Search

CN-121997114-A - Fault detection classification method for energy storage battery and storage medium

CN121997114ACN 121997114 ACN121997114 ACN 121997114ACN-121997114-A

Abstract

The invention discloses a fault detection classification method and a storage medium of an energy storage battery, wherein the method utilizes normal operation data to complete unsupervised abnormal detection, obtains fault characteristics of the abnormal data aiming at the obtained abnormal data, and completes classification of the characteristics; and after triggering the abnormality, calling a supervision classification model to give a fault type probability. The storage medium has stored therein a computer program for executing the above method. The invention has the advantages of simple principle, easy realization, wide application range and the like.

Inventors

  • LIU ZHIWEI
  • ZHANG JINGYA
  • XIONG CHAO

Assignees

  • 湖南云储循环新能源科技有限公司

Dates

Publication Date
20260508
Application Date
20251211

Claims (10)

  1. 1. A fault detection and classification method for an energy storage battery is characterized in that normal operation data are utilized to complete unsupervised abnormal detection, fault characteristics of the abnormal data are obtained aiming at the obtained abnormal data, classification of the characteristics is completed, and after triggering of the abnormality, a supervised classification model is called to give out fault type probability.
  2. 2. The fault detection classification method of an energy storage battery according to claim 1, characterized in that the steps include: step S1, collecting voltage, current, temperature and SoC data of a battery cell/module in actual operation Collecting data D of normal operation and data when faults occur, and performing fault classification labeling to obtain a fault label data set ; Step S2, aiming at normal operation data The method comprises the steps of carrying out channel normalization on window data, and dividing the data into a training set, a testing set and a verification set, wherein the training set is used for training an MC-LSTM-AE self-coding model, the verification set is used for adjusting model parameters, and the testing set is used for evaluating model performance; Step S3, training an MC-LSTM-AE self-coding model by adopting a training set in normal operation data, and training and adjusting model parameters through verification set to obtain a trained self-coder model and a reconstruction matrix with the same dimension as the training set; s4, training the SVDD model by adopting errors between a training data set and a reconstruction matrix to obtain the minimum hypersphere radius ; Step S5, failure tag data set The method comprises the steps of dividing the data into a training set, a verification machine and a test set, training DRSN a classification model, and carrying out fault diagnosis classification on abnormal data; And S6, carrying out service deployment on the trained self-encoder model and the classification model to realize two-stage online fault diagnosis, firstly judging whether the data to be detected is abnormal through an abnormality diagnosis module, and if so, inputting the data to be detected into the fault diagnosis classification model to judge which type of fault belongs specifically.
  3. 3. The method of claim 2, wherein the fault type comprises internal short circuit, electrolyte leakage, thermal runaway, and excessive aging of the battery.
  4. 4. The fault detection classification method of the energy storage battery according to claim 1, wherein the unsupervised anomaly detection is based on a multi-channel LSTM self-encoder model, the multi-channel LSTM self-encoder model is used for extracting multi-dimensional time series features for normal operation state data of the battery, attention layer enhancement information extraction is performed, core features of normal operation data of the battery are learned and captured, a reconstruction error threshold of an MC-LSTM-AE model is calculated by adopting a support vector data description algorithm, and the abnormal data is judged according to the threshold.
  5. 5. The method for detecting and classifying faults of an energy storage battery according to claim 4, wherein the updating and utilizing of the historical sequence data are controlled by an input gate, an output gate and a forgetting gate in an LSTM structure based on a multi-channel LSTM self-encoder model, and the multi-channel LSTM is adopted as an encoder and combined with an attention mechanism to realize the connection between different characteristic co-channels of a long-time sequence.
  6. 6. The method of claim 5, wherein the data input to the encoder in the multi-channel LSTM based self-encoder model comprises a time window of battery cells Including voltage, current, temperature, soC, as follows: in the formula, , , , Respectively represent the first Time frame of individual battery cells Voltage, current, temperature and SoC characteristic data; in the encoder, for each characteristic data, an intermediate characteristic representation is obtained by LSTM and linear layers, including voltage data characteristics, Wherein And Respectively obtaining the weight and bias value of the linear layer, and the current, temperature and SoC characteristic channel data by the same operation , , Then all the characteristic channel data are spliced to obtain The input of the multi-head attention mechanism is obtained.
  7. 7. The method of claim 6, wherein in a multi-head attention mechanism, for each attention head, input is made Conversion into three different matrices: , , , wherein, Is a trainable weight, then the attention is calculated as: Wherein, the Is that Is a dimension of (2); Feature fusion is performed by adopting a multi-head attention mechanism, and the expression is expressed as follows: Wherein, the , Is a trainable weight of the attention header.
  8. 8. The method for fault detection classification of an energy storage battery as claimed in claim 7, wherein, In the deep network training process, a residual error normalization layer is adopted and expressed as: Wherein, the Is a feature normalization function that is used to normalize the features, Is the attention layer output; Output by the attention layer Processing is performed through a linear residual normalization layer to further optimize battery characteristics, which are calculated by the following formula: Wherein, therein Representing the output of the linear residual normalization layer, And Training weights and bias terms representing the layer, respectively; Finally, the step of obtaining the product, By a name of Is used for converting the input characteristics into a low-dimensional space and obtaining the output of the encoder through a linear layer The expression is: Wherein, the And Representing training weights and biases of the last linear layer, respectively.
  9. 9. The method of claim 8, wherein the supervised classification model is configured to classify the obtained abnormal data by obtaining a fault feature of the abnormal data using the deep residual shrinkage network DRSN.
  10. 10. A storage medium readable by a computer or a processor, characterized in that the storage medium has stored therein a computer program for executing the method of any of the preceding claims 1 to 9.

Description

Fault detection classification method for energy storage battery and storage medium Technical Field The invention mainly relates to the technical field of battery fault early warning, in particular to a fault detection classification method and a storage medium of an energy storage battery. Background Lithium batteries have become the first choice of energy sources in various fields by virtue of the advantages of high energy ratio, long service life, low self-discharge rate and the like. However, in recent years, fire and explosion accidents caused by faults such as thermal runaway and internal short circuit frequently occur, so that serious economic loss and social influence are caused. The battery fault evolution has concealment, nonlinearity and time variability, and the traditional method is difficult to realize high-precision and low-delay early warning, so that the early prediction and real-time monitoring technology of the battery fault still face significant challenges. The existing battery fault early warning technology mostly adopts a mode of a single threshold value or an empirical model, but has the defects of false alarm, high missing report rate and the like, and cannot adapt to battery aging drift. Practitioners have also proposed methods employing supervised deep learning, but rely on a large number of fault labels, with practical fault samples being scarce and models being prone to overfitting. In addition, practitioners propose an unsupervised abnormality detection mode, and although the mode can give an alarm in advance, specific fault types cannot be informed, fault problems cannot be positioned, and operation and maintenance staff are difficult to maintain in a targeted mode. If abnormal detection and fault classification are directly trained end-to-end in series, a normal sample dominates the loss function, so that abnormal characteristics are submerged, and classification accuracy is low. Therefore, a dual-stage fault detection framework is needed, which is capable of fully mining normal data boundaries by using an unsupervised model and finely distinguishing fault types by using the supervised model, and is capable of engineering landing. Disclosure of Invention The invention aims to solve the technical problems in the prior art, and provides a fault detection classification method and a storage medium for an energy storage battery, which are simple in principle, easy to implement and wide in application range. In order to solve the technical problems, the invention adopts the following technical scheme: a fault detection classification method of an energy storage battery utilizes normal operation data to complete unsupervised abnormal detection, obtains fault characteristics of the abnormal data aiming at the obtained abnormal data, completes classification of the characteristics, and calls a supervised classification model to give fault type probability after triggering abnormality. As a further improvement of the invention, the method comprises the following steps: step S1, collecting voltage, current, temperature and SoC data of a battery cell/module in actual operation Collecting data D of normal operation and data when faults occur, and performing fault classification labeling to obtain a fault label data set; Step S2, aiming at normal operation dataThe method comprises the steps of carrying out channel normalization on window data, and dividing the data into a training set, a testing set and a verification set, wherein the training set is used for training an MC-LSTM-AE self-coding model, the verification set is used for adjusting model parameters, and the testing set is used for evaluating model performance; Step S3, training an MC-LSTM-AE self-coding model by adopting a training set in normal operation data, and training and adjusting model parameters through verification set to obtain a trained self-coder model and a reconstruction matrix with the same dimension as the training set; s4, training the SVDD model by adopting errors between a training data set and a reconstruction matrix to obtain the minimum hypersphere radius ; Step S5, failure tag data setThe method comprises the steps of dividing the data into a training set, a verification machine and a test set, training DRSN a classification model, and carrying out fault diagnosis classification on abnormal data; And S6, carrying out service deployment on the trained self-encoder model and the classification model to realize two-stage online fault diagnosis, firstly judging whether the data to be detected is abnormal through an abnormality diagnosis module, and if so, inputting the data to be detected into the fault diagnosis classification model to judge which type of fault belongs specifically. As a further improvement of the present invention, the failure type includes internal short circuit, electrolyte leakage, thermal runaway, excessive aging of the battery, and the like. As a further improvement of the invention,