parameters, regularization parameters, and number of learning rounds are optimized by the performance of the model on the validation set. The training mechanism used by the twin network calculates the learning rate of each parameter and updates the model parameters according to the following rules.
(6-27)
wherein ,is the learning rate at the t training round,>regarding variables for loss function->Gradient of-> and />Representing a first moment estimate and a second moment estimate, < +.> and />Representing the corrected first and second moment estimates,/->For learning rate step size, ++>Is a constant parameter->。
The implementation flow of the electric energy metering fault diagnosis model based on the twin network is shown as follows, firstly, the data after the data loss repair is divided into a training set and a sample set, then the training and the testing are respectively carried out on the data after the data loss repair, and model parameters are optimized, as shown in fig. 5.
Therefore, the small sample learning electric energy fault diagnosis lightweight method provided by the application operates in the edge internet of things agent, such as a diagnosis terminal, and aims at the characteristic that the data correlation of the current, the voltage, the power and the like collected in the electric energy metering sample is strong, and the fault diagnosis of the electric energy metering device is realized by using the low-cost intelligent terminal. And combining an improved KNN and a data missing repair model for correlation analysis to repair and complement missing or abnormal electric energy metering data. And realizes rapid diagnosis and type judgment of faults. The fault rapid diagnosis efficiency is improved, and the electric energy metering risk is controlled, so that timeliness and scientificity of overall process supervision, management and control of electric energy metering are realized.
It should be understood that the sequence number of each step in the foregoing embodiment does not mean that the execution sequence of each process should be determined by the function and the internal logic, and should not limit the implementation process of the embodiment of the present application.
Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and the various illustrative elements and steps are described above in terms of functions generally in order to clearly illustrate the interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
The electrical energy metering fault diagnosis method based on small sample learning provided by the application is units and algorithm steps of each example described in connection with the embodiment disclosed in the application, can be realized by electronic hardware, computer software or a combination of the two, and in order to clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described in terms of functions in the above description. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (9)

1. Electric energy metering fault diagnosis method based on small sample learning, wherein the method defines collected data set asI.e. the data set contains m electrical parameters; the fault diagnosis method is characterized by comprising the following steps:
s1: for data set X i Extracting characteristics of the electrical parameters in the data processing system, classifying the abnormal data and the normal data to form an abnormal data sample set and a normal data sample set;
s2: selecting a normal data sample set as a training set sample, and calculating the pearson coefficients among the electrical parameters by using correlation analysis; based on abnormal electrical parameters, screening out electrical parameters which are strongly related to the abnormal electrical parameters, establishing a pearson correlation coefficient matrix sigma among the electrical parameters, and calculating the weight of each electrical parameter by using an entropy weight method；
S3: definition of electrical parametersIs expressed as +.>For electrical parameters->Is a strongly associated attribute +.>In data set->Find and +.>Data of the same time sequence point, denoted +.>Configuration->The cluster center of the class is +.>；
S4: based on electrical parametersSelecting all m electrical parameters, and carrying out weighted summation on the m electrical parameters to obtain abnormal data +.>Is the optimum correction result of (a): />Further realizing the repair and the repair of abnormal data;
s5: based on the repaired electrical data, a twin network construction model is adopted, and the nearest neighbor class is found to determine the final classification result by calculating the distance between the sample to be detected and the known label sample;
the twin network adopts LSTM and CNN networks as feature extraction networks, two input samples are received simultaneously by using two sub-networks with shared weights, and the output result is the similarity of the two samples by combining the improved loss function.
2. The method for diagnosing a power metering failure based on small sample learning as claimed in claim 1, wherein,
and step S1, a data deletion repair model is constructed in a mode of combining clustering and correlation analysis, all input electrical parameters are gathered into different sets by using a KNN clustering algorithm, and data deletion repair is realized through data correlation analysis.
3. The method for diagnosing a power metering failure based on small sample learning as claimed in claim 2, wherein,
in step S2, the pearson correlation coefficient is used to evaluate the correlation between two data, and when the pearson correlation coefficient is greater than a preset characteristic value, it is determined that the two data have a strong correlation.
4. The method for diagnosing a power metering failure based on small sample learning as claimed in claim 3, wherein,
the pearson coefficients between the various electrical parameters are expressed as:
(1-1)
wherein ,two n-dimensional data variables, +.>Respectively->Standard deviation of>Is the covariance between the two variables;
the covariance calculation formula is as follows:
(1-2)
wherein 、/>For n-dimensional variables-> and />Is a data average of (a).
5. The method for diagnosing a power metering failure based on small sample learning as claimed in claim 2, wherein,
in step S2, the entropy weighting method determines the weight by indicating the information content, calculates the entropy between the variable j and other variablesExpressed as:
(1-3)
wherein S is the number of variables.
6. The method for diagnosing a power metering failure based on small sample learning as claimed in claim 2, wherein,
weighting of the respective electrical parameters in step S2Expressed as:
(1-4)
wherein the weight isIndicating the size of the index information amount, +.>。
7. The method for diagnosing an electric energy metering fault based on small sample learning according to claim 2, wherein the KNN clustering algorithm performs weighting processing based on a gaussian function, and the gaussian function is as follows:
(1-15)
wherein a is the height of the peak of the Gaussian curve, and a >0; b is the x coordinate value of the peak center of the Gaussian curve; c is the standard deviation, and the width of the Gaussian curve is controlled;
weight function of KNN algorithmExpressed as:
(1-16)
for the distance between the electrical parameter to be measured and the first adjacent point, k is selectedAdjacent number.
8. The small sample learning-based power metering fault diagnosis method according to claim 1 or 2, further comprising the step of performing linear proportion normalization processing on the collected electrical parameters, wherein for the collected electrical parametersiTime of day original electrical parametersNormalized data ∈>Can be expressed as:
(1-13)
in the formula ,is the maximum value of the corresponding data channel;
the method also defines a current inversion flagAnd average current +.>；
The concrete steps are as follows:
(1-11)
(1-12)。
9. the diagnosis terminal is characterized in that the diagnosis terminal adopts a twin network to construct a model based on the repaired electrical data, and the nearest neighbor class is found to determine the final classification result by calculating the distance between a sample to be detected and a known label sample; the twin network adopts LSTM and CNN networks as feature extraction networks, two input samples are received simultaneously by using two sub-networks with shared weights, and the output result is the similarity of the two samples by combining the improved loss function;
in particular comprising a memory, a processor and a computer program stored on said memory and executable on said processor, said processor implementing the steps of the small sample learning based power metering fault diagnosis method according to any one of claims 1 to 8 when said program is executed.
CN202311119877.8A 2023-09-01 2023-09-01 Electric energy metering fault diagnosis method and diagnosis terminal based on small sample learning Active CN116842459B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311119877.8A CN116842459B (en) 2023-09-01 2023-09-01 Electric energy metering fault diagnosis method and diagnosis terminal based on small sample learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311119877.8A CN116842459B (en) 2023-09-01 2023-09-01 Electric energy metering fault diagnosis method and diagnosis terminal based on small sample learning

Publications (2)

Publication Number Publication Date
CN116842459A true CN116842459A (en) 2023-10-03
CN116842459B CN116842459B (en) 2023-11-21

Family

ID=88162099

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311119877.8A Active CN116842459B (en) 2023-09-01 2023-09-01 Electric energy metering fault diagnosis method and diagnosis terminal based on small sample learning

Country Status (1)

Country Link
CN (1) CN116842459B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117093946A (en) * 2023-10-20 2023-11-21 国网天津市电力公司营销服务中心 Method and device for analyzing abnormality of gateway electric energy metering device

Citations (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109976318A (en) * 2019-04-28 2019-07-05 郑州万特电气股份有限公司 Electrical energy measurement fault diagnosis Internet-based checks expert system
CN110412417A (en) * 2019-07-15 2019-11-05 中国人民解放军陆军工程大学 Micro-capacitance sensor data fault diagnostic method based on intelligent power monitoring instrument table
CN110727669A (en) * 2019-10-10 2020-01-24 国网辽宁省电力有限公司电力科学研究院 Device and method for cleaning sensor data of power system
WO2020119481A1 (en) * 2018-12-11 2020-06-18 深圳先进技术研究院 Network traffic classification method and system based on deep learning, and electronic device
CN111950526A (en) * 2020-09-01 2020-11-17 国网河北省电力有限公司检修分公司 Fault diagnosis method for energy storage mechanism of circuit breaker based on deep learning
CN112651435A (en) * 2020-12-22 2021-04-13 中国南方电网有限责任公司 Self-learning-based detection method for flow abnormity of power network probe
CN113049035A (en) * 2021-03-12 2021-06-29 辽宁工程技术大学 Transformer state monitoring system based on Internet of things
US11080127B1 (en) * 2018-02-28 2021-08-03 Arizona Public Service Company Methods and apparatus for detection of process parameter anomalies
US20210241041A1 (en) * 2020-01-31 2021-08-05 Element Ai Inc. Method of and system for joint data augmentation and classification learning
CN113505531A (en) * 2021-07-06 2021-10-15 北京化工大学 Diagnosis model construction method for reciprocating mechanical valve clearance abnormal fault based on combination of traditional diagnosis and machine learning method
CN114020906A (en) * 2021-10-20 2022-02-08 杭州电子科技大学 Chinese medical text information matching method and system based on twin neural network
CN114069838A (en) * 2021-10-05 2022-02-18 国网辽宁省电力有限公司电力科学研究院 Transformer substation robot intelligent inspection system and method with intelligent sensor actively cooperated
CN114723583A (en) * 2022-03-24 2022-07-08 天津三源电力信息技术股份有限公司 Unstructured electric power big data analysis method based on deep learning
CN114818940A (en) * 2022-04-28 2022-07-29 华北电力科学研究院有限责任公司 GAN-based line-variable relation identification method for processing unbalanced data
CN114970950A (en) * 2022-04-13 2022-08-30 国家能源集团国源电力有限公司 Fan fault alarm method and device, storage medium and electronic equipment
CN115127192A (en) * 2022-05-20 2022-09-30 中南大学 Semi-supervised water chilling unit fault diagnosis method and system based on graph neural network
CN115204302A (en) * 2022-07-22 2022-10-18 贵州大学 Unmanned aerial vehicle small sample fault diagnosis system and method
CN115270883A (en) * 2022-08-02 2022-11-01 泾县市场监督检验所(安徽省电机产品及零部件质量监督检验中心) Motor bearing fault diagnosis method based on twin neural network small sample learning
CN115659174A (en) * 2022-10-13 2023-01-31 中国科学院空间应用工程与技术中心 Multi-sensor fault diagnosis method, medium and equipment based on graph regularization CNN-BilSTM
CN115907984A (en) * 2022-11-14 2023-04-04 合肥工业大学 Point-to-point electric power transaction method and system based on supply and demand prediction under block chain environment
CN116358871A (en) * 2023-03-29 2023-06-30 哈尔滨理工大学 Rolling bearing weak signal composite fault diagnosis method based on graph rolling network
US20230267721A1 (en) * 2022-02-24 2023-08-24 Vinbrain Joint Stock Company Method and system for training a machine learning model for medical image classification

Patent Citations (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11080127B1 (en) * 2018-02-28 2021-08-03 Arizona Public Service Company Methods and apparatus for detection of process parameter anomalies
WO2020119481A1 (en) * 2018-12-11 2020-06-18 深圳先进技术研究院 Network traffic classification method and system based on deep learning, and electronic device
CN109976318A (en) * 2019-04-28 2019-07-05 郑州万特电气股份有限公司 Electrical energy measurement fault diagnosis Internet-based checks expert system
CN110412417A (en) * 2019-07-15 2019-11-05 中国人民解放军陆军工程大学 Micro-capacitance sensor data fault diagnostic method based on intelligent power monitoring instrument table
CN110727669A (en) * 2019-10-10 2020-01-24 国网辽宁省电力有限公司电力科学研究院 Device and method for cleaning sensor data of power system
US20210241041A1 (en) * 2020-01-31 2021-08-05 Element Ai Inc. Method of and system for joint data augmentation and classification learning
CN111950526A (en) * 2020-09-01 2020-11-17 国网河北省电力有限公司检修分公司 Fault diagnosis method for energy storage mechanism of circuit breaker based on deep learning
CN112651435A (en) * 2020-12-22 2021-04-13 中国南方电网有限责任公司 Self-learning-based detection method for flow abnormity of power network probe
CN113049035A (en) * 2021-03-12 2021-06-29 辽宁工程技术大学 Transformer state monitoring system based on Internet of things
CN113505531A (en) * 2021-07-06 2021-10-15 北京化工大学 Diagnosis model construction method for reciprocating mechanical valve clearance abnormal fault based on combination of traditional diagnosis and machine learning method
CN114069838A (en) * 2021-10-05 2022-02-18 国网辽宁省电力有限公司电力科学研究院 Transformer substation robot intelligent inspection system and method with intelligent sensor actively cooperated
CN114020906A (en) * 2021-10-20 2022-02-08 杭州电子科技大学 Chinese medical text information matching method and system based on twin neural network
US20230267721A1 (en) * 2022-02-24 2023-08-24 Vinbrain Joint Stock Company Method and system for training a machine learning model for medical image classification
CN114723583A (en) * 2022-03-24 2022-07-08 天津三源电力信息技术股份有限公司 Unstructured electric power big data analysis method based on deep learning
CN114970950A (en) * 2022-04-13 2022-08-30 国家能源集团国源电力有限公司 Fan fault alarm method and device, storage medium and electronic equipment
CN114818940A (en) * 2022-04-28 2022-07-29 华北电力科学研究院有限责任公司 GAN-based line-variable relation identification method for processing unbalanced data
CN115127192A (en) * 2022-05-20 2022-09-30 中南大学 Semi-supervised water chilling unit fault diagnosis method and system based on graph neural network
CN115204302A (en) * 2022-07-22 2022-10-18 贵州大学 Unmanned aerial vehicle small sample fault diagnosis system and method
CN115270883A (en) * 2022-08-02 2022-11-01 泾县市场监督检验所(安徽省电机产品及零部件质量监督检验中心) Motor bearing fault diagnosis method based on twin neural network small sample learning
CN115659174A (en) * 2022-10-13 2023-01-31 中国科学院空间应用工程与技术中心 Multi-sensor fault diagnosis method, medium and equipment based on graph regularization CNN-BilSTM
CN115907984A (en) * 2022-11-14 2023-04-04 合肥工业大学 Point-to-point electric power transaction method and system based on supply and demand prediction under block chain environment
CN116358871A (en) * 2023-03-29 2023-06-30 哈尔滨理工大学 Rolling bearing weak signal composite fault diagnosis method based on graph rolling network

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
蒲天骄等: "人工智能技术在电力设备运维检修中的研究及应用", 高电压技术 *
陈淑梅等: "卷积神经网络多变量过程特征学习与故障诊断", 哈尔滨工业大学学报 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117093946A (en) * 2023-10-20 2023-11-21 国网天津市电力公司营销服务中心 Method and device for analyzing abnormality of gateway electric energy metering device

Also Published As

Publication number Publication date
CN116842459B (en) 2023-11-21

Similar Documents

Publication Publication Date Title
CN108734355B (en) Short-term power load parallel prediction method and system applied to power quality comprehensive management scene
Kapucu et al. A supervised ensemble learning method for fault diagnosis in photovoltaic strings
CN108133225A (en) A kind of icing flashover fault early warning method based on support vector machines
CN116842459B (en) Electric energy metering fault diagnosis method and diagnosis terminal based on small sample learning
CN112767106B (en) Automatic auditing method, system, computer readable storage medium and auditing equipment
CN114676742A (en) Power grid abnormal electricity utilization detection method based on attention mechanism and residual error network
CN116721537A (en) Urban short-time traffic flow prediction method based on GCN-IPSO-LSTM combination model
CN116449218B (en) Lithium battery health state estimation method
Song et al. Short-term forecasting based on graph convolution networks and multiresolution convolution neural networks for wind power
CN110879377A (en) Metering device fault tracing method based on deep belief network
CN112305441B (en) Power battery health state assessment method under integrated clustering
CN112287980B (en) Power battery screening method based on typical feature vector
CN115374995A (en) Distributed photovoltaic and small wind power station power prediction method
CN109686402A (en) Based on key protein matter recognition methods in dynamic weighting interactive network
CN111562541B (en) Software platform for realizing electric energy meter detection data management by applying CART algorithm
CN115757103A (en) Neural network test case generation method based on tree structure
CN116186633A (en) Power consumption abnormality diagnosis method and system based on small sample learning
CN114548199A (en) Multi-sensor data fusion method based on deep migration network
Yin et al. Non-intrusive load monitoring by load trajectory and multi feature based on DCNN
CN113642255A (en) Photovoltaic power generation power prediction method based on multi-scale convolution cyclic neural network
CN113033898A (en) Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network
CN116706888A (en) Intelligent power grid load prediction method based on federal learning
CN115734274A (en) Cellular network fault diagnosis method based on deep learning and knowledge graph
CN115659258A (en) Power distribution network fault detection method based on multi-scale graph convolution twin network
Wang et al. LTE network quality analysis method based on MR data and XGBoost algorithm

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