Electric vehicle battery thermal runaway diagnosis method based on unsupervised adversarial learning

Abstract

The invention belongs to the crossing field of batteries and artificial intelligence, and relates to an electric vehicle battery thermal runaway diagnosis method based on unsupervised adversarial learning, and the method comprises the steps: obtaining an electric vehicle battery data set, carrying out the preprocessing of the electric vehicle battery data set, and obtaining a training sample set and a test sample set; each training sample in the training sample set is normal electric vehicle battery data; constructing an integrated model, and performing two-stage adversarial training on the integrated model by using the training sample set to obtain a trained integrated model; inputting the test sample set into the trained integrated model to obtain a reconstruction result; obtaining a thermal runaway diagnosis result according to a reconstruction result by adopting a self-adaptive integrated voting method; according to the self-adaptive integrated voting method, the voting weight can be dynamically adjusted according to the performance of each model in the test stage, so that the model with better performance plays a greater role in final decision making, and the accuracy of a diagnosis result is further improved.

Description

Technical Field

[0001] The present invention belongs to the intersection field of batteries and artificial intelligence, and relates to a thermal runaway diagnosis method for electric vehicle batteries based on unsupervised adversarial learning. Background Art

[0002] Vigorously developing electric vehicles is one of the necessary means to deal with the depletion of fossil energy and carbon emissions. An important component of electric vehicles is the battery system, which plays a key role in many aspects such as vehicle safety. With the continuous development of new energy vehicles, car fire accidents are increasing day by day, of which more than 90% are caused by battery thermal runaway, which seriously threatens the property and life safety of drivers and passengers.

[0003] To solve the thermal runaway warning problem, existing methods can be divided into four categories: rule-based methods, experimental methods, model-based methods, and data-driven methods. Among them, the data-driven method can avoid complex mechanism analysis by adaptively mining the thermal runaway data characteristics of electric vehicle batteries from massive data. It has the advantages of simple calculation and strong scalability. Therefore, it has attracted more and more attention and has become one of the emerging research hotspots in this field.

[0004] At present, most data-driven methods are supervised methods, which learn the mapping relationship between the operating characteristics of electric vehicle batteries and the probability of thermal runaway through traditional machine learning and neural network methods. However, these supervised methods have two main problems: i) First, due to the scarcity of thermal runaway battery data, there is a serious imbalance in the number of normal samples and fault samples. ii) More importantly, due to the diversity of thermal runaway types, the existing thermal runaway data cannot cover all possible thermal runaways. Therefore, the existing thermal runaway fault diagnosis supervision methods are difficult to apply in practice.

[0005] Unsupervised methods can perform pattern mining on unlabeled data sets and are insensitive to the scarcity of thermal runaway battery data. This method is suitable for electric vehicle battery abnormality judgment problems with a large number of normal samples. At present, clustering methods are mainly used to diagnose battery failures. Batteries are classified by feature distance, but the difference between electric vehicle batteries in the early and middle stages of thermal runaway and normal batteries is small. Therefore, it is difficult to distinguish thermal runaway electric vehicle batteries by unsupervised clustering using only simple distance calculation methods. The integrated model combines the prediction results of multiple basic models according to the integrated method, which can effectively avoid the one-sidedness of the judgment of a single model and improve the stability of the overall model.

[0006] However, existing ensemble models lack robustness, and their ensemble methods are usually voting, averaging or weighted averaging, etc., which fail to consider the performance of each basic model in the test phase and dynamically adjust the voting weights, resulting in low accuracy of the diagnostic results. Summary of the invention

[0007] In order to solve the above-mentioned problems in the prior art, the present invention adopts an electric vehicle battery thermal runaway diagnosis method based on unsupervised adversarial learning, comprising:

[0008] S1. Obtain an electric vehicle battery data set, preprocess the electric vehicle battery data set, and obtain a training sample set and a test sample set; each training sample in the training sample set is normal electric vehicle battery data;

[0009] S2. Construct an integrated model, and perform two-stage adversarial training on the integrated model using the training sample set to obtain a trained integrated model; the integrated model includes N basic models; the two-stage adversarial training includes: a first-stage reconstruction training and a second-stage adversarial training;

[0010] S3. Input the test sample set into the trained integrated model to obtain the reconstruction result of each basic model; and use the adaptive integrated voting method to obtain the thermal runaway diagnosis result based on the reconstruction result.

[0011] The training sample set includes N sub-training sample sets; the basic model n includes: model ED n，1 and Model ED n，2 , ED is the encoder-decoder model; the two-stage adversarial training process for each basic model n includes:

[0012] S21, input the training samples X of the nth sub-training sample set into ED n，1 and ED n，2 Performing reconstruction training to obtain a first reconstruction result and a second reconstruction result;

[0013] S22, input the first reconstruction result into ED n，2 Conduct adversarial training to obtain generated results;

[0014] S23. Calculate the loss function value according to the sample X, the first reconstruction result, the second reconstruction result and the generated result, update the model parameters according to the loss function value, and obtain the trained basic model n when the number of iterations reaches the preset maximum number of iterations.

[0015] The loss function is:

[0016]

[0017] in, For ED n，1 The loss function is For ED n，2 The loss function of .

[0018] ED n，1Includes: Encoder n and Decoder n，1 , E.D. n，2 Including Encoder n and Decoder n，2 ;ED n，1 Processing the training sample X includes: inputting the training sample X into the encoder Encoder n , the encoder Encoder n The output of the decoder n，1 , and obtain the first reconstruction result.

[0019] Encoder n , Decoder n，1 And the decoder Decoder n，2 Both include: Bi-LSTM, attention layer and fully connected layer; among them, encoder Encoder n Processing the training sample X includes: inputting the training sample X into the Bi-LSTM, inputting the output of the Bi-LSTM into the attention layer, and inputting the output of the attention layer into the fully connected layer to obtain an output result; the Bi-LSTM is a bidirectional long short-term memory network.

[0020] Electric vehicle battery data includes: battery voltage, state of charge, current and temperature.

[0021] The test sample set includes battery data of M electric vehicles; calculating the thermal runaway diagnosis results based on the reconstruction results includes: calculating the reconstruction error of each basic model for the battery data of each electric vehicle based on the reconstruction results, constructing the reconstruction error vector of each basic model based on the reconstruction error, and constructing a reconstruction error matrix based on the reconstruction error vector of each basic model; calculating the thermal runaway diagnosis results based on the reconstruction error matrix; each column of the reconstruction error matrix is ​​a reconstruction error vector of the basic model, and the reconstruction error vector of the basic model is the reconstruction error of the basic model for the battery data of all electric vehicles.

[0022] Calculated thermal runaway diagnostic results include:

[0023] S31, sorting the reconstruction errors of each column in the reconstruction error matrix Q from large to small to obtain a reconstruction error matrix Q′;

[0024] S32, replacing the reconstruction error in the reconstruction error matrix Q′ with the corresponding electric vehicle to obtain the electric vehicle matrix V; setting a threshold according to the 3-point rule, marking the electric vehicles in the electric vehicle matrix V whose reconstruction error is greater than the threshold as potential thermal runaway electric vehicles, to obtain the potential thermal runaway electric vehicle matrix;

[0025] S33, counting the initial total number of votes in the first row of the potential thermal runaway electric vehicle matrix, and counting the number of times each electric vehicle in the first row is marked as a potential thermal runaway electric vehicle, and when the ratio of the number of times an electric vehicle is marked to the initial total number of votes is greater than 50%, it is determined to be a thermal runaway electric vehicle;

[0026] S34, counting the total valid votes in the i+1th row in the potential thermal runaway electric vehicle matrix, and counting the number of times each electric vehicle in the i+1th row is marked as a potential thermal runaway electric vehicle, and when the ratio of the number of times an electric vehicle is marked to the total valid votes is greater than 50%, it is determined to be a thermal runaway electric vehicle; wherein i is the number of iterations of step S34;

[0027] S35. Repeat step S34 until there is no electric vehicle marked as a potential thermal runaway electric vehicle in the potential thermal runaway electric vehicle matrix.

[0028] The total number of valid votes in the i+1th row in the potential thermal runaway electric vehicle matrix is ​​equal to the number of basic models minus the number of electric vehicles that have been determined to be thermal runaway electric vehicles.

[0029] Setting the threshold includes: calculating the reconstruction error matrix Q ′ The mean μ and standard deviation σ of the reconstruction error in , and the threshold μ+3σ is set according to the mean and standard deviation.

[0030] Beneficial effects:

[0031] 1. The adaptive integrated voting method of the present invention sorts the reconstruction errors of each column in the reconstruction error matrix Q from large to small, so that the first few rows of the sorted reconstruction error matrix are all likely to be judged as thermal runaway vehicles, and then the 3σ rule is used to mark potential thermal runaway electric vehicles in the sorted reconstruction error matrix; when judging the thermal runaway vehicles for each row in turn, the number of basic models minus the number of vehicles that have been judged as runaway vehicles is taken as the total valid votes, and the thermal runaway electric vehicle is judged according to the ratio of the number of times the electric vehicle is marked as a potential thermal runaway electric vehicle to the total valid votes, which can further reduce the impact of non-potential thermal runaway electric vehicles and vehicles that have been judged as thermal runaway vehicles on subsequent voting, enhance the sensitivity of the model to runaway vehicles, and thus improve the recall rate and accuracy of the model.

[0032] 2. The present invention adopts a basic model for mining electric vehicle battery data features based on Bi-LSTM and attention mechanism; Bi-LSTM can effectively extract the time series characteristics of electric vehicle battery voltage, temperature, SOC and other data during thermal runaway; the attention mechanism realizes the differentiated weight learning of each dimension of electric vehicle battery data at different times, further reducing the basic model's missed judgment and misjudgment of thermal runaway electric vehicle batteries;

[0033] 3. The present invention adopts a two-stage adversarial training algorithm for electric vehicle battery anomaly mining; the first stage is the reconstruction training stage, in which each basic model is trained separately with the goal of minimizing the reconstruction error, providing stable and reliable initial parameters for the second stage adversarial training; the second stage is the adversarial training stage, in which the model discrimination ability is gradually improved in the confrontation between forging high-quality electric vehicle battery data and discriminating real electric vehicle battery data to the maximum extent; BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A flowchart of a method for diagnosing thermal runaway of an electric vehicle battery based on unsupervised adversarial learning provided by an embodiment of the present invention;

[0035] Figure 2 A battery thermal runaway warning framework diagram based on unsupervised and integrated learning provided in an embodiment of the present invention;

[0036] Figure 3 A schematic diagram of a basic model based on Bi-LSTM and attention mechanism provided for an embodiment of the present invention;

[0037] Figure 4 A schematic diagram of a two-stage adversarial training method provided by an embodiment of the present invention;

[0038] Figure 5 A schematic diagram of the reconstruction error difference between the M6, M7 and M8 thermal runaway samples and normal samples provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] The reconstruction error can quantify the difference between the original data of the electric vehicle battery and the reconstructed data output by the model. The present invention uses normal electric vehicle battery data as a training set, so the trained model can reconstruct normal electric vehicle battery data well, and the reconstruction error is small. When thermal runaway data is input, the reconstruction error will increase significantly due to the inconsistency between the distribution of thermal runaway data and normal data; therefore, the size of the reconstruction error based on the electric vehicle battery is an important basis for judging whether the battery is thermally runaway.

[0041] like Figure 1 , Figure 2 As shown, the present invention adopts an electric vehicle battery thermal runaway diagnosis method based on unsupervised adversarial learning, comprising:

[0042] S1. Obtain an electric vehicle battery data set, preprocess the electric vehicle battery data set, and obtain a training sample set and a test sample set; each training sample in the training sample set is normal electric vehicle battery data;

[0043] S2. Construct an integrated model, and perform two-stage adversarial training on the integrated model using the training sample set to obtain a trained integrated model; the integrated model includes N basic models; the two-stage adversarial training includes: a first-stage reconstruction training and a second-stage adversarial training;

[0044] S3. Input the test sample set into the trained integrated model to obtain the reconstruction result of each basic model; and use the adaptive integrated voting method to obtain the thermal runaway diagnosis result based on the reconstruction result.

[0045] Electric vehicle battery data includes but is not limited to: battery voltage (V), state of charge (SOC), current (I), and temperature (T), that is, X = [V, SOC, I, T...].

[0046] Preprocessing the electric vehicle battery data set includes: performing maximum and minimum normalization on the electric vehicle battery data, and dividing the normalized data into a training sample set and a test sample set.

[0047] The training sample set includes N sub-training sample sets; Figure 4 As shown, each basic model includes: model ED1 and model ED2, ED1 includes: encoder Encoder and decoder Decoder1, ED2 includes encoder Encoder and decoder Decoder2; basic model n includes: model ED n，1 and Model ED n，2 , ED is the encoder-decoder model; ,ED n，1 Includes: Encoder n and Decoder n，1 , E.D. n，2 Including encoder Encoder n and Decoder n，2 ; The two-stage adversarial training process for each base model n includes:

[0048] S21, input the training samples X of the nth sub-training sample set into ED n，1 and ED n，2 Performing the first stage of reconstruction training to obtain a first reconstruction result and a second reconstruction result;

[0049] ED n，1 Processing the training sample X includes: inputting the training sample X into the encoder Encoder n, the encoder Encoder n The output of the decoder n，1 , and obtain the first reconstruction result;

[0050] The training objectives are:

[0051]

[0052] in, is the loss function, ED n，1 The neural network parameters are abbreviated as θ1, ED n，2 The neural network parameters are abbreviated as θ2. After the parameters are initialized, they are transferred and optimized between the two stages.

[0053] S22, input the first reconstruction result into ED n，2 Conduct the second stage of adversarial training to obtain generated fake samples;

[0054] The training objectives of this phase are:

[0055]

[0056] In the adversarial training phase, Decoder n，1 The goal of the generator is to minimize the relationship between the training sample X and the decoder. n，2 The difference between the outputs creates fake data that is as close to the real data as possible; Decoder n，2 The goal of the discriminator is to maximize this difference and compare the real data with the ED n，1 The reconstructed data are separated to improve the ability to distinguish complex scene noise between thermal runaway samples and normal samples with complex scene noise, and finally to expand the reconstruction error difference between thermal runaway samples and normal samples;

[0057] S23. Calculate the loss function value according to the training sample X, the first reconstruction result, the second reconstruction result and the generated result, update the model parameters according to the loss function value, and obtain the trained basic model n when the number of iterations reaches the preset maximum number of iterations.

[0058] It should be noted that the first and second stages are performed alternately, and the reconstruction stage and the adversarial stage gain each other in each iterative training. The reconstruction stage provides stability and reliability for the adversarial stage, and the adversarial stage further clarifies the boundary between normal data and thermal runaway data for the reconstruction stage. Therefore, the reconstruction stage and the adversarial stage are regarded as equally important for adversarial training of electric vehicle battery anomaly mining, and the coefficients are set to 1 respectively. Therefore, the total training goal of the two-stage training algorithm is:

[0059]

[0060] in, For ED n，1 The loss function is For ED n，2 The loss function of θ1 is ED n，1 The neural network parameters are abbreviated as , θ2 is ED n，2 The neural network parameters, It means that the calculation The minimum value of parameter θ1.

[0061] like Figure 3 As shown, the encoder n , Decoder n，1 And the decoder n，2 They all include: Bi-LSTM, attention layer and fully connected layer; among them, Bi-LSTM is composed of two layers of long short-term memory (LSTM), encoder Encoder n Processing the training sample X includes:

[0062] The training sample X is input into Bi-LSTM to extract the time series characteristics of each dimension of electric vehicle data during thermal runaway:

[0063] h t =LSTM(X, h t-1 )

[0064] h i =LSTM(X, h i+1 )

[0065] h′ t =a t h t +b t h i +c t

[0066] Among them, h t 、h t-1 are the outputs of the first layer of Bi-LSTM at time t and time t-1, respectively. i 、h i+1 are the outputs of the second layer of Bi-LSTM at time t and time t+1, respectively, h′ t is the output of the Bi-LSTM output layer, a t 、b t is the weight, c t is the bias, LSTM(·) represents the operation process of the LSTM network, and the LSTM unit updates and utilizes the multi-dimensional historical information of the electric vehicle battery through three gate structures.

[0067] Feed the output of the Bi-LSTM into the attention layer:

[0068] Bi-LSTM can effectively extract the time series correlation of electric vehicle batteries, but it does not effectively consider the different importance of various operating data of electric vehicle batteries at different times to thermal runaway warning. In fact, when calculating the cumulative reconstruction error of electric vehicle batteries in multiple time periods, there are differences in the relative importance of each input variable of the electric vehicle battery in the reconstruction error at the current moment. It is not reasonable to directly consider the reconstruction error calculation of the multi-dimensional time series of electric vehicle batteries as the accumulation of reconstruction errors for abnormal judgment of single-variable time series data. It is necessary to assign dynamic weights to the reconstruction errors of each dimension at the current moment as time changes. Therefore, the present invention adds an attention mechanism to the Bi-LSTM network, and assigns different weight coefficients to the feature vector after the attention mechanism layer. The weight coefficient calculation formula is:

[0069]

[0070] Among them, h′ t , t, h′ j，t , h′ k，t They represent h′ in the Bi-LSTM output layer at time t respectively. t The i-th, j-th and k-th elements of , l is the dimension of the Bi-LSTM output layer; the weight coefficients assigned by the electric vehicle battery attention mechanism are assigned to the Bi-LSTM output layer data to obtain h″ t , the allocation process is:

[0071] h″ t =A t h′ t

[0072] Among them, A t is the weight matrix of the attention mechanism layer, which consists of the weight coefficient element α i，j composition.

[0073] Input the output of the attention layer into the fully connected layer to get the output result:

[0074] X out =W fc ×h″ t +b fc

[0075] Among them, W fc is the weight, b fc For bias.

[0076] The test sample set includes battery data of M electric vehicles; calculating the thermal runaway diagnosis results based on the reconstruction results includes: calculating the reconstruction error of each basic model for the battery data of each electric vehicle based on the reconstruction results, constructing the reconstruction error vector of each basic model based on the reconstruction error, and constructing a reconstruction error matrix based on the reconstruction error vector of each basic model; calculating the thermal runaway diagnosis results based on the reconstruction error matrix; each column of the reconstruction error matrix is ​​a reconstruction error vector of the basic model, and the reconstruction error vector of the basic model is the reconstruction error of the basic model for the battery data of all electric vehicles.

[0077] Calculated thermal runaway diagnostic results include:

[0078] S31, sorting the reconstruction errors of each column in the reconstruction error matrix Q from large to small to obtain a reconstruction error matrix Q′;

[0079] S32, replacing the reconstruction errors in the reconstruction error matrix Q′ with the corresponding electric vehicles to obtain the electric vehicle matrix V; setting a threshold using the 3-point rule according to the reconstruction error matrix Q′, marking the electric vehicles in the electric vehicle matrix V whose reconstruction errors are greater than the threshold as potential thermal runaway electric vehicles, to obtain the potential thermal runaway electric vehicle matrix;

[0080] S33, counting the initial total number of votes in the first row of the potential thermal runaway electric vehicle matrix, and counting the number of times each electric vehicle in the first row is marked as a potential thermal runaway electric vehicle, and when the ratio of the number of times an electric vehicle is marked to the initial total number of votes is greater than 50%, it is determined to be a thermal runaway electric vehicle;

[0081] S34, counting the total valid votes in the i+1th row in the potential thermal runaway electric vehicle matrix, and counting the number of times each electric vehicle in the i+1th row is marked as a potential thermal runaway electric vehicle, and when the ratio of the number of times an electric vehicle is marked to the total valid votes is greater than 50%, it is determined to be a thermal runaway electric vehicle; wherein i is the number of iterations of step S34;

[0082] S35. Repeat step S34 until there is no electric vehicle marked as a potential thermal runaway vehicle in the potential thermal runaway electric vehicle matrix.

[0083] The total number of valid votes in the i+1th row in the potential thermal runaway electric vehicle matrix is ​​equal to the number of basic models minus the number of electric vehicles that have been determined to be thermal runaway electric vehicles.

[0084] Setting the threshold includes: calculating the reconstruction error matrix Q ′ The mean μ and standard deviation σ of the reconstruction error in , and the threshold μ+3σ is set according to the mean and standard deviation.

[0085] In one embodiment, the open source electric vehicle battery data of the National New Energy Vehicle Monitoring and Management Center (NMMC-NEV) is used to verify the experimental results. The NMMC-NEV public data set involves a total of 21 groups of batteries, each of which includes voltage, current, state of charge, temperature, and battery cell data. The time span is about half a year, and the sampling frequency is 0.1Hz. Among them, 7 groups of electric vehicle batteries caused combustion due to thermal runaway.

[0086] In one embodiment, the following methods (M1-M8) are compared, and the settings are shown in Table 1, wherein M8 is the method proposed by the present invention.

[0087] M1: supervised method; using a fully connected neural network, with 2 hidden layers, 32 neurons in each layer, a learning rate of 0.001, and an optimizer of Adam (Adaptive Method);

[0088] M2: unsupervised clustering method (K-means method);

[0089] M3: Unsupervised method based on reconstruction error; using fully connected neural network, other parameters are set the same as M1;

[0090] M4: Electric vehicle battery thermal runaway warning framework based on unsupervised and ensemble learning; based on M3, an ensemble learning framework is further used to form an integrated model;

[0091] M5: Basic model for electric vehicle battery data feature mining based on Bi-LSTM; based on M4, the neural network of the hidden layer is replaced by Bi-LSTM, the number of Bi-LSTM layers is 2, and other parameter settings are the same as M1.

[0092] M6: A basic model for mining electric vehicle battery data features based on Bi-LSTM and attention mechanism. Based on M5, an attention layer is added after the Bi-LSTM neural network, and the number of neurons is also 32.

[0093] M7: Unsupervised anomaly detection of multivariate time series (USAD method); use 3 fully connected neural network models (the number of neural network layers is the same as M6) to build an adversarial algorithm.

[0094] M8: The method proposed in the present invention; based on M6, three M6 models are used to build a two-stage adversarial algorithm, and other parameter settings are the same as M7.

[0095] Table 1 Comparison method settings and their purpose

[0096]

[0097] Because the focus of the anomaly detection task is to pay attention to abnormal samples, the following indicators are selected:

[0098]

[0099]

[0100]

[0101] Among them, TP represents normal samples identified as normal samples by the model, TN represents abnormal samples identified as abnormal samples by the model, FP represents abnormal samples identified as normal samples by the model, and FN represents normal samples identified as abnormal samples by the model. TNR is the true negative rate, also known as the negative sample recall rate; NPV is the negative prediction value, also known as the negative sample prediction accuracy; MCC is the Matthews correlation coefficient, which can provide more accurate performance evaluation in unbalanced data sets. The larger the values ​​of the above indicators, the better the model classification performance.

[0102] 1. Verification of the effectiveness of unsupervised algorithms based on reconstruction error

[0103] In order to compare the effects of the basic model of electric vehicle battery thermal runaway warning based on unsupervised learning and the unsupervised clustering method based on distance calculation, M1, M2 and M3 are compared. The following example is set: M1 and M3 randomly select 7 groups of normal electric vehicle battery samples from 21 groups of NMMC-NEV public battery data sets as training sets, and the test set data is the remaining 14 groups of electric vehicle battery data (including 7 groups of thermal runaway batteries); M2 directly performs cluster analysis on the test set without training, and the test set data value is consistent with M1 and M3. M1, M2 and M3 are tested 3 times respectively, and the average is taken as the result. Different normal batteries are randomly selected as training sets for the 3 tests, and the remaining batteries are used as test sets. The NMMC-NEV public data set battery thermal runaway prediction results of M1, M2 and M3 are shown in Table 2;

[0104] Table 2 Comparison of thermal runaway fault diagnosis results of M1, M2 and M1 on the NMMC-NEV public dataset:

[0105]

[0106] Among them, the negative sample recall rate of the M2 method is 7% higher than that of the M1 method, the negative sample prediction accuracy is 56% higher than that of the M1 method, and the Matthews correlation coefficient is 27% higher than that of the M1 method, which can effectively illustrate the superiority of unsupervised methods in the field of battery thermal runaway detection over supervised methods in dealing with unbalanced data and complex and variable anomalies. The negative sample prediction accuracy of the M3 method is 29% higher than that of the M2 method, and the Matthews correlation coefficient is 14% higher than that of the M2 method, which can illustrate that the unsupervised thermal runaway anomaly detection method based on reconstruction error can distinguish the slight differences between electric vehicle batteries and normal battery data before and after the runaway, reflecting the superiority of unsupervised clustering methods over simple distance calculations. The above results verify the effectiveness of the unsupervised algorithm based on reconstruction error.

[0107] 2. Stability verification of thermal runaway warning integrated model

[0108] In order to illustrate the stability performance of the integrated model, the present invention compares the early warning indicators of M and M4. Set the following example: The training set and test set of M3 are set as above, and the training set and test set of M4 are the same as M3. The difference is that M4 uses each group of normal battery data to train a basic model. The NMMC-NEV public data set includes 7 basic models. M3 and M4 are tested 3 times respectively and displayed one by one to verify the stability of the M4 integrated model. Different normal batteries are randomly selected as training sets in the 3 tests, and the remaining batteries are used as test sets.

[0109] According to the NMMC-NEV public data set battery thermal runaway prediction results of M3 and M4, as shown in Table 3, the evaluation index of M3 fluctuated greatly in the three tests, among which the first and third evaluation accuracy were low, while the second evaluation accuracy was high. In contrast, the evaluation index of the integrated model M4 was high in the three tests, and no fluctuations were detected. The above results show that the electric vehicle battery thermal runaway warning integrated model has higher stability compared with the basic model.

[0110] Table 3 Comparison of thermal runaway fault diagnosis results between M3 and M4 on the NMMC-NEV public dataset

[0111]

[0112] 3. Validation of the basic model based on Bi-LSTM and attention mechanism

[0113] The warning accuracy of the integrated framework depends on the feature mining ability of the basic model. In order to effectively improve the feature mining ability of the basic model, the present invention proposes an electric vehicle battery data feature mining method based on Bi-LSTM and attention mechanism, and uses it as the basic model. In order to illustrate the effect of this method on improving the accuracy of the model, the following example is set: the training set and test set settings of M4 are the same as the previous section, and the training set and test set of M5 and M6 are the same as M4. M4, M5 and M6 were tested 3 times respectively, and the average was taken as the result. Different normal batteries were randomly selected as training sets in the 3 tests, and the remaining batteries were used as test sets. The prediction results of battery thermal runaway of the NMMC-NEV public data set for M4, M5 and M6 are shown in Table 5;

[0114] Table 5 Comparison of thermal runaway fault diagnosis results of M4, M5 and M6 on the NMMC-NEV public dataset

[0115]

[0116] As shown in Table 5, since M6 takes into account the temporal characteristics of the data, compared with M4, the negative sample recall rate and the comprehensive evaluation index Matthews correlation coefficient are increased by 3% and 17% respectively. The Bi-LSTM network set by M5 and the attention mechanism set by M6 effectively improve the model accuracy and verify the effectiveness of the basic model for mining electric vehicle battery data features based on Bi-LSTM and attention mechanism.

[0117] 4. Verification of the effectiveness of a two-stage training algorithm for electric vehicle battery anomaly mining based on adversarial training

[0118] The two-stage training algorithm for electric vehicle battery anomaly mining based on adversarial training aims to improve the ability to distinguish complex scene noise between thermal runaway samples and normal samples with complex scene noise, improve the reconstruction error difference between thermal runaway samples and normal samples, and finally improve the model accuracy. In order to illustrate the effect of this method on improving model accuracy, the following example is set: the training set and test set of M6 are set as above, and the training set and test set of M7 and M8 are the same as M6. M6, M7, and M8 are tested 3 times respectively, and the average is taken as the result. Different normal batteries are randomly selected as training sets for the 3 tests, and the remaining batteries are used as test sets.

[0119] The reconstruction error differences between the thermal runaway samples of M6, M7 and M8 and the normal samples are shown in Figure 5 As shown in the figure, compared with M6 and M7, the reconstruction error difference between the thermal runaway samples and normal samples of M8 has been significantly improved: in M6 and M7, the reconstruction error value of the thermal runaway samples is about 2-2.5 times that of the normal samples; after two-stage electric vehicle battery anomaly mining adversarial training, in M8, the reconstruction error value of the thermal runaway samples is about 6-8 times that of the normal samples, and the model discrimination ability has been significantly improved.

[0120] The prediction results of battery thermal runaway of NMMC-NEV public dataset for M6, M7 and M8 are shown in Table 6, which proves the effectiveness of the two-stage training algorithm for electric vehicle battery anomaly mining based on adversarial training.

[0121] Table 6 Comparison of thermal runaway fault diagnosis results between M6 and M7 on the NMMC-NEV public dataset

[0122]

[0123] The negative sample recall rate of the adversarial training method of M7 and M8 is 11% higher than that of the M6 ​​method, reaching 100%, which means that all thermal runaway vehicles can be fully detected without missed detection. The negative sample prediction accuracy of M8 is 17% higher than that of M7, which means that the robustness is improved and thermal runaway vehicles can be selected more accurately. The Matthews correlation coefficient of M8 is 6% higher than that of M6 and 11% higher than that of M7, which can illustrate the effectiveness of the two-stage training algorithm for electric vehicle battery anomaly mining based on adversarial training in the field of battery thermal runaway detection.

[0124] In summary, the unsupervised learning method based on reconstruction error and ensemble learning can solve the problem of scarce thermal runaway battery data and improve the stability and credibility of the diagnostic results; the two-stage adversarial training algorithm can expand the difference in reconstruction error between thermal runaway battery data and normal battery data, and improve the robustness and accuracy of the model; the adaptive ensemble voting method uses the total valid votes and the 3σ rule for ensemble voting, which improves the recall and accuracy of the model.

[0125] The above embodiments further illustrate the purpose, technical solutions and advantages of the present invention in detail. It should be understood that the above embodiments are only preferred implementation modes of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for diagnosing thermal runaway of electric vehicle batteries based on unsupervised adversarial learning, characterized in that: include: S1. Obtain an electric vehicle battery data set, preprocess the electric vehicle battery data set, and obtain a training sample set and a test sample set; Each training sample in the training sample set is normal electric vehicle battery data; S2. Construct an integrated model, and use the training sample set to perform two-stage adversarial training on the integrated model to obtain a trained integrated model; the integrated model includes N basic models; S3. Input the test sample set into the trained integrated model to obtain the reconstruction result of each basic model; and use the adaptive integrated voting method to obtain the thermal runaway diagnosis result based on the reconstruction result.

2. According to claim 1, a method for diagnosing thermal runaway of electric vehicle batteries based on unsupervised adversarial learning is characterized in that: The training sample set includes N sub-training sample sets; the basic model n includes: model ED n,1 and Model ED n,2 , ED is the encoder-decoder model; the two-stage adversarial training process for each basic model n includes: S21, input the training samples X of the nth sub-training sample set into ED n,1 and ED n,2 Performing reconstruction training to obtain a first reconstruction result and a second reconstruction result; S22, input the first reconstruction result into ED n,2 Conduct adversarial training to obtain generated results; S23. Calculate the loss function value according to the sample X, the first reconstruction result, the second reconstruction result and the generated result, update the model parameters according to the loss function value, and obtain the trained basic model n when the number of iterations reaches the preset maximum number of iterations.

3. The electric vehicle battery thermal runaway diagnosis method based on unsupervised adversarial learning according to claim 2, characterized in that: The loss function is: in, For ED n,1 The loss function is For ED n,2 The loss function of .

4. The electric vehicle battery thermal runaway diagnosis method based on unsupervised adversarial learning according to claim 2, characterized in that: ED n,1 Includes: Encoder n and Decoder n,1 , E.D. n,2 Including encoder Encoder n and Decoder n,2 ;ED n,1 Processing the training sample X includes: inputting the training sample X into the encoder Encoder n , the encoder Encoder n The output of the decoder n,1 , and obtain the first reconstruction result.

5. The electric vehicle battery thermal runaway diagnosis method based on unsupervised adversarial learning according to claim 4, characterized in that: Encoder n , Decoder n,1 And the decoder Decoder n,2 Both include: Bi-LSTM, attention layer and fully connected layer; among them, encoder Encoder n Processing the training sample X includes: inputting the training sample X into the Bi-LSTM, inputting the output of the Bi-LSTM into the attention layer, and inputting the output of the attention layer into the fully connected layer to obtain an output result; the Bi-LSTM is a bidirectional long short-term memory network.

6. The electric vehicle battery thermal runaway diagnosis method based on unsupervised adversarial learning according to claim 1, characterized in that: Electric vehicle battery data includes: battery voltage, battery state of charge, battery current and battery temperature.

7. The electric vehicle battery thermal runaway diagnosis method based on unsupervised adversarial learning according to claim 1, characterized in that: The test sample set includes battery data of M electric vehicles; calculating the thermal runaway diagnosis results based on the reconstruction results includes: calculating the reconstruction error of each basic model for the battery data of each electric vehicle based on the reconstruction results, constructing a reconstruction error vector of each basic model based on the reconstruction error, and constructing a reconstruction error matrix based on the reconstruction error vector of each basic model; calculating the thermal runaway diagnosis results based on the reconstruction error matrix; each column of the reconstruction error matrix is ​​a reconstruction error vector of a basic model, and the reconstruction error vector of a basic model includes the reconstruction error of the basic model for the battery data of M electric vehicles.

8. The electric vehicle battery thermal runaway diagnosis method based on unsupervised adversarial learning according to claim 7, characterized in that: Calculated thermal runaway diagnostic results include: S31, sort the reconstruction errors of each column in the reconstruction error matrix Q from large to small, and obtain the reconstruction error matrix Q ′ ; S32, reconstruct the error matrix Q ′ The reconstruction error in is replaced by the corresponding electric vehicle to obtain the electric vehicle matrix V; a threshold is set according to the 3σ rule, and the electric vehicles with reconstruction errors greater than the threshold in the electric vehicle matrix V are marked as potential thermal runaway electric vehicles to obtain the potential thermal runaway electric vehicle matrix; S33, counting the initial total number of votes in the first row of the potential thermal runaway electric vehicle matrix, and counting the number of times each electric vehicle in the first row is marked as a potential thermal runaway electric vehicle, and when the ratio of the number of times an electric vehicle is marked to the initial total number of votes is greater than 50%, it is determined to be a thermal runaway electric vehicle; S34, counting the total valid votes in the i+1th row in the potential thermal runaway electric vehicle matrix, and counting the number of times each electric vehicle in the i+1th row is marked as a potential thermal runaway electric vehicle, and when the ratio of the number of times an electric vehicle is marked to the total valid votes is greater than 50%, it is determined to be a thermal runaway electric vehicle; wherein i is the number of iterations of step S34; S35. Repeat step S34 until there is no electric vehicle marked as a potential thermal runaway electric vehicle in the potential thermal runaway electric vehicle matrix.

9. The electric vehicle battery thermal runaway diagnosis method based on unsupervised adversarial learning according to claim 8, characterized in that: The total number of valid votes in the i+1th row in the potential thermal runaway electric vehicle matrix is ​​equal to the number of basic models minus the number of electric vehicles that have been determined to be thermal runaway electric vehicles.

10. The electric vehicle battery thermal runaway diagnosis method based on unsupervised adversarial learning according to claim 8, characterized in that: Setting the threshold includes: calculating the reconstruction error matrix Q ′ The mean μ and standard deviation σ of the reconstruction error in , and the threshold λ+3σ is set according to the mean and standard deviation.

Images