A method for identifying the risk of thermal runaway of a lithium-ion battery pack for an electric vehicle

Abstract

The present invention provides a method for identifying the thermal runaway risk of a lithium-ion battery pack for an electric vehicle. First, strongly correlated data features of the thermal runaway risk in the charging cycles of all single cells in the battery pack are screened out. Then, the Gaussian mixture model is used to calculate the risk feature distance and the cumulative risk feature distance of each single cell sample relative to the reference sample. Based on the cumulative risk feature distance threshold, it can be used as a quantitative index for effective identification of the thermal runaway risk. Standardization processing can also be performed on the risk feature distance and the cumulative one, so that the determined quantitative index for risk identification has high generality for different types of power battery packs for electric vehicles.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault diagnosis of lithium-ion batteries for electric vehicles, and particularly relates to a method for identifying the thermal runaway risk of a lithium-ion battery pack for an electric vehicle. Background Art

[0002] In the actual use process of the power battery system of new energy vehicles, abuse behaviors including thermal abuse, electrical abuse, and mechanical abuse often cause thermal runaway. If not handled properly, it is very likely to induce serious safety risks. Therefore, it is particularly important to identify battery system thermal runaway in a timely manner. At present, the existing battery thermal runaway identification mainly includes three categories: rule-based, model-based, and data-driven methods. Among them, the rule-based method has a simple algorithm and high computational efficiency, but the preset threshold needs to be continuously adjusted by experience, and the same threshold cannot be universal among different types of battery packs; the model-based method depends on the modeling accuracy of the system model, has poor robustness to modeling errors, and the performance of this type of algorithm is only verified under laboratory and simulation conditions, without fully considering the complex and variable working conditions of actual vehicle driving; the data-driven method constructs statistical features by combining the power battery operation monitoring data stored in the cloud data platform of the actual vehicle, combines machine learning methods to identify abnormal single cells in the battery pack, and quantifies the thermal runaway risk of the power battery through data features. Compared with the previous two methods, it has obvious advantages, but there are still many problems to be solved. Due to the characteristics of multi-source, heterogeneous data of actual vehicle data and large differences in the correlation with thermal runaway, at present, feature screening still mainly relies on manual trial and error, lacking high-efficiency automatic analysis of feature effectiveness, and the strong correlation between features and thermal runaway risk is difficult to meet the requirements; at the same time, the data-driven method itself has high requirements for computing power, while the real-time data collected during actual vehicle driving is more vulnerable to noise interference. Therefore, real-time processing of each time frame makes the computational load of the battery management system remain high, and the accuracy of the risk identification result is also affected by noise. For example, the above problems are obvious in Chinese patents CN116184229A, CN115166533A, CN116125290A, CN111812535A, and the data items such as SOC and SOH used by them cannot be directly collected by measurement, which will further increase the computational burden, and the extracted temperature data items usually cannot reflect the true temperature of the battery surface. Therefore, there is an urgent need in this field for a thermal runaway risk identification method with high data feature screening efficiency, low computational load, and reduced influence of data collection noise. Summary of the Invention

[0003] In view of this, aiming at the technical problems existing in the field, the present invention provides a method for identifying the thermal runaway risk of a lithium-ion battery pack for an electric vehicle, which specifically includes the following steps:

[0004] Step 1: Extract the single-cell voltage time series data V of the last parking and charging cycle in the whole life cycle of several electric vehicles with single-cell thermal runaway, and add classification labels indicating safety or failure status to the single-cell voltage time series of each single cell to obtain the classification label set Y;

[0005] Step 2: Use the Tsfresh automated feature engineering tool to extract multiple feature data of each single-cell battery voltage time series to form the feature set F, and jointly construct it with the classification label set Y as the training set; use the training set to train the random forest model, and output the total information gain f obtained when splitting each feature with the classification label of the safe state in the decision tree, a which is used to reflect the actual importance of each feature;

[0006] Step 3: After randomly shuffling the classification label set Y k times, obtain a new classification label set Y1, reconstruct the training set with it and the feature set F, and train the random forest model again. Output the total information gain f obtained when splitting each feature after the classification label is shuffled, n which is used to reflect the importance of each feature when the classification label is shuffled;

[0007] Step 4: Based on the importance of each feature obtained in Steps 2 and 3, perform feature screening and retain several strongly correlated features of the thermal runaway risk;

[0008] Step 5: For the electric vehicle to be detected, extract the single-cell voltage time series data of the last parking and charging cycle before the accident caused by thermal runaway, and extract the feature sample set F composed of the values of each strongly correlated feature; s ; perform a clustering algorithm on the feature sample set F s and output the classification labels of the safety status or the presence of thermal runaway risk corresponding to each single cell after clustering, which is used to automatically mark and locate the single-cell battery with the thermal runaway risk;

[0009] Step 6: Establish a Gaussian mixture model to calculate the probability likelihood of the distribution of each feature in the feature sample set F; s After inputting the single-cell voltage time series data of any kth parking and charging cycle in the whole life cycle of the electric vehicle to be detected into the Gaussian mixture model for calculation, set the single cell with the maximum log-likelihood of the feature samples of all single-cell batteries as the reference single-cell battery; calculate the Euclidean distance between the feature samples of other single-cell batteries and the reference single-cell battery as the risk feature distance; traverse the cumulative risk feature distances of each single-cell battery in the K parking and charging cycles in the whole life cycle of the electric vehicle to be detected, and set the corresponding threshold;

[0010] Step 7: According to the change trends of the risk feature distance and the cumulative risk feature distance, judge the time when the thermal runaway risk single cell first occurs and the subsequent risk evolution law.

[0011] Further, the monomer voltage time series data V extracted in step one is specifically in the following matrix form:

[0012]

[0013] where M is the number of monomer cells in the electric vehicle battery pack, T is the data length of the time series of this charging cycle, and the data acquisition frequency is 10 s / frame;

[0014] Add a classification label y indicating the safe or faulty state to the monomer voltage time series of each monomer cell i (i ∈ M), to obtain the following classification label set Y:

[0015] Y = [y1, y2,..., y M T

[0016] If any monomer cell is a thermal runaway accident monomer cell, its label is 1; if it is a safe monomer cell, the label is 0.

[0017] Further, in step two, specifically use the Tsfresh tool to extract 785 time-frequency domain feature data of the voltage time series of each monomer cell, to obtain the following feature set F:

[0018]

[0019] where f MN represents the Nth feature of the voltage time series of the Mth monomer cell;

[0020] Based on the feature set F and the classification label set Y, jointly construct the following training set D for training the random forest model:

[0021]

[0022] In step three, randomly shuffle the classification label set Y to obtain a new classification label set Y1, and then construct a training set D1 = [F, Y1] for retraining the random forest model.

[0023] Further, in step four, based on the information gain f a and f n and use the following importance scoring formula to screen the strongly correlated features of thermal runaway risk:

[0024]

[0025] where percentile(f n , 0.75) represents the 75th percentile of f n ;​

[0026] Select and retain w feature items in the feature set F with importance scores imp_s greater than 0 as the strongly correlated features f for thermal runaway risk s :

[0027] f s = [f s1 , f s2 ,..., f sw .

[0028] Furthermore, in step five, the following feature sample set F composed of each strongly correlated feature value is extracted for the electric vehicle to be detected s :

[0029]

[0030] In the formula, f sMw represents the w-th strongly correlated feature for thermal runaway risk in the voltage time series data of the M-th single battery of the thermal runaway electric vehicle;

[0031] Perform clustering on F s using the following clustering algorithm model specifically:

[0032] First, set the ε-neighborhood parameter range list and Minpts parameter of the density clustering model, traverse the ε-neighborhood parameter range list, and construct clustering models with different ε-neighborhood parameters;

[0033] After that, input F s into the clustering models with different ε-neighborhood parameters, and calculate the silhouette coefficient s of each model through the following formula:

[0034]

[0035]

[0036] In the formula, s(i) represents the silhouette coefficient of the i-th single battery sample; a(i) represents the dissimilarity within the same cluster, that is, the average of the dissimilarities of all samples in the cluster to other points in the same cluster; b(i) represents the dissimilarity between different clusters, that is, the minimum of the average dissimilarity degrees of all samples in the same cluster to other clusters;

[0037] The value range of the silhouette coefficient is [-1, 1]. The closer it is to 1, the higher the score of the clustering algorithm and the better the clustering effect. Select the clustering model with the highest silhouette coefficient and output its labeling result Y for the single battery after clustering cluster , as shown in the following formula:

[0038] Y cluster = [y c1 , yc2 ,...,y cM T

[0039] In the formula, y ci represents the thermal runaway risk label of the i-th single battery of the electric vehicle. If y ci =-1, it means that the clustering algorithm marks this single battery as a single battery with thermal runaway risk; if y ci =0, it means that the clustering algorithm marks this single battery as a safe single battery. Thus, the automatic marking and positioning of single batteries with thermal runaway risk is realized.

[0040] Furthermore, the Gaussian mixture model established in step six specifically describes the probability density of the distribution of each feature in the strong thermal runaway risk association feature set F s with p(x|θ); where x represents the observed value of the statistical sample, θ represents the distribution that x follows, and is composed of g clusters of Gaussian components; the number of clusters g is specifically determined based on the AIC and BIC criteria, and the parameters of θ are obtained by iterative expectation maximization method;

[0041] For the statistical characteristics of the voltage sequences of each single battery in the k-th charging cycle sequence input to the Gaussian mixture model, the specific distribution parameters are obtained by iterative expectation maximization method: calculate the probability that the observed value x i comes from the i-th cluster of Gaussian components, so that the logarithm of the probability likelihood reaches the maximum, and the algorithm converges by iterative calculation, and finally the Gaussian distributions θ k and their corresponding weights α k are obtained; the r-th single battery with the maximum log-likelihood of the feature samples of all single batteries is used as the reference single battery, and the strong thermal runaway risk association feature sets of each item are expressed as:

[0042] F sr =[f sr1 ,f sr2 ,...,f srw

[0043]

[0044] The strong thermal runaway risk association features of other single batteries are expressed as:

[0045] F si =[f si1 ,f si2 ,...,f siw

[0046] Calculate the Euclidean distance between the samples of other single batteries and the reference single battery samples:

[0047] ​​​

[0048] Perform the following normalization to obtain the risk characteristic distance within the range of [0, 1]:

[0049]

[0050] d sk =[d sk1 ,d sk2 ,...,d skM

[0051] Traverse the K parking and charging cycles in the whole life cycle of the electric vehicle to be detected to obtain the following risk characteristic distance matrix D sk :

[0052]

[0053] The cumulative risk characteristic distance of the corresponding single battery i is calculated by the following formula:

[0054]

[0055] Furthermore, for different electric vehicle models using the same battery, the standardized cumulative risk characteristic distance is calculated through the following normalization to meet the needs of thermal runaway risk identification for different models:

[0056]

[0057] The method for identifying the thermal runaway risk of the lithium-ion battery pack of an electric vehicle provided by the present invention first screens out the strong correlation data features of the thermal runaway risk in the charging cycles of all single batteries in the battery pack, and then uses the Gaussian mixture model to calculate the risk characteristic distance and cumulative risk characteristic distance of each single battery sample relative to the reference sample. Based on the cumulative risk characteristic distance threshold, it can be used as a quantitative index for effective identification of the thermal runaway risk. The risk characteristic distance and cumulative can also be standardized, so that the determined quantitative index for risk identification has high generality for different types of power battery packs of electric vehicles. Description of the Drawings

[0058] Figure 1 is the flowchart for automatic extraction of strong correlation features of thermal runaway risk and positioning and marking of risk single cells of the present invention;

[0059] Figure 2 is the result diagram of positioning and marking of single cells with thermal runaway risk achieved by clustering;

[0060] Figure 3 is the analysis process diagram of single cell risk characteristic measurement and risk evolution law based on the Gaussian mixture model;

[0061] ​Figure 4 Schematic diagram for setting a safety threshold quantization index based on the cumulative risk feature distance

[0062] Figure 5 It is a diagram of the voltage time-domain characteristic change of the thermal runaway risk monomer corresponding to the risk feature distance Specific implementation manners

[0063] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention

[0064] The present invention provides a method for identifying the thermal runaway risk of an electric vehicle lithium-ion battery pack, as Figure 1 、 2 shown, which specifically includes the following steps

[0065] Step 1: Extract the single-cell voltage time series data V of the last parking and charging cycle in the whole life cycle of several electric vehicles with single-cell thermal runaway, and add a classification label indicating the safe or faulty state to the single-cell voltage time series of each cell to obtain a classification label set Y

[0066] Step 2: Use the Tsfresh automated feature engineering tool to extract multiple feature data of each single-cell voltage time series to form a feature set F, and jointly construct it with the classification label set Y into a training set; use the training set to train a random forest model, and output the sum f of the information gains obtained when splitting each feature with the classification label of the safe state in the decision tree a , which is used to reflect the actual importance of each feature

[0067] Step 3: After randomly shuffling the classification label set Y k times, a new classification label set Y1 is obtained, which is reconstructed with the feature set F into a training set and the random forest model is trained again, and the sum f of the information gains obtained when splitting each feature with the shuffled classification label is output n , which is used to reflect the importance of each feature when the classification label is shuffled

[0068] Step 4: Perform feature screening based on the importance of each feature obtained in Steps 2 and 3, and retain several strongly correlated features of the thermal runaway risk

[0069] Step 5: For the electric vehicle to be detected, extract the single-cell voltage time series data of the last parking and charging cycle before the accident caused by thermal runaway, and extract the feature sample set F composed of the values of each strongly correlated feature s ; for the feature sample set F sExecute the clustering algorithm to output the safety status or the classification label of the risk of thermal runaway corresponding to each monomer after clustering, which is used to automatically mark and locate the monomer battery with the risk of thermal runaway;

[0070] Step Six: Establish a Gaussian mixture model to calculate the probability likelihood of the distribution of each feature in the feature sample set F s After inputting the monomer voltage time series data of any kth parking and charging cycle in the whole life cycle of the electric vehicle to be detected into the Gaussian mixture model for calculation, set the monomer with the maximum logarithmic likelihood of the feature samples of all monomer batteries as the reference monomer battery; calculate the Euclidean distance between the feature samples of other monomer batteries and the reference monomer battery as the risk feature distance; traverse the cumulative risk feature distances of each monomer battery in the K parking and charging cycles in the whole life cycle of the electric vehicle to be detected, and set the corresponding threshold;

[0071] Step Seven: According to the change trends of the risk feature distance and the cumulative risk feature distance, judge the time when the risk of the thermal runaway monomer first occurs and the subsequent risk evolution law.

[0072] In the preferred embodiment of the present invention, the monomer voltage time series data V extracted in Step One specifically adopts the following matrix form:

[0073]

[0074] In the formula, M is the number of monomer batteries in the electric vehicle battery pack, T is the data length of the time series of this charging cycle, and the data acquisition frequency is 10 s / frame;

[0075] Add a classification label y representing the safe or faulty state to the monomer voltage time series of each monomer battery i (i ∈ M) to obtain the following classification label set Y:

[0076] Y = [y1, y2,..., y M T

[0077] If any monomer battery is a thermal runaway accident monomer battery, its label is 1; if it is a safe monomer battery, the label is 0.

[0078] In Step Two, specifically use the Tsfresh tool to extract 785 time-frequency domain feature data of the voltage time series of each monomer battery to obtain the following feature set F:

[0079]

[0080] In the formula, f MN represents the Nth feature of the voltage time series of the Mth monomer battery;

[0081] ​Based on the feature set F and the classification label set Y, the following training set D is jointly constructed for training the random forest model:

[0082]

[0083] In step three, the classification label set Y is randomly shuffled to obtain a new classification label set Y1, and then the training set D1 = [F, Y1] is constructed for retraining the random forest model.

[0084] In step four, based on the information gain f a and f n and the following importance scoring formula is used to screen the features strongly associated with the thermal runaway risk:

[0085]

[0086] In the formula, percentile(f n , 0.75) represents the 75th percentile of f n ;

[0087] Screen and retain w feature items in the feature set F with an importance score imp_s greater than 0 as the features strongly associated with the thermal runaway risk f s :

[0088] f s = [f s1 , f s2 ,..., f sw .

[0089] In the preferred example of this invention, the following 15 features strongly associated with the thermal runaway risk are automatically screened by performing the above steps:

[0090] Feature 1: Sum of the monomer voltage sequences

[0091]

[0092] In the formula, T represents the data length of the charging cycle being analyzed, and u t represents the monomer voltage value at the t-th time frame of the charging cycle;

[0093] Feature 2: Sum of the squares of the monomer voltage sequences

[0094]

[0095] In the formula, T represents the data length of the charging cycle being analyzed, and u t represents the monomer voltage value at the t-th time frame of the charging cycle;

[0096] Feature 3: Sum of the absolute values of the continuous changes in the monomer voltage sequences

[0097]

[0098] In the formula, T represents the data length of the charging cycle to be analyzed, and u t represents the cell voltage value at the t-th time frame of the charging cycle;

[0099] Feature 4: Mean of the continuous change of the cell voltage sequence

[0100]

[0101] In the formula, T represents the data length of the charging cycle to be analyzed, and u t represents the cell voltage value at the t-th time frame of the charging cycle;

[0102] Feature 5: Mean of the cell voltage sequence

[0103]

[0104] In the formula, T represents the data length of the charging cycle to be analyzed, and u t represents the cell voltage value at the t-th time frame of the charging cycle;

[0105] Feature 6: Standard deviation of the cell voltage sequence

[0106]

[0107] In the formula, T represents the data length of the charging cycle to be analyzed, and u t represents the cell voltage value at the t-th time frame of the charging cycle, and μ is the mean cell voltage of this charging cycle;

[0108] Feature 7: Coefficient of variation of the cell voltage

[0109]

[0110] In the formula, T represents the data length of the charging cycle to be analyzed, and u t represents the cell voltage value at the t-th time frame of the charging cycle, μ is the mean cell voltage of this charging cycle, and σ is the standard deviation of the cell voltage of this charging cycle;

[0111] Feature 8: Variance of the cell voltage

[0112]

[0113] In the formula, T represents the data length of the charging cycle to be analyzed, and u t represents the cell voltage value at the t-th time frame of the charging cycle, and μ is the mean cell voltage of this charging cycle;

[0114] Feature 9: Kurtosis of the cell voltage

[0115]

[0116] In the formula, T represents the data length of the charging cycle to be analyzed, u t represents the single-cell voltage value at the t-th time frame of the charging cycle, μ is the average single-cell voltage of this charging cycle, and σ is the standard deviation of the single-cell voltage of this charging cycle;

[0117] Feature 10: Root mean square of single-cell voltage

[0118]

[0119] In the formula, T represents the data length of the charging cycle to be analyzed, u t represents the single-cell voltage value at the t-th time frame of the charging cycle;

[0120] Feature 11: Proportion of singular values in the single-cell voltage period sequence

[0121]

[0122] In the formula, T represents the data length of the charging cycle to be analyzed, and the numerator is the number of singular values of the single-cell voltage in this charging cycle;

[0123] Feature 12: 60% quantile of the single-cell voltage period sequence

[0124] f s12 = percentile(U t , 0.6)

[0125] In the formula, T represents the data length of the charging cycle to be analyzed, U t represents the single-cell voltage sequence at the t-th time frame of the charging cycle;

[0126] Feature 13: Statistical quantity of time-reversal asymmetry degree of single-cell voltage at lag order

[0127]

[0128] In the formula, T represents the data length of the charging cycle to be analyzed, u t represents the single-cell voltage value at the t-th time frame of the charging cycle, lag is the order, and it takes positive integers;

[0129] Feature 14: Autocorrelation coefficient of single-cell voltage at lag order

[0130]

[0131] In the formula, T represents the data length of the charging cycle to be analyzed, u t represents the single-cell voltage value at the t-th time frame of the charging cycle, μ is the average single-cell voltage of this charging cycle, lag is the order, and it takes positive integers;

[0132] Feature 15: Sample entropy of single-cell voltage sequence

[0133] Calculate the absolute value of the difference between the voltage values u corresponding to two moments i and j in the single-cell voltage vector of a charging cycle i and u j :

[0134] d[u i ,u j = |u i - u j |

[0135] Define B i as the number of times the distance between these two voltage values is less than or equal to r

[0136]

[0137] Increase the dimension to m + 1 and calculate the number Ai of times the distance between two voltage values is less than or equal to r

[0138]

[0139] Calculate the sample entropy of the single-cell voltage

[0140]

[0141] In the formula, T is the length of the voltage sequence of this charging cycle, and m = 1

[0142] In step five, extract the following feature sample set F composed of each strong correlation eigenvalue for the electric vehicle to be detected s :

[0143]

[0144] In the formula, f sMw represents the w-th strong correlation feature of the thermal runaway risk of the voltage time series data of the M-th single cell of this thermal runaway electric vehicle

[0145] Perform clustering on F s Specifically, adopt the following clustering algorithm model

[0146] First, set the ε-neighborhood parameter range list and Minpts parameter of the density clustering model, traverse the ε-neighborhood parameter range list, and construct clustering models with different ε-neighborhood parameters

[0147] After that, input F s into the clustering models with different ε-neighborhood parameters, and calculate the silhouette coefficient s of each model through the following formula

[0148]

[0149]

[0150] In the formula, s(i) represents the silhouette coefficient of the i-th single battery sample; a(i) represents the dissimilarity within the same cluster, that is, the average of the dissimilarities of all samples in the cluster to other points in the same cluster; b(i) represents the dissimilarity between different clusters, that is, the minimum of the average dissimilarity degrees of all samples in the same cluster to other clusters;

[0151] The value range of the silhouette coefficient is [-1, 1]. The closer it is to 1, the higher the score of the clustering algorithm and the better the clustering effect. Select the clustering model with the highest silhouette coefficient and output the labeling result Y of the single battery after clustering cluster , as shown in the following formula:

[0152] Y cluster = [y c1 , y c2 ,..., y cM T

[0153] In the formula, y ci represents the thermal runaway risk label of the i-th single battery of the electric vehicle. If y ci = -1, it means that the clustering algorithm labels this single battery as a thermal runaway risk single battery; if y ci = 0, it means that the clustering algorithm labels this single battery as a safe single battery. Thus, the automatic labeling and positioning of single batteries with thermal runaway risk are realized.

[0154] Of course, those skilled in the art can also adopt other relatively simple or more complex and accurate existing clustering algorithms according to considerations such as computational overhead.

[0155] As Figure 3 shown, the Gaussian mixture model established in step six specifically describes the probability density of the distribution of each feature in the strong correlation feature set F of thermal runaway risk with p(x|θ); among them, x represents the observed value of the statistical sample, θ represents the distribution that x follows, and is composed of g cluster Gaussian components; the number of clusters g is specifically determined based on the AIC and BIC criteria, and the parameters of θ are obtained by iterative expectation maximization method; s The Gaussian mixture model aims at the statistical features of the voltage sequences of each single battery in the k-th charging cycle sequence input arbitrarily, and obtains specific distribution parameters through iterative expectation maximization method: calculate the probability that the observed value x

[0156] comes from the i-th cluster Gaussian component, so that the logarithm of the probability likelihood i reaches the maximum, and iterative calculation makes the algorithm converge, and finally obtains each Gaussian distribution θ of the probability density; among them, x represents the observed value of the statistical sample, θ represents the distribution that x follows, and is composed of g cluster Gaussian components; the number of clusters g is specifically determined based on the AIC and BIC criteria, and the parameters of θ are obtained by iterative expectation maximization method; k ​and its corresponding weight α k ; The r-th single cell with the maximum logarithmic likelihood of the characteristic samples of all single cells is used as the reference single cell, and its strongly associated feature set of various thermal runaway risks is expressed as:

[0157] F sr =[f sr1 ,f sr2 ,...,f srw

[0158]

[0159] The strongly associated features of various thermal runaway risks of other single cells are expressed as:

[0160] F si =[f si1 ,f si2 ,...,f siw

[0161] Calculate the Euclidean distance between the samples of other single cells and the reference single cell sample:

[0162]

[0163] Perform the following normalization to obtain the risk feature distance within the range of [0,1]:

[0164]

[0165] d sk =[d sk1 ,d sk2 ,...,d skM

[0166] Traverse the K parking and charging cycles in the whole life cycle of the electric vehicle to be detected to obtain the following risk feature distance matrix D sk :

[0167]

[0168] The cumulative risk feature distance of the corresponding single cell i is calculated by the following formula:

[0169]

[0170] For different electric vehicle models using the same battery, the standardized cumulative risk feature distance can be further calculated through the following normalization to meet the needs of thermal runaway risk identification for different models:

[0171]

[0172] Figure 4 ​​​(a) and (b) show the evolution law of the cumulative risk characteristic distance, as well as the safety threshold and the standardized safety threshold respectively determined by comparing with normal monomers. Both thresholds can be used as quantitative indicators for identifying the thermal runaway risk of single cells. Figure 5 It shows the changes in the thermal runaway risk characteristic distance and the corresponding time-domain characteristics of the thermal runaway single-cell voltage during the 50th, 350th, 650th, 672nd, and 690th charging cycles of the power battery and the charging cycle when thermal runaway occurs.

[0173] The strong correlation data characteristics of the battery thermal runaway risk and the thermal runaway risk identification results in the present invention are extracted, calculated, and verified by real vehicle data, and can accurately quantify the thermal runaway risk of the power battery under real vehicle operating conditions. For in-service new energy vehicles, it can give early warnings about the potential thermal runaway risk of the power battery system; for vehicles that have experienced thermal runaway, it can timely identify the single cell with thermal runaway and trace back its risk evolution law. At the same time, the present invention only extracts and calculates the strong correlation data characteristics of the thermal runaway risk and calculates the thermal runaway risk characteristic distance for the single-cell voltage sequence of the entire charging cycle after each charging ends, without real-time calculation, and only needs to use single-cell voltage data, reducing the calculation load and data acquisition cost of the battery management system, and reducing the impact of noise in real vehicle data acquisition on the accuracy of the calculation results.

[0174] It should be understood that the magnitudes of the sequence numbers of the steps in the embodiments of the present invention do not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0175] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for identifying the risk of thermal runaway of a lithium-ion battery pack for an electric vehicle, characterized in that: Specifically, it includes the following steps: Step 1: Extract the single-cell voltage time-series data V of the last parking and charging cycle in the whole life cycle of several electric vehicles with single-cell thermal runaway, and add classification labels indicating safety or failure status to the single-cell voltage time series of each single cell to obtain a classification label set Y; Step 2: Use the Tsfresh automated feature engineering tool to extract multiple feature data of each single cell voltage time series to form a feature set F, and jointly construct it with the classification label set Y as the training set; use the training set to train the random forest model, and output the total information gain f obtained when splitting each feature with the classification label of the safe state in the decision tree, which is used to reflect the actual importance of each feature. a , which is used to reflect the actual importance of each feature. Step 3: After randomly shuffling the classification label set Y for k times, a new classification label set Y1 is obtained. Reconstruct the training set with it and the feature set F and retrain the random forest model, and output the total information gain f obtained when each feature is split with the shuffled classification labels, which is used to reflect the importance of each feature when the classification labels are shuffled; n , which is used to reflect the importance of each feature when the classification labels are shuffled; Step 4: Based on the feature importance obtained in Steps 2 and 3, perform feature screening to retain several strongly correlated features of thermal runaway risk; Step 5: Extract the single-cell voltage time series data of the last charging cycle before the accident caused by thermal runaway for the electric vehicle to be detected, and extract the feature sample set F composed of each strongly correlated eigenvalue s ; For the characteristic sample set F s Execute the clustering algorithm, and output the safety status or the classification label of the risk of thermal runaway corresponding to each monomer after clustering, which is used to automatically mark and locate the monomer battery with the risk of thermal runaway; Step 6. Establish a Gaussian mixture model for calculating the probability likelihood of the distribution of each feature in the feature sample set F s After inputting the single-cell voltage time series data of any k-th charging cycle in the whole life cycle of the electric vehicle to be detected into the Gaussian mixture model for calculation, set the single cell with the maximum log-likelihood of the feature samples of all single cells as the reference single cell battery; Calculate the Euclidean distance between the feature samples of other single cells and the reference single cell as the risk feature distance; Traverse the cumulative risk feature distances of each single cell in K parking and charging cycles in the whole life cycle of the electric vehicle to be detected, and set corresponding thresholds; Step 7: According to the change trends of the risk feature distance and the cumulative risk feature distance, judge the time when the thermal runaway risk single cell first occurs and the subsequent risk evolution law.

2. The method according to claim 1, characterized in that: The single-cell voltage time-series data V extracted in Step 1 is specifically in the following matrix form: In the formula, M is the number of single cells in the electric vehicle battery pack, T is the data length of the time series of this charging cycle, and the data acquisition frequency is 10 s / frame; Add a classification label y indicating the safe or faulty state to the time series of the single-cell voltage of each single cell i (i ∈ M), to obtain the following classification label set Y: Y = [y1, y2,..., y M T ​ If any single cell is a single cell in a thermal runaway accident, its label is 1; if it is a safe single cell, the label is 0.

3. The method according to claim 2, characterized in that: In Step 2, use the Tsfresh tool to specifically extract 785 time-frequency domain feature data of the voltage time series of each single cell to obtain the following feature set F: where f MN represents the Nth feature of the voltage time series of the Mth single cell; Based on the feature set F and the classification label set Y, jointly construct the following training set D for training the random forest model: In Step 3, randomly shuffle the classification label set Y to obtain a new classification label set Y1, and then construct a training set D1 = [F, Y1] for retraining the random forest model.

4. The method according to claim 3, characterized in that: In step 4, based on the information gain f a and f n and the following importance scoring formula is used to screen the strongly correlated features of thermal runaway risk: where percentile(f n , 0.75) represents the 75th percentile of f n ; Select and retain w feature items in the feature set F whose importance scores imp_s are greater than 0 as the strong correlation features f for thermal runaway risk s : f s = [f s1 , f s2 ,..., f sw .

5. The method according to claim 4, characterized in that: In step five, the following feature sample set F composed of each strongly correlated eigenvalue is extracted for the electric vehicle to be detected s : where f sMw represents the w-th strong correlation feature of thermal runaway risk in the voltage-time series data of the M-th single battery of the thermal runaway electric vehicle; For F s The following clustering algorithm model is specifically used for clustering: First, set the ε-neighborhood parameter range list and Minpts parameter of the density clustering model, traverse the ε-neighborhood parameter range list, and construct clustering models with different ε-neighborhood parameters; After that, input F s into clustering models with different ε-neighborhood parameters, and calculate the silhouette coefficient s of each model through the following formula: In the formula, s(i) represents the silhouette coefficient of the i-th single cell sample; a(i) represents the dissimilarity within the same clustering cluster, that is, the average of the dissimilarities of all samples in the cluster to other points in the same cluster; b(i) represents the dissimilarity between different clustering clusters, that is, the minimum of the average dissimilarity degrees of all samples in the same clustering cluster to other clusters; Select the clustering model with the highest silhouette coefficient and output the labeling result Y of the single cells after clustering cluster , as shown in the following formula: Y cluster = [y c1 , y c2 ,..., y cM T ​ where y ci represents the thermal runaway risk label of the i-th single battery of the electric vehicle. If y ci = -1, it means that the clustering algorithm marks this single battery as a thermally runaway risk single battery; if y ci = 0, it means that the clustering algorithm marks this single battery as a safe single battery.

6. The method according to claim 5, wherein: The Gaussian mixture model established in Step 6 specifically describes the probability density of the distribution of each feature in the strong correlation feature set F of the thermal runaway risk with p(x|θ); where x represents the observed value of the statistical sample, θ represents the distribution that x follows, which is composed of g clusters of Gaussian components; the number of clusters g is specifically determined based on the AIC and BIC criteria to obtain the optimal value, and the parameters of θ are obtained by iterative expectation maximization method; s in which, x represents the observed value of the statistical sample, θ represents the distribution that x follows, which is composed of g clusters of Gaussian components; the number of clusters g is specifically determined based on the AIC and BIC criteria to obtain the optimal value, and the parameters of θ are obtained by iterative expectation maximization method; The Gaussian mixture model iteratively obtains specific distribution parameters for the statistical characteristics of the voltage sequences of each single battery in the k-th charging cycle sequence for any input: calculate the observable x i The probability from the i-th cluster of Gaussian components, such that the probability likelihood The logarithm of reaches the maximum, and iteratively calculates to converge the algorithm, and finally obtains each Gaussian distribution θ k And its corresponding weight α k ; The r-th single battery with the maximum log-likelihood of the feature samples of all single batteries is used as the reference single battery, and its strong correlation feature set of each thermal runaway risk is expressed as: F sr = [f sr1 , f sr2 ,..., f srw ​ The strongly correlated features of thermal runaway risk for other single cells are expressed as: F si = [f si1 , f si2 ,..., f siw ​ Calculate the Euclidean distance between other single cell samples and the reference single cell sample: Perform the following normalization processing to obtain a risk feature distance in the range of [0, 1]: d sk = [d sk1 , d sk2 ,..., d skM ​ Traverse K parking and charging cycles in the entire life cycle of the electric vehicle to be detected to obtain the following risk characteristic distance matrix D sk : The cumulative risk feature distance of the corresponding single cell i is calculated through the following formula:

7. The method according to claim 6, wherein: For different electric vehicle models using the same battery, calculate the standardized cumulative risk feature distance through the following normalization processing to meet the needs of thermal runaway risk identification for different models:

Images