Fault diagnosis method and device for battery pack under random working condition

Abstract

The invention relates to a fault diagnosis method and device for a random working condition battery pack. The method comprises the steps of performing residual calculation and feature extraction based on an obtained state parameter prediction matrix and an actual observation parameter matrix of a single battery to obtain a feature matrix of the single battery; carrying out dimension reduction processing on the characteristic matrix of the battery monomer by utilizing a principal component analysis method to obtain a comprehensive evaluation index of the battery monomer; comparing the comprehensive evaluation index of each battery monomer with a preset threshold value, and determining the battery monomer of which the comprehensive evaluation index exceeds the preset threshold value as a fault battery monomer; according to the feature matrix corresponding to the faulty single battery and the initial state matrix of the single battery, constructing a fault type matrix of the faulty single battery; and inputting the fault type matrix of the fault single battery into the trained fault type probability prediction model to obtain a trigger probability corresponding to the fault type of the fault single battery so as to carry out fault diagnosis on the battery pack. According to the scheme, the fault type of the battery pack can be accurately diagnosed.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of power battery safety monitoring, and in particular to a fault diagnosis method and device for a random operating condition battery pack. Background Art

[0002] The power battery system composed of lithium-ion battery packs has the advantages of high specific energy, long life, low self-discharge and wide applicable temperature range, and is currently the mainstream choice for electric vehicles. As the market share of electric vehicles increases, the safety failure of battery modules has become a major bottleneck restricting the development of electric vehicles. In recent years, vehicle battery failures have occurred frequently, and the number of market recalls is countless, which has increased users' safety anxiety and the economic burden of the market.

[0003] Research on lithium-ion battery fault diagnosis often analyzes the monitoring object for specific fault types and then designs corresponding diagnostic methods. In previous research on lithium-ion battery fault diagnosis, the main fault types detected were external short circuit, internal short circuit, electrolyte leakage, abnormal aging and thermal runaway. The monitoring objects mainly include voltage, temperature, current, gas production, and battery appearance images. Due to the limitations on the types of macroscopic characterization data that can be obtained, different degradation phenomena may have similar macroscopic characterizations, the most typical of which are temperature rise and voltage drop, making it impossible to accurately diagnose the fault type of the battery pack.

[0004] Therefore, there is an urgent need to provide a fault diagnosis method and device for a random operating condition battery pack. Summary of the invention

[0005] In order to solve the problem that the traditional fault diagnosis method cannot accurately diagnose the fault type of the battery pack, the embodiment of the present invention provides a fault diagnosis method and device for a battery pack under random working conditions.

[0006] In a first aspect, an embodiment of the present invention provides a fault diagnosis method for a random working condition battery pack, the method comprising:

[0007] Perform residual calculation based on the obtained state parameter prediction matrix of the battery cell and the actual observation parameter matrix to obtain a residual matrix of the battery cell; wherein the state parameter prediction matrix of the battery cell is obtained by prediction using a pre-trained deep learning network;

[0008] Performing feature calculation on the residual matrix of the battery cell to obtain a feature matrix of the battery cell;

[0009] Using principal component analysis to reduce the dimension of the characteristic matrix of the battery cell to obtain a comprehensive evaluation index of the battery cell;

[0010] Comparing the comprehensive evaluation index of each battery cell with a preset threshold, and determining the battery cell whose comprehensive evaluation index exceeds the preset threshold as a faulty battery cell;

[0011] Constructing a fault type matrix of the faulty battery cell according to the characteristic matrix corresponding to the faulty battery cell and the initial state matrix of the battery cell;

[0012] The fault type matrix of the faulty battery cell is input into a trained fault type probability prediction model to obtain the trigger probability corresponding to the fault type of the faulty battery cell, so as to perform fault diagnosis on the battery pack.

[0013] In a second aspect, an embodiment of the present invention further provides a fault diagnosis device for a random working condition battery pack, the device comprising:

[0014] A first calculation unit is used to perform residual calculation based on the obtained state parameter prediction matrix of the battery cell and the actual observation parameter matrix to obtain a residual matrix of the battery cell; wherein the state parameter prediction matrix of the battery cell is obtained by prediction using a pre-trained deep learning network;

[0015] A second calculation unit, configured to perform characteristic calculation on the residual matrix of the battery cell to obtain a characteristic matrix of the battery cell;

[0016] a third calculation unit, configured to perform dimensionality reduction processing on the characteristic matrix of the battery cell by using a principal component analysis method to obtain a comprehensive evaluation index of the battery cell;

[0017] A fault determination unit, used to compare the comprehensive evaluation index of each battery cell with a preset threshold value, and determine the battery cell whose comprehensive evaluation index exceeds the preset threshold value as a faulty battery cell;

[0018] A fault type matrix construction unit, used to construct a fault type matrix of the faulty battery cell according to a characteristic matrix corresponding to the faulty battery cell and an initial state matrix of the battery cell;

[0019] The fault diagnosis unit is used to input the fault type matrix of the faulty battery cell into the trained fault type probability prediction model to obtain the trigger probability corresponding to the fault type of the faulty battery cell, so as to perform fault diagnosis on the battery pack.

[0020] In a third aspect, an embodiment of the present invention further provides a computing device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method described in any embodiment of this specification is implemented.

[0021] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, enables the computer to execute the method described in any embodiment of this specification.

[0022] In a fifth aspect, an embodiment of the present invention further provides a computer program product, including a computer program, which implements the steps of the method described above when executed by a processor.

[0023] An embodiment of the present invention provides a fault diagnosis method and device for a battery pack under random working conditions. After a state parameter prediction matrix of a battery cell is obtained by using a pre-trained deep learning network, residual and feature extraction is first performed on it and the actual observed parameter matrix, thereby obtaining a feature matrix composed of several parameters affecting the fault state of the battery pack. Then, the principal component analysis method is used to further screen several influencing parameters in the feature matrix, and the influence of each screened influencing parameter on the battery cell is further quantified, thereby accurately locating the faulty battery cell according to the quantitative index. Finally, the probability corresponding to the fault type of the faulty battery cell is predicted based on the trained fault type probability prediction model. In this way, the embodiment of the present invention combines a deep learning network with massive cloud data for training, which can realize the probability prediction of the fault type of a battery pack under random working conditions, thereby facilitating the accurate diagnosis of the fault type of a battery pack under random working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0025] Figure 1 is a flow chart of a fault diagnosis method for a random working condition battery pack provided by an embodiment of the present invention;

[0026] Figure 2 is a hardware architecture diagram of a computing device provided by an embodiment of the present invention;

[0027] Figure 3 It is a structural diagram of a fault diagnosis device for a random operating condition battery pack provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0029] The specific implementation of the above concept is described below.

[0030] Please refer to Figure 1 The embodiment of the present invention provides a fault diagnosis method for a random working condition battery pack, the method comprising:

[0031] Step 100, performing residual calculation based on the obtained state parameter prediction matrix of the battery cell and the actual observation parameter matrix to obtain the residual matrix of the battery cell; wherein the state parameter prediction matrix of the battery cell is obtained by prediction using a pre-trained deep learning network;

[0032] Step 102, performing feature calculation on the residual matrix of the battery cell to obtain a feature matrix of the battery cell;

[0033] Step 104, using principal component analysis to perform dimension reduction processing on the characteristic matrix of the battery cell to obtain a comprehensive evaluation index of the battery cell;

[0034] Step 106, comparing the comprehensive evaluation index of each battery cell with a preset threshold, and determining the battery cell whose comprehensive evaluation index exceeds the preset threshold as a faulty battery cell;

[0035] Step 108, constructing a fault type matrix of the faulty battery cell according to the characteristic matrix corresponding to the faulty battery cell and the initial state matrix of the battery cell;

[0036] Step 110 , inputting the fault type matrix of the faulty battery cell into the trained fault type probability prediction network model to obtain the trigger probability corresponding to the fault type of the faulty battery cell, so as to perform fault diagnosis on the battery pack.

[0037] The embodiment of the present invention provides that after using a pre-trained deep learning network to predict the state parameter prediction matrix of the battery cell, firstly, residual and feature extraction are performed on it and the actual observation parameter matrix, so as to obtain a feature matrix composed of several parameters affecting the fault state of the battery pack, and then the principal component analysis method is used to further screen several influencing parameters in the feature matrix, and the influence of each screened influencing parameter on the battery cell is further quantified, so as to accurately locate the faulty battery cell according to the quantitative index, and finally, the probability corresponding to the fault type of the faulty battery cell is predicted based on the trained fault type probability prediction model; in this way, the embodiment of the present invention combines the deep learning network and massive cloud data for training, so as to realize the probability prediction of the fault type of the battery pack under random working conditions, which is conducive to the accurate diagnosis of the fault type of the battery pack under random working conditions.

[0038] For step 100:

[0039] In the embodiment of the present invention, the state prediction matrix and the actual observation parameter matrix of the battery cell are both composed of the voltage value and the state of charge value of the battery cell in a sampling period.

[0040] In some embodiments, in step 100, the deep learning network includes a filter, a temporal network prediction model, and a physical enhancement encoder connected in sequence;

[0041] The state parameter prediction matrix of the battery cell is predicted in the following way:

[0042] For each current moment, execute:

[0043] Step S1, constructing a battery pack state model and a battery cell deviation model according to the voltage parameters of the battery cell obtained at the current moment;

[0044] Step S2, using filters to perform state prediction on the battery pack state model and the battery cell deviation model at the current moment, respectively, to obtain filter prediction results of the battery pack at the next moment and filter prediction results of the battery cell at the next moment, respectively; wherein the filter prediction results include the voltage and charge state of the battery pack and the battery cell at the next moment;

[0045] Step S3, inputting the battery state parameters and vehicle state parameters obtained at the current moment into the trained time series network prediction model to obtain the time series network prediction result of the battery pack at the next moment; wherein, during the training process of the time series network prediction model, the filter prediction result of the battery pack is used as a loss function to reversely adjust the time series network prediction result;

[0046] Step S4, inputting the timing network prediction result of the battery pack and the filter prediction result of the battery cell into the trained physical enhancement encoder to obtain the state parameter prediction matrix of the battery cell.

[0047] In an embodiment of the present invention, a battery pack state model and a battery cell deviation model are first constructed based on the voltage parameters of the battery cells obtained from a cloud data platform, and then filters are used to predict the states of the battery pack and the battery cells respectively, so as to obtain prediction results with better stability; a pre-trained timing network prediction model is then used to perform a secondary prediction on the state of the battery pack, and the prediction results of the timing network model are reversely adjusted using the prediction results of the filter in the trained timing network prediction model, so as to enhance the accuracy of the prediction results; finally, a trained physical enhancement encoder is used to further compress and decompress the prediction results of the filter and the prediction results of the timing network model, so as to obtain a battery cell prediction result with both good stability and accuracy.

[0048] In this way, in the embodiment of the present invention, filters, timing network prediction models and physical enhanced encoders are used in sequence to predict the state of the battery pack, and massive cloud data is used as a training sample set, so that the prediction results take into account both robustness and accuracy, which is conducive to the accurate diagnosis of the fault state of the battery pack under random working conditions.

[0049] For step S1:

[0050] In some embodiments, before constructing the battery pack state model and the battery cell deviation model, the method further includes:

[0051] Fitting the inverse Gaussian distribution as the probability distribution of the battery cell parameters to calculate the voltage distribution probability of the battery cell;

[0052] According to the voltage distribution probability of the battery cell, the battery pack voltage and the battery cell deviation voltage at the current moment are obtained;

[0053] A battery pack state model and a battery cell deviation model are constructed based on the battery pack voltage and the battery cell deviation voltage.

[0054] In the embodiment of the present invention, due to the different production batches of battery cells or the mutual influence between local cells in the circuit, the distribution of battery cell voltage is often skewed. Therefore, in the embodiment of the present invention, the inverse Gaussian distribution is selected as the probability distribution fitting of the battery cell voltage parameters, so that a detailed description of the distribution characteristics of the battery cells can be achieved without increasing the parameters to be solved.

[0055] In some preferred embodiments, the voltage distribution probability of the battery cell is calculated by the following formula:

[0056] f(U i )=[λ / (2πU i 3 )] 1 / 2 *exp[-λ(U i -μ) 2 / (2μ 2 U i )]

[0057] Where μ = U m ,λ=μ 3 / Var(U i ), U i is the voltage of the battery cell, U m is the average voltage of the battery cell;

[0058] The voltage of the battery pack is corrected based on the obtained voltage distribution probability of the battery cell, thereby obtaining the battery pack voltage and the battery cell deviation voltage. In some preferred embodiments, the battery pack voltage and the battery cell deviation voltage are respectively calculated by the following formulas:

[0059] U T，p =(f(U i )*U i ) / ∑f(U i )

[0060] ΔU i =U i -U T，p

[0061] Where U T，p The battery pack voltage, f(U i ) is the voltage distribution probability of the battery cell, ΔU i is the battery cell deviation voltage.

[0062] In some embodiments, the battery pack state model and the battery cell deviation state model are respectively as follows:

[0063]

[0064] ΔU i =ΔU OCV,i (ΔSOC i )-ΔR i I

[0065] Where U T,p is the terminal voltage of the battery pack state model, R P,p is the polarization internal resistance of the battery pack, C P,p is the polarization capacitance of the battery pack, U P,p is the polarization voltage of the battery pack, R0,p is the ohmic resistance of the battery pack, I is the instantaneous current, U OCV，p is the open circuit voltage of the battery pack, ΔU i is the terminal voltage of the battery cell deviation state model, ΔU OCV，i is the open circuit voltage of the battery cell, ΔSOC i is the state of charge of the battery cell, ΔR i is the ohmic internal resistance of the battery cell.

[0066] Since the working circuit of the battery pack is extremely complex and there are many factors affecting the battery pack state during actual vehicle operation, in this embodiment, the battery pack voltage and battery cell deviation obtained after distribution optimization processing are used as model output parameters to construct the above-mentioned battery pack and battery cell deviation state model. This model can be used as an equivalent model of a complex battery circuit to dynamically describe the dynamic process of the battery pack system. The filter is further combined with the above-mentioned state model to achieve online closed-loop prediction of the battery pack and battery cell state.

[0067] For step S2:

[0068] In some preferred embodiments, step S2 comprises:

[0069] Using the least square method to identify the coefficients of the battery pack state model and the battery cell deviation model at the current moment, to obtain the state parameters of the battery pack state model and the battery cell deviation model at the current moment;

[0070] The state parameters of the battery pack state model and the state parameters of the battery cell deviation model at the current moment are respectively input into the filter to obtain the first prediction result of the battery pack at the next moment and the first prediction result of the battery cell at the next moment respectively.

[0071] In the embodiment of the present invention, it can be seen from the battery pack state model and the battery cell deviation state model that the state parameters of the battery pack state model include the polarization internal resistance of the battery pack, the polarization capacitance of the battery pack and the ohmic resistance of the battery pack, and the state parameters of the battery cell deviation model include the ohmic internal resistance of the battery cell; in order to facilitate the identification of the state parameters of the battery pack state model and the state parameters of the battery cell deviation model, it is necessary to preprocess the battery pack state model and the battery cell deviation model.

[0072] For step S3:

[0073] The state prediction method of the battery pack and the battery cell using the filter combined with the state model in step S2 has good stability, but the accuracy of the prediction results is relatively poor, and in the state prediction results of the battery pack and the battery cell obtained in step S3, the state of the battery pack has a greater impact on the entire battery pack. In an embodiment of the present invention, the trained timing network model is used to further predict the battery pack state prediction results output in the filter, which can enhance the accuracy of the battery pack system prediction, so that the prediction results take into account both robustness and accuracy.

[0074] In step S3, the time series network prediction model is trained in the following way:

[0075] Build a preset time series network learning model;

[0076] The hyperparameters and loss function of the temporal network learning model are set; wherein the loss function is as follows:

[0077]

[0078] In the formula, loss1 is the loss function, t is the sampling time, is the prediction result of the time series network learning model, is the first prediction result of the battery pack, U T，p，k+1 is the terminal voltage of the battery pack at time k+1, x T，p，k+1 is the state of charge of the battery pack at time k+1;

[0079] The known vehicle state parameters and battery state parameters are used as input sample sets, and the known battery pack voltage and state of charge parameters are used as output sample sets, and the Adam algorithm is used to train the time series network learning model; wherein the vehicle state parameters include vehicle speed, vehicle mileage and vehicle operating status, and the battery state parameters include battery pack total voltage, current, battery pack average temperature and battery pack state of charge;

[0080] During the training process, the loss and accuracy of the time series network learning model are calculated, and the network parameters of the time series network learning model are iterated until the model converges or reaches a preset number of iterations to obtain the time series network prediction model.

[0081] For step S4:

[0082] In some preferred embodiments, the physical enhancement encoder in step S4 includes an encoding module, a latent space sampling module and a decoding module connected in sequence; wherein:

[0083] The encoding module is used to perform feature fusion and dimensionality reduction operations on the input first prediction result of the battery cell and the second prediction result of the battery pack to obtain latent variables;

[0084] The latent space sampling module is used to sum the latent variables output by the encoding module and the first prediction results of the battery cells to obtain sampling points corresponding to the number of battery cells;

[0085] The decoding module is used to map the sampling points corresponding to the number of battery cells output by the latent space sampling module to obtain a second prediction matrix of the battery cells.

[0086] In an embodiment of the present invention, a physical enhancement encoder is first constructed and trained, so that the physical enhancement encoder can be used to compress and decompress the second prediction result of the battery pack and the first prediction result of the battery cell, and realize the prediction of the battery cell state, thereby realizing accurate identification of battery pack system faults using the battery cell state prediction result.

[0087] During the training of the physical enhancement encoder, the MSE function is used as the loss function to reversely adjust the prediction result of the physical enhancement encoder; wherein the loss function is:

[0088]

[0089] Where loss2 is the MSE function, is the second prediction result of the battery cell output by the physical enhancement encoder, k is the sampling time, is the actual observation matrix of the battery cell, x n，k is the state of charge of the nth battery cell, U n，k is the terminal voltage of the nth battery cell.

[0090] Regarding step 102:

[0091] In some embodiments, the characteristic matrix of the battery cell is as follows:

[0092] X=[ewu,ewx,zu1(k),zx1(k),zu2(k),zx2(k)],

[0093]

[0094] In the formula, zu1(k) and zx1(k) are the maximum outlier coefficients of the corresponding variables, zu2(k) and zx2(k) are the deviations between the maximum and the second largest outlier coefficients, k is the sampling time, is the voltage value of the first row in the battery cell residual matrix, is the average voltage value of the first row in the battery cell residual matrix, is the state of charge value of the second row in the battery cell residual matrix, is the standard deviation of the state of charge value in the second row of the battery cell residual matrix, is the standard deviation of the voltage value in the first row of the battery cell residual matrix, is the average value of the state of charge value in the second row of the battery cell residual matrix, is the maximum voltage value in the first row of the residual matrix of the battery cell, is the maximum value of the state of charge value in the second row of the residual matrix of the battery cell, is the second largest voltage value in the first row of the residual matrix of the battery cell, is the second largest value of the state of charge value in the second row of the residual matrix of the battery cell, ewu is the exponentially weighted moving average filtered signal of the maximum voltage value in the first row of the residual matrix, and ewx is the exponentially weighted moving average filtered signal of the maximum state of charge value in the second row of the residual matrix.

[0095] In order to identify the fault state of the battery pack, most of the related technologies compare the residual value of the calculated state parameter prediction matrix of the battery cell with a manually set threshold value. However, in the actual working process of the battery pack, due to the relatively random working conditions of the battery pack and the instability of the state of each cell, the method of simply comparing the size of the residual value with the preset threshold value cannot accurately reflect the fault state of the battery pack. Therefore, in the embodiment of the present invention, after obtaining the residual matrix of the battery cell, the residual matrix is ​​further feature extracted and calculated to obtain a six-dimensional feature matrix composed of parameter indicators closely related to the fault state of the battery pack. This is not only conducive to improving the accuracy of the diagnosis of the fault state of the battery pack under random working conditions, but also can achieve accurate identification of the fault type of the battery pack.

[0096] Regarding step 104 and step 106:

[0097] In the embodiment of the present invention, the principal component analysis method is used to perform dimensionality reduction screening processing on the feature matrix of the battery cell, so as to extract the main characteristic parameters that affect the fault state of the battery pack from the feature matrix, reduce the influence of redundant information, obtain a low-dimensional feature matrix, and further design a comprehensive evaluation index to quantify the low-dimensional feature matrix. Finally, based on the quantified index and the preset threshold, the accurate diagnosis of the fault state of the battery pack can be achieved quickly and efficiently.

[0098] In some implementations, step 104 includes:

[0099] Using principal component analysis to reduce the dimension of the characteristic matrix of the battery cell to obtain a low-dimensional characteristic matrix;

[0100] Obtaining principal component space statistics and residual space statistics according to the characteristic matrix of the battery cell and the low-dimensional characteristic matrix;

[0101] Based on the principal component space statistics and the residual space statistics, a comprehensive evaluation index of the battery cell is obtained.

[0102] In the embodiment of the present invention, the method of reducing the dimension of the characteristic matrix of the battery cell by using principal component analysis is as follows:

[0103] First, the characteristic matrix of the battery cell is centralized. Each characteristic element in the characteristic matrix is ​​centralized using the following formula:

[0104]

[0105] Where, X ij is the characteristic element after centralization, x ij is the characteristic element of the i-th row and j-th column in the characteristic matrix of the battery cell, is the average value of the characteristic elements in the jth column, s j is the standard deviation of the characteristic elements in the jth column;

[0106] Then, calculate the covariance matrix of the feature matrix X1 after centering:

[0107]

[0108] According to |R-λI n |=0Calculate the eigenvalues ​​and eigenvectors of the covariance matrix, and calculate the eigenvalue λ=(λ1,λ2,…,λ n ), the corresponding eigenvector matrix P = (p1, p2, ..., p n ), thereby determining the principal component vector;

[0109] Furthermore, the feature matrix X1 after centralization is decomposed as follows:

[0110]

[0111] Where T is the score matrix of feature matrix X1, t k =X1P k , which is essentially the projection of the characteristic matrix X1 along the direction of the load vector. Each column vector t of T k represents the kth principal component, The space where E is located is called the principal component subspace, k is the number of principal components, E is the residual matrix in the principal component space, which contains the information that is not explained in the principal component model, and the space where E is located is called the residual subspace;

[0112] The number of principal components k is determined according to the cumulative contribution rate CPV of the principal components:

[0113]

[0114] For example, k principal components corresponding to eigenvalues ​​with cumulative contribution rates exceeding 80% may be taken to obtain the number of principal components.

[0115] In some implementations, the principal component space statistics and the residual space statistics are calculated by the following formulas:

[0116] T 2 =X T PS -1 P T X

[0117] SPE = X(I-PP T )X T

[0118] Where, T 2 is the principal component space statistic, SPE is the residual space statistic, X is the residual matrix of the battery cell, P is the low-dimensional feature matrix, I is the instantaneous current, S is the sample set X T X.

[0119] In some embodiments, the comprehensive evaluation index of the battery cell is calculated by the following formula:

[0120]

[0121] In the formula, is the total evaluation index of the battery cell, δ 2 is the control limit of the principal component space statistics, is the control limit of the residual space statistic, SPE is the residual space statistic, T 2 is the principal component space statistic, is a symmetric positive definite matrix, P k is the low-dimensional feature matrix, and Λ is the diagonal matrix composed of the eigenvalues ​​of the test sample covariance matrix.

[0122] Taking into account that the principal component space statistics and the residual space statistics can respectively reflect the degree to which the data deviates from the predicted value of the principal component model in the residual subspace and in the principal component space, therefore, in an embodiment of the present invention, the comprehensive indicators of the battery pack system at the current moment are calculated based on the principal component space statistics and the residual space statistics, which is conducive to accurately diagnosing whether there is a fault in the battery pack system at the current moment.

[0123] In the embodiment of the present invention, the comprehensive evaluation index of the battery cell conforms to the chi-square distribution with a degree of freedom of h and a coefficient of g. According to the probability function of the chi-square distribution The comprehensive evaluation index of the battery cell can be calculated, among which:

[0124]

[0125] In the formula, S is the training sample set

[0126] It should be noted that the preset threshold in step 106 can be set by expert experience threshold, 3σ criterion and other methods, and the absolute value of the indicator, the frequency of exceeding the threshold, etc. can be monitored to achieve quantification and graded alarm of battery pack failure.

[0127] Regarding step 108:

[0128] In some embodiments, step 108 includes:

[0129] Based on the characteristic matrix corresponding to the faulty battery cell, calculating the contribution of each element in the characteristic matrix to the principal component space statistics and the residual space statistics;

[0130] According to the contribution of each element in the feature matrix to the principal component space statistics and the residual space statistics, the contribution of each element in the feature matrix to the comprehensive index is obtained;

[0131] Based on the contribution of each element in the characteristic matrix to the comprehensive index and the initial state matrix of the battery cell, the fault type matrix of the faulty battery cell is constructed.

[0132] In an embodiment of the present invention, in order to determine the cause of the failure of a battery cell, the contribution of each characteristic element in the characteristic matrix corresponding to the faulty cell to the comprehensive index is first calculated, and then based on the contribution of each characteristic element to the comprehensive index and further combined with the initial state matrix of the battery cell, classification features of different fault types are extracted to obtain a fault type matrix of the faulty battery cell, and the matrix is ​​used as the input matrix of the fault type probability prediction model, so as to achieve accurate estimation of the probability of the fault type of the faulty battery cell.

[0133] In some embodiments, the contribution of each element in the feature matrix to the comprehensive index is calculated by the following formula:

[0134]

[0135]

[0136] CONTq=e 2

[0137] In the formula, CONT is the contribution of each element in the feature matrix to the comprehensive index, CONTq is the contribution of each element in the feature matrix to the residual space statistics, CONTj is the contribution of each element in the feature matrix to the principal component space statistics, t i is the score of the observations in the i-th dimension, λ i is the eigenvalue corresponding to the i-th principal component, P(i,j) is the element in the i-th row and j-th column of the low-dimensional matrix, x j is the jth variable in the battery cell characteristic matrix, u j is a column vector consisting of the mean values ​​of the jth column in the battery cell characteristic matrix;

[0138] In the embodiment of the present invention, it is necessary to calculate the total contribution of the characteristic matrix corresponding to the faulty battery cell to the statistics of the principal component space. Specifically, firstly, the standardized score of each actual observation matrix is ​​calculated. Assume that the preset threshold is T α ,by Determine the r scores that cause the battery pack failure state, determine the time when the failure occurs, and extract the principal component P at that time i , calculate the score t in the i-th dimension i :

[0139] t i =X T P i

[0140] After that, the contribution of each characteristic element in the characteristic matrix corresponding to the faulty battery cell to the score at the time of fault occurrence is calculated:

[0141]

[0142] In the formula, λ i is the eigenvalue corresponding to the i-th principal component, P(i, j) is the element in the i-th row and j-th column of the k-th low-dimensional matrix, x j is the jth variable in the battery cell characteristic matrix, u j is the mean value of the jth variable in the battery cell characteristic matrix;

[0143] Finally, the j calculated process variables x j Add the contribution of , and get the total contribution of the feature matrix to the statistics of the principal component space:

[0144]

[0145] Based on this, the fault type matrix of the faulty battery cell is constructed as follows:

[0146]

[0147] In the formula, CONT is the contribution of each element in the feature matrix to the comprehensive index, is the maximum value of the comprehensive evaluation index of the battery cell in a sampling period, ra is the proportion of the maximum value of the comprehensive evaluation index of the battery cell in 3000 consecutive sampling points, and T m For in [t m -3000,t m ]The maximum temperature value during the sampling period, t m is the time corresponding to the maximum value of the comprehensive evaluation index of the battery cell within a sampling period, r isc is the actual insulation resistance of the battery cell, U min It is the lowest voltage value of the battery cell within a sampling period.

[0148] Regarding step 110:

[0149] In an embodiment of the present invention, the prediction network model structure is composed of a linear layer and an activation function, and the probability of outputting a corresponding fault type is determined by evaluating the similarity with the label sample. Specifically, based on the fault types of the battery cells in the cloud discrete data, each fault type has a corresponding trainable weight and bias, and the fault types of the battery cells and the probability values ​​corresponding to each fault type are used as the input sample set and output sample set of the fault type probability learning model for training, thereby obtaining a fault type probability prediction model; at this point, the fault type matrix of the faulty battery cell constructed in step 108 is used as the input of the fault type probability prediction model, and the probability of each fault type causing the battery cell failure can be estimated respectively.

[0150] like Figure 2 , Figure 3 As shown, an embodiment of the present invention provides a fault diagnosis device for a random working condition battery pack. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. From the hardware level, Figure 2 As shown, it is a hardware architecture diagram of a computing device where a fault diagnosis device for a random working condition battery pack provided by an embodiment of the present invention is located, except Figure 2 In addition to the processor, memory, network interface, and non-volatile memory shown in the figure, the computing device in which the device is located in the embodiment may also generally include other hardware, such as a forwarding chip responsible for processing messages, etc. Taking software implementation as an example, Figure 3 As shown, as a device in a logical sense, the CPU of the computing device in which it is located reads the corresponding computer program in the non-volatile memory into the memory and runs it. This embodiment provides a fault diagnosis device for a random working condition battery pack, the device comprising:

[0151] A first calculation unit 301 is used to perform residual calculation based on the obtained state parameter prediction matrix of the battery cell and the actual observation parameter matrix to obtain a residual matrix of the battery cell; wherein the state parameter prediction matrix of the battery cell is obtained by prediction using a pre-trained deep learning network;

[0152] A second calculation unit 302 is used to perform characteristic calculation on the residual matrix of the battery cell to obtain a characteristic matrix of the battery cell;

[0153] The third calculation unit 303 is used to perform dimension reduction processing on the characteristic matrix of the battery cell by using the principal component analysis method to obtain a comprehensive evaluation index of the battery cell;

[0154] A fault determination unit 304 is used to compare the comprehensive evaluation index of each battery cell with a preset threshold value, and determine the battery cell whose comprehensive evaluation index exceeds the preset threshold value as a faulty battery cell;

[0155] A fault type matrix construction unit 305 is used to construct a fault type matrix of the faulty battery cell according to the characteristic matrix corresponding to the faulty battery cell and the initial state matrix of the battery cell;

[0156] The fault diagnosis unit 306 is used to input the fault type matrix of the faulty battery cell into the trained fault type probability prediction model to obtain the trigger probability corresponding to the fault type of the faulty battery cell, so as to perform fault diagnosis on the battery pack.

[0157] In one embodiment of the present invention, in the first computing unit 301, the deep learning network includes a filter, a temporal network prediction model, and a physical enhancement encoder connected in sequence;

[0158] The state parameter prediction matrix of the battery cell is predicted in the following way:

[0159] According to the voltage parameters of the battery cells obtained at the current moment, a battery pack state model and a battery cell deviation model are constructed;

[0160] Using filters to predict the state of the battery pack state model and the battery cell deviation model at the current moment, respectively, to obtain the filter prediction results of the battery pack at the next moment and the filter prediction results of the battery cell at the next moment; wherein the filter prediction results include the voltage and charge state of the battery pack and the battery cell at the next moment;

[0161] The battery state parameters and vehicle state parameters obtained at the current moment are input into the trained time series network prediction model to obtain the time series network prediction result of the battery pack at the next moment; wherein, during the time series network prediction model training process, the filter prediction result of the battery pack is used as a loss function to reversely adjust the time series network prediction result;

[0162] The timing network prediction result of the battery pack and the filter prediction result of the battery cell are input into the trained physical enhancement encoder to obtain the state parameter prediction matrix of the battery cell.

[0163] In one embodiment of the present invention, in the first calculation unit 302, the characteristic matrix of the battery cell is as follows:

[0164] X=[ewu,ewx,zu1(k),zx1(k),zu2(k),zx2(k)],

[0165]

[0166] In the formula, zu1(k) and zx1(k) are the maximum outlier coefficients of the corresponding variables, zu2(k) and zx2(k) are the deviations between the maximum and the second largest outlier coefficients. is the voltage value of the first row in the battery cell residual matrix, is the average voltage value of the first row in the battery cell residual matrix, is the state of charge value of the second row in the battery cell residual matrix, is the standard deviation of the state of charge value in the second row of the battery cell residual matrix, is the standard deviation of the voltage value in the first row of the battery cell residual matrix, is the average value of the state of charge value in the second row of the battery cell residual matrix, is the maximum voltage value in the first row of the residual matrix of the battery cell, is the maximum value of the state of charge value in the second row of the residual matrix of the battery cell, is the second largest voltage value in the first row of the residual matrix of the battery cell, is the second largest value of the state of charge value in the second row of the residual matrix of the battery cell, ewu is the exponentially weighted moving average filtered signal of the maximum voltage value in the first row of the residual matrix, and ewx is the exponentially weighted moving average filtered signal of the maximum state of charge value in the second row of the residual matrix.

[0167] In one embodiment of the present invention, the third calculation unit 303 is used to perform the following operations when the principal component analysis method is used to perform dimensionality reduction processing on the characteristic matrix of the battery cell:

[0168] Using principal component analysis to reduce the dimension of the characteristic matrix of the battery cell to obtain a low-dimensional characteristic matrix;

[0169] Obtaining principal component space statistics and residual space statistics according to the characteristic matrix of the battery cell and the low-dimensional characteristic matrix;

[0170] Based on the principal component space statistics and the residual space statistics, a comprehensive evaluation index of the battery cell is obtained.

[0171] In one embodiment of the present invention, in the third calculation unit 303, the principal component space statistics and the residual space statistics are calculated by the following formulas:

[0172] T 2 =X T PS -1 P T X

[0173] SPE = X(I-PP T )X T

[0174] Where, T 2 is the principal component space statistic, SPE is the residual space statistic, X is the residual matrix of the battery cell, P is the low-dimensional feature matrix, and I is the instantaneous current.

[0175] In one embodiment of the present invention, in the third calculation unit 303, the comprehensive evaluation index of the battery cell is calculated by the following formula:

[0176]

[0177] In the formula, is the total evaluation index of the battery cell, δ 2 is the control limit of the principal component space statistics, is the control limit of the residual space statistic, SPE is the residual space statistic, T 2 is the principal component space statistic, is a symmetric positive definite matrix, P is the low-dimensional feature matrix, and Λ is a diagonal matrix composed of eigenvalues ​​of the test sample covariance matrix.

[0178] In one embodiment of the present invention, when constructing the fault type matrix unit 305, it is used to perform the following operations:

[0179] Based on the characteristic matrix corresponding to the faulty battery cell, calculating the contribution of each element in the characteristic matrix to the principal component space statistics and the residual space statistics;

[0180] According to the contribution of each element in the feature matrix to the principal component space statistics and the residual space statistics, the contribution of each element in the feature matrix to the comprehensive index is obtained;

[0181] Based on the contribution of each element in the characteristic matrix to the comprehensive index and the initial state matrix of the battery cell, the fault type matrix of the faulty battery cell is constructed.

[0182] In one embodiment of the present invention, in the fault type matrix construction unit 305, the contribution of each element in the characteristic matrix to the comprehensive index is calculated by the following formula:

[0183]

[0184] CONTq=e 2

[0185] In the formula, CONT is the contribution of each element in the feature matrix to the comprehensive index, CONTq is the contribution of each element in the feature matrix to the residual space statistics, CONTj is the contribution of each element in the feature matrix to the principal component space statistics, t i is the observed value score of the i-th load vector, λ i is the eigenvalue corresponding to the i-th principal component, P(i, j) is the element in the i-th row and j-th column of the low-dimensional matrix, x j is the jth column in the battery cell characteristic matrix, u j is the column vector consisting of the mean of the jth column in the battery cell characteristic matrix

[0186] In one embodiment of the present invention, in the fault type matrix construction unit 305, the fault type matrix of the faulty battery cell is as follows:

[0187]

[0188] In the formula, CONT is the contribution of each element in the feature matrix to the comprehensive index, is the maximum value of the comprehensive evaluation index of the battery cell in a sampling period, ra is the proportion of the maximum value of the comprehensive evaluation index of the battery cell in 3000 consecutive sampling points, and T m For in [t m -3000,t m ]The maximum temperature value during the sampling period, t m is the time corresponding to the maximum value of the comprehensive evaluation index of the battery cell within a sampling period, r isc is the actual insulation resistance of the battery cell, U min It is the lowest voltage value of the battery cell within a sampling period.

[0189] It is to be understood that the structure illustrated in the embodiment of the present invention does not constitute a specific limitation on a fault diagnosis device for a random working condition battery pack. In other embodiments of the present invention, a fault diagnosis device for a random working condition battery pack may include more or fewer components than shown in the figure, or combine certain components, or split certain components, or arrange the components differently. The components shown in the figure may be implemented in hardware, software, or a combination of software and hardware.

[0190] The information interaction, execution process and other contents between the modules in the above-mentioned device are based on the same concept as the embodiment of the method of the present invention. For the specific contents, please refer to the description in the embodiment of the method of the present invention, and no further description is given here.

[0191] An embodiment of the present invention further provides a computing device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, a fault diagnosis method for a random operating condition battery pack in any embodiment of the present invention is implemented.

[0192] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the processor executes a fault diagnosis method for a random operating condition battery pack in any embodiment of the present invention.

[0193] Specifically, a system or device equipped with a storage medium can be provided, on which software program code that implements the functions of any of the above-mentioned embodiments is stored, and a computer (or CPU or MPU) of the system or device can be enabled to read and execute the program code stored in the storage medium.

[0194] In this case, the program code itself read from the storage medium can realize the function of any one of the above-mentioned embodiments, and thus the program code and the storage medium storing the program code constitute a part of the present invention.

[0195] The storage medium embodiments for providing the program code include a floppy disk, a hard disk, a magneto-optical disk, an optical disk (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), a magnetic tape, a non-volatile memory card, and a ROM. Alternatively, the program code can be downloaded from a server computer by a communication network.

[0196] In addition, it should be clear that the functions of any of the above embodiments can be implemented not only by executing the program code read by the computer, but also by enabling an operating system operating on the computer to complete part or all of the actual operations based on instructions from the program code.

[0197] In addition, it can be understood that the program code read from the storage medium is written to a memory provided in an expansion board inserted into the computer or to a memory provided in an expansion module connected to the computer, and then based on the instructions of the program code, a CPU installed on the expansion board or expansion module is enabled to perform part or all of the actual operations, thereby realizing the functions of any of the above-mentioned embodiments.

[0198] It should be noted that, in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0199] A person of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk, etc., various media that can store program codes.

[0200] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A fault diagnosis method for a random working condition battery pack, characterized in that: include: Perform residual calculation based on the obtained state parameter prediction matrix of the battery cell and the actual observation parameter matrix to obtain a residual matrix of the battery cell; wherein the state parameter prediction matrix of the battery cell is obtained by prediction using a pre-trained deep learning network; Performing feature calculation on the residual matrix of the battery cell to obtain a feature matrix of the battery cell; Using principal component analysis to reduce the dimension of the characteristic matrix of the battery cell to obtain a comprehensive evaluation index of the battery cell; Comparing the comprehensive evaluation index of each battery cell with a preset threshold, and determining the battery cell whose comprehensive evaluation index exceeds the preset threshold as a faulty battery cell; Constructing a fault type matrix of the faulty battery cell according to the characteristic matrix corresponding to the faulty battery cell and the initial state matrix of the battery cell; The fault type matrix of the faulty battery cell is input into a trained fault type probability prediction model to obtain the trigger probability corresponding to the fault type of the faulty battery cell, so as to perform fault diagnosis on the battery pack.

2. The method according to claim 1, characterized in that The deep learning network includes a filter, a temporal network prediction model and a physical enhancement encoder connected in sequence; The state parameter prediction matrix of the battery cell is predicted in the following way: For each current moment, execute: According to the voltage parameters of the battery cells obtained at the current moment, a battery pack state model and a battery cell deviation model are constructed; The filter is used to predict the state of the battery pack state model and the battery cell deviation model at the current moment, and the filter prediction results of the battery pack and the battery cell at the next moment are obtained respectively; The battery state parameters and vehicle state parameters obtained at the current moment are input into the trained time series network prediction model to obtain the time series network prediction result of the battery pack at the next moment; wherein, during the time series network prediction model training process, the filter prediction result of the battery pack is used as a loss function to reversely adjust the time series network prediction result; The timing network prediction result of the battery pack and the filter prediction result of the battery cell are input into the trained physical enhancement encoder to obtain the state parameter prediction matrix of the battery cell.

3. The method according to claim 1, characterized in that The characteristic matrix of the battery cell is as follows: X=[ewu,ewx,zu1(k),zx1(k),zu2(k),zx2(k)], In the formula, zu1(k) and zx1(k) are the maximum outlier coefficients of the corresponding variables, zu2(k) and zx2(k) are the deviations between the maximum and the second largest outlier coefficients, k is the sampling time, is the voltage value of the first row in the battery cell residual matrix, is the average voltage value of the first row in the battery cell residual matrix, is the state of charge value of the second row in the battery cell residual matrix, is the standard deviation of the state of charge value in the second row of the battery cell residual matrix, is the standard deviation of the voltage value in the first row of the battery cell residual matrix, is the average value of the state of charge value in the second row of the battery cell residual matrix, is the maximum voltage value in the first row of the residual matrix of the battery cell, is the maximum value of the state of charge value in the second row of the residual matrix of the battery cell, is the second largest voltage value in the first row of the residual matrix of the battery cell, is the second largest value of the state of charge value in the second row of the residual matrix of the battery cell, ewu is the exponentially weighted moving average filtered signal of the maximum voltage value in the first row of the residual matrix, and ewx is the exponentially weighted moving average filtered signal of the maximum state of charge value in the second row of the residual matrix.

4. The method according to claim 1, characterized in that: The principal component analysis method is used to perform dimension reduction processing on the characteristic matrix of the battery cell to obtain the comprehensive evaluation index of the battery cell, including: Using principal component analysis to reduce the dimension of the characteristic matrix of the battery cell to obtain a low-dimensional characteristic matrix; Obtaining principal component space statistics and residual space statistics according to the characteristic matrix of the battery cell and the low-dimensional characteristic matrix; Based on the principal component space statistics and the residual space statistics, a comprehensive evaluation index of the battery cell is obtained.

5. The method according to claim 1, characterized in that The principal component space statistics and the residual space statistics are calculated by the following formulas: T 2 =X T PS -1 P T X SPE=X(I-PP T )X T Where, T 2 is the principal component space statistic, SPE is the residual space statistic, X is the residual matrix of the battery cell, P is the low-dimensional feature matrix, and I is the instantaneous current; and / or, The comprehensive evaluation index of the battery cell is calculated by the following formula: In the formula, is the total evaluation index of the battery cell, δ 2 is the control limit of the principal component space statistics, is the control limit of the residual space statistic, SPE is the residual space statistic, T 2 is the principal component space statistic, is a symmetric positive definite matrix, P is the low-dimensional feature matrix, and Λ is a diagonal matrix composed of eigenvalues ​​of the test sample covariance matrix.

6. The method according to claim 4, characterized in that The step of constructing a fault type matrix of the faulty battery cell according to the characteristic matrix corresponding to the faulty battery cell and the initial state matrix of the battery cell comprises: Based on the characteristic matrix corresponding to the faulty battery cell, calculating the contribution of each element in the characteristic matrix to the principal component space statistics and the residual space statistics; According to the contribution of each element in the feature matrix to the principal component space statistics and the residual space statistics, the contribution of each element in the feature matrix to the comprehensive index is obtained; Based on the contribution of each element in the characteristic matrix to the comprehensive index and the initial state matrix of the battery cell, the fault type matrix of the faulty battery cell is constructed.

7. The method according to claim 6, characterized in that The contribution of each element in the feature matrix to the comprehensive index is calculated by the following formula: CONTq=e 2 In the formula, CONT is the contribution of each element in the feature matrix to the comprehensive index, CONTq is the contribution of each element in the feature matrix to the residual space statistics, CONTj is the contribution of each element in the feature matrix to the principal component space statistics, ti is the observed value score of the i-th load vector, λ i is the eigenvalue corresponding to the i-th principal component, P(i, j) is the element in the i-th row and j-th column of the low-dimensional matrix, x j is the jth column in the battery cell characteristic matrix, u j is a column vector consisting of the mean values ​​of the jth column in the battery cell characteristic matrix; and / or, The fault type matrix of the faulty battery cell is as follows: In the formula, CONT is the contribution of each element in the feature matrix to the comprehensive index, is the maximum value of the comprehensive evaluation index of the battery cell in a sampling period, ra is the proportion of the maximum value of the comprehensive evaluation index of the battery cell in 3000 consecutive sampling points, and T m is the maximum temperature value in the sampling period of [tm-3000, tm], tm is the time corresponding to the maximum value of the comprehensive evaluation index of the battery cell in a sampling period, r isc is the actual insulation resistance of the battery cell, U min It is the lowest voltage value of the battery cell within a sampling period.

8. A fault diagnosis device for a random working condition battery pack, characterized in that: include: A first calculation unit is used to perform residual calculation based on the obtained state parameter prediction matrix of the battery cell and the actual observation parameter matrix to obtain a residual matrix of the battery cell; wherein the state parameter prediction matrix of the battery cell is obtained by prediction using a pre-trained deep learning network; A second calculation unit, configured to perform characteristic calculation on the residual matrix of the battery cell to obtain a characteristic matrix of the battery cell; a third calculation unit, configured to perform dimensionality reduction processing on the characteristic matrix of the battery cell by using a principal component analysis method to obtain a comprehensive evaluation index of the battery cell; A fault determination unit, used to compare the comprehensive evaluation index of each battery cell with a preset threshold value, and determine the battery cell whose comprehensive evaluation index exceeds the preset threshold value as a faulty battery cell; A fault type matrix construction unit, used to construct a fault type matrix of the faulty battery cell according to a characteristic matrix corresponding to the faulty battery cell and an initial state matrix of the battery cell; The fault diagnosis unit is used to input the fault type matrix of the faulty battery cell into the trained fault type probability prediction model to obtain the trigger probability corresponding to the fault type of the faulty battery cell, so as to perform fault diagnosis on the battery pack.

9. A computing device, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to execute the method according to any one of claims 1 to 7.

Images