CN-121980497-A - Power equipment fault diagnosis method and system based on multi-source data fusion

CN121980497ACN 121980497 ACN121980497 ACN 121980497ACN-121980497-A

Abstract

The application discloses a power equipment fault diagnosis method and system based on multi-source data fusion, wherein the application extracts the linear characteristics of each data source from preprocessed multi-source data through an improved Principal Component Analysis (PCA) algorithm; the linear characteristics of multiple data sources are fused through a weighted fusion strategy, a fault diagnosis model based on the mixture of the attention mechanism two-way long-short-term memory network BiLSTM and the convolutional neural network CNN is constructed, the pre-trained fault diagnosis model is utilized for carrying out fault diagnosis on the power equipment, and a fault type identification result and a fault degree evaluation value are output. The application realizes accurate and real-time diagnosis and quantitative evaluation of fault degree of the power equipment through multi-source data acquisition, layered fusion and mixed diagnosis models, and simultaneously captures spatial information and time sequence information of the features, highlights key features and improves the accuracy of fault diagnosis by adopting a BiLSTM and CNN mixed model based on an attention mechanism.

Inventors

HUANG FUQIANG
YANG HAI
ZHANG PAN
QU WENFENG
REN GANG

Assignees

三峡金沙江川云水电开发有限公司

Dates

Publication Date: 20260505
Application Date: 20260106

Claims (10)

1. The power equipment fault diagnosis method based on the multi-source data fusion is characterized by comprising the following steps of: s1, acquiring multi-source data of power equipment and preprocessing the multi-source data; S2, extracting linear characteristics of each data source from the preprocessed multi-source data by adopting an improved Principal Component Analysis (PCA) algorithm, and optimizing principal component selection by introducing an adaptive weight coefficient so as to improve the Principal Component Analysis (PCA) algorithm; s3, fusing linear characteristics of multiple data sources through a weighted fusion strategy to obtain multiple-source fusion characteristics; s4, constructing a fault diagnosis model of the mixture of the two-way long-short-term memory network BiLSTM and the convolutional neural network CNN based on an attention mechanism; and extracting local spatial features and time sequence dependent features from the multisource fusion features by using the trained fault diagnosis model, carrying out weighted fusion on the local spatial features and the time sequence dependent features to obtain weighted fusion features, and analyzing the weighted fusion features to obtain a fault type identification result and a fault degree evaluation value.
2. The power equipment fault diagnosis method based on multi-source data fusion according to claim 1, wherein in step S1, the multi-source data includes electrical quantity data, physical state data, environmental data, and historical fault data of the power equipment.
3. The method for diagnosing a power equipment failure based on multi-source data fusion according to claim 1, wherein in step S2, the improved principal component analysis PCA is used to extract the linear features of the data source, specifically comprising: s21, calculating single data source Covariance matrix of (2) The method comprises the following steps: In the formula, Representing the PCA weight matrix Converting into a diagonal matrix, weighting the original features by a PCA weight matrix, For the number of samples to be taken, Representing a transpose operation; s22, for covariance matrix Performing eigenvalue decomposition, and calculating eigenvectors corresponding to the principal components according to the accumulated variance contribution ratio, wherein: In the formula, As a vector of the eigenvalues, For the corresponding feature vector matrix, Is the data source Is a feature of the original feature dimension of (a); The cumulative variance contribution rate is expressed as: Setting a threshold value Screening for minimum So that Obtaining the front part Feature vectors corresponding to the principal components ; S23, based on final PCA weight matrix and front Feature vectors corresponding to the principal components Calculating linear characteristics The method comprises the following steps: ; In the formula, And (5) obtaining a final PCA weight matrix through multiple iterations.
4. The method for diagnosing a power equipment failure based on multi-source data fusion according to claim 3, wherein in step S3, linear features of the multi-source data are fused by a weighted fusion strategy to obtain multi-source fusion features, and the method specifically comprises the following steps: S31, adopting mutual information entropy quantization Linear characteristics of individual data sources And fault labels The degree of association, i.e. contribution : In the formula, Is that Is used for the information entropy of (a), Is the information entropy of the failure tag, Is the joint information entropy; S32 degree of contribution Normalization processing is carried out to obtain the fusion weight of each data source The method comprises the following steps: In the formula, The number of the data sources; s33, utilizing the fusion weight Linear characteristics for multiple data sources Weighting and splicing to obtain final multisource fusion characteristics The method comprises the following steps: In the formula, Representing the will be Linear characteristics of individual data sources Each element multiplied by a weight And realizing weighted fusion according to the contribution degree.
5. The method for diagnosing a fault of an electrical device based on multi-source data fusion as recited in claim 1, wherein in step S4, the fault diagnosis model includes an input layer, a CNN spatial feature extraction module, a BiLSTM time sequence feature extraction module, an attention fusion module, and an output layer; extracting local spatial features in the multisource fusion features by adopting a CNN spatial feature extraction module; Extracting time sequence dependent features in the multi-source fusion features by adopting BiLSTM time sequence feature extraction modules; And carrying out weighted fusion on the local spatial features and the time sequence dependent features by the attention fusion module to obtain weighted fusion features.
6. The method for diagnosing a power apparatus failure based on multi-source data fusion as recited in claim 5, wherein in step S4, the BiLSTM timing feature extraction module is composed of a forward LSTM and a reverse LSTM, and controls information flow through a gating mechanism.
7. The method for diagnosing a power apparatus fault based on multi-source data fusion as recited in claim 5, wherein in step S4, said attention fusion module outputs local spatial features to the CNN spatial feature extraction module by introducing a time-space joint attention mechanism And BiLSTM time sequence dependency characteristics output by time sequence characteristic extraction module The weighted fusion is carried out, and the method specifically comprises the following steps: s41, local spatial features are processed And timing dependent features Splicing according to time steps to obtain joint characteristics , ; S42, learning the attention weight of each time step through the single hidden layer neural network The method comprises the following steps: ; In the formula, In order to activate the function, As the weight value of the weight value, As a result of the bias term, In order for the attention vector to be of interest, For the transpose operation, Representing joint features Is the first of (2) The characteristics of the individual time steps are that, As a hyperbolic tangent function; S43, local space feature And timing dependent features Weighted summation is carried out according to the attention weight to obtain weighted fusion characteristics The method comprises the following steps: in order to balance the system, 、 Respectively represent local spatial features And timing dependent features Is the first of (2) The characteristics of the individual time steps are that, Is the total time step.
8. The method for diagnosing a fault in a power apparatus based on multi-source data fusion as recited in claim 7, wherein in step S4, the output layer adopts a dual-task parallel structure, which is a fault type classification branch and a fault degree evaluation branch, respectively; The output results are respectively fault type probability distribution , wherein, As the number of types of faults to be detected, For the sample to belong to Probability of class failure and failure degree evaluation value And mapping the output to the [0,1] interval through Sigmoid activation, multiplying by 10 to obtain a final evaluation value, and simultaneously realizing fault type classification and fault degree evaluation.
9. The method for diagnosing a power plant fault based on multi-source data fusion as recited in claim 8, wherein said fault type classification branch loss function in step S4 is cross entropy loss The method comprises the following steps: Wherein, the For the sample Is the first of (2) The type of fault-like one-time encoding, For the sample Is the first of (2) The probability of a class fault type, Is the number of samples; The loss function of the fault degree evaluation branch is mean square error loss : Wherein, the In order to predict the extent of the failure, The real fault degree is obtained; Total loss function The method comprises the following steps: Wherein, the Task weights are categorized.
10. A power equipment fault diagnosis system based on multi-source data fusion, characterized in that the power equipment fault diagnosis system adopts the power equipment fault diagnosis method according to any one of claims 1 to 9 to complete power equipment fault diagnosis; The power equipment fault diagnosis system includes: The multi-source data acquisition module is used for acquiring multi-source data of the power equipment and preprocessing the multi-source data; the linear feature extraction module is used for extracting the linear features of each data source from the preprocessed multi-source data by adopting an improved Principal Component Analysis (PCA) algorithm; The feature fusion module fuses the linear features of the multiple data sources through a weighted fusion strategy to obtain multiple source fusion features; The fault diagnosis model construction module is used for constructing a fault diagnosis model of the mixture of the two-way long-short-term memory network BiLSTM and the convolutional neural network CNN based on an attention mechanism; The fault diagnosis module is used for calling a pre-trained fault diagnosis model, extracting local spatial features and time sequence dependent features from the multisource fusion features by using the trained fault diagnosis model, carrying out weighted fusion on the local spatial features and the time sequence dependent features to obtain weighted fusion features, and analyzing the weighted fusion features to obtain a fault type identification result and a fault degree evaluation value.

Description

Power equipment fault diagnosis method and system based on multi-source data fusion Technical Field The application belongs to the technical field of power equipment fault diagnosis, and particularly relates to a power equipment fault diagnosis method and system based on multi-source data fusion. Background The power equipment is a core guarantee for safe and stable operation of the power system, and the operation state of the power equipment directly influences the reliability and safety of power supply. Along with the development of the power system to intellectualization and large capacity, the structure of the power equipment is increasingly complex, the running environment is worse, and the probability of faults is gradually increased. The power equipment fault can be diagnosed timely and accurately, and the method has important significance for reducing operation and maintenance cost and avoiding major safety accidents. The existing power equipment fault diagnosis method is mostly based on a single data source, for example, fault judgment is carried out only through electrical quantity data or partial discharge signals. However, the single data source has the defects of one-sided information and weak anti-interference capability, and the running state of the power equipment is difficult to comprehensively reflect, so that the fault diagnosis accuracy is low, and the missed diagnosis rate and the misdiagnosis rate are high. For example, when the fault of a transformer is diagnosed by only depending on the content of dissolved gas in oil, the transformer is easily influenced by factors such as ambient temperature, humidity and the like, so that the deviation of diagnosis results is caused, and when the fault of a circuit breaker is diagnosed based on a vibration signal, if the signal is interfered by the outside, the effectiveness of fault feature extraction is reduced. While some prior art attempts to employ multi-source data for fault diagnosis, there are shortcomings in data fusion strategies and diagnostic model designs. On one hand, the traditional data fusion method is mainly simple in characteristic splicing or weighted summation, internal association and complementary information among multi-source data cannot be fully mined, and on the other hand, the existing diagnosis model is mainly a single machine learning algorithm or a deep learning model, so that spatial characteristics and time sequence characteristics of the multi-source data are difficult to process simultaneously, and real-time performance and accuracy of fault diagnosis are difficult to meet actual requirements. Therefore, there is a need for a power equipment fault diagnosis method capable of efficiently fusing multi-source data, accurately extracting fault characteristics, and rapidly identifying fault types. Disclosure of Invention The application aims to overcome the problems in the prior art and discloses a power equipment fault diagnosis method and system based on multi-source data fusion. On the one hand, the aim of the application is achieved by the following technical scheme: a power equipment fault diagnosis method based on multi-source data fusion, the power equipment fault diagnosis method based on multi-source data fusion comprising: s1, acquiring multi-source data of power equipment and preprocessing the multi-source data; S2, extracting linear characteristics of each data source from the preprocessed multi-source data by adopting an improved Principal Component Analysis (PCA) algorithm, and optimizing principal component selection by introducing an adaptive weight coefficient so as to improve the Principal Component Analysis (PCA) algorithm; s3, fusing linear characteristics of multiple data sources through a weighted fusion strategy to obtain multiple-source fusion characteristics; s4, constructing a fault diagnosis model of the mixture of the two-way long-short-term memory network BiLSTM and the convolutional neural network CNN based on an attention mechanism; and extracting local spatial features and time sequence dependent features from the multisource fusion features by using the trained fault diagnosis model, carrying out weighted fusion on the local spatial features and the time sequence dependent features to obtain weighted fusion features, and analyzing the weighted fusion features to obtain a fault type identification result and a fault degree evaluation value. According to a preferred embodiment, in step S1, the multi-source data includes electrical quantity data, physical state data, environmental data, and historical fault data of the electrical device. According to a preferred embodiment, in step S2, the linear features of the PCA extracted data source are analyzed using the modified principal component, comprising in particular: s21, calculating single data source Covariance matrix of (2)The method comprises the following steps: In the formula, Representing the PCA weight matrixConverting into a diagon