An intelligent identification method for monomer difference risk of new energy vehicle battery system

Abstract

A method for intelligently identifying the risk of single-cell differences in a new energy vehicle battery system. The method provides six data items, namely, SOC, charging state, highest voltage battery cell code, highest value of battery cell voltage, lowest voltage battery cell code, and lowest value of battery cell voltage, according to the national standard GB / T32960. By statistically analyzing the extreme difference and the battery cells where the extreme difference occurs, a difference matrix is ​​constructed to structure the difference characteristics; the extreme difference matrix is ​​accumulated within the observation window to eliminate the influence of outliers; the average value within the historical influence window is used to square it to highlight the abnormal changes in consistency; a safety risk signal is constructed to calculate the safety risk feature vector; a logistic regression model is constructed to realize intelligent identification of safety risks. The method of the present invention has the advantages of high real-time performance, simple data requirements, and the ability to realize the quantification of differences between batteries and intelligent identification of safety risks.

Description

Technical Field

[0001] The present invention relates to the technical field of power batteries, and in particular to a method for intelligently identifying risks of cell differences in a new energy vehicle battery system. Background Art

[0002] Most accidents involving new energy vehicles (electric vehicles), especially fire accidents, are caused by power battery failure. In practical applications, the internal status of the battery pack can be analyzed through various indicators monitored by the BMS (battery management system), thereby achieving safety warning and risk assessment for new energy vehicles.

[0003] In automotive applications, many independent battery cells are connected in series to form a battery group and a battery pack. Each battery cell is the smallest monitoring unit in the battery pack. The deterioration of the internal state of the battery pack is often manifested by the inconsistency of the performance between the battery cells. Therefore, the risk state of the battery pack can be analyzed by analyzing whether the performance of each battery cell is consistent. Since each battery cell is subject to process restrictions, the internal resistance, connection resistance, internal impurities, etc. of each battery are different, so the initial state of each battery cell when leaving the factory is different, but these inconsistencies are determined by their inherent physical and chemical properties. It can be considered that this inconsistency is stable and usually does not lead to accidents. At present, the commonly used methods for analyzing the consistency of battery cells in the battery pack include: the information entropy method for analyzing the instability of battery cell voltage switching, the internal resistance method for calculating the internal resistance based on the equivalent circuit model to identify the internal short circuit, and the feature analysis method based on a large number of examples for supervised machine learning to build an analysis model. However, these methods have defects such as low differentiation between accident vehicles and normal vehicles, high data quality requirements, and large amount of calculation. Summary of the invention

[0004] The purpose of the present invention is to address the deficiencies in the prior art and to provide a method for intelligently identifying the risks of cell differences in a new energy vehicle battery system. The method has high real-time performance, simple data requirements, and can achieve quantification of differences among batteries and intelligent identification of safety risks.

[0005] The technical solution of the present invention is: a method for intelligently identifying the risk of cell differences in a new energy vehicle battery system, and the specific steps are as follows:

[0006] 1) Data collection: real-time collection of six operating parameters of the vehicle i (battery level at time i), status i (charging and discharging status at time i), umax i (the highest voltage value at time i), umin i (the lowest voltage value at time i), smax i (the number of the single cell with the highest voltage at time i), smini (the lowest voltage cell number at time i), the above data are all time-varying data with time stamp;

[0007] 2) Structuring of battery cell inconsistency characteristics: defining the range matrix A m×n :

[0008] A m×n =(a ij )∈R m×n

[0009] in

[0010] In the formula, m represents the number of battery cells in the battery pack, n represents the number of valid data, and a ij represents the element in the i-th row and j-th column of the range matrix, R m×n Represents a real matrix with m rows and n columns;

[0011] 3) Quantification of battery cell inconsistency:

[0012] 3-1) According to historical data, set the filter f;

[0013] 3-2) Convolve the range matrix with the filter f to obtain the difference matrix D that reflects the current inconsistency of the battery pack m×n =cov(A,f), the calculation formula is as follows,

[0014]

[0015] Where, d ij represents the element in the i-th row and j-th column of the difference matrix, f i represents the i-th element of the filter, a i+1-k,j represents the element in the i+1-kth row and jth column of the range matrix, and l represents the length of the historical observation window;

[0016] 4) Identification of abnormal changes in battery cell inconsistency: Based on the mean and standard deviation of L data in the historical impact window, the difference matrix is ​​scaled to obtain the relative difference matrix D′. The calculation formula is as follows:

[0017]

[0018] Where d′ ij represents the element in the i-th row and j-th column of the relative difference matrix, D ij represents the difference matrix, x ij represents the mean of L data in the historical impact window, S ij Indicates the standard deviation of L data within the historical impact window;

[0019] 5) Construction of characteristic signal of inconsistent safety risk: According to the sampling frequency Fs, the safety risk signal of the i-th battery cell at the j-th moment is set to a ij , extract the mean μ of the safety risk signal ij , standard deviation σ ij , Kurtosis K ij , rectified average Form Factor Crest Factor Information entropy H(a ij ), and obtain its security risk feature α ij ,in::

[0020] a ij =(D′ i,j-600*Fs+1 ,D′ i,j-600*Fs+2 ,…,D′ i,j ),

[0021]

[0022] 6) Intelligent risk identification of battery system safety risks: Use the safety risk characteristics of the problematic battery cells of the accident vehicle 3 hours before the accident as the risk training set, and use the safety risk characteristics of the battery cells of normal vehicles and the normal telecommunications of the accident vehicle as the safety training set to train a nonlinear logistic regression classification model, thereby realizing intelligent risk identification of battery system safety risks.

[0023] Furthermore, in step 3-1), the filter f is a first-order half-Gaussian filter with a length of 2l and a standard deviation of σ=2l / 3.75, and the calculation formula is as follows:

[0024] f=(f1,f2,…,f l ),in

[0025] Where l is the set historical observation window length, which is used to observe the historical data within the observation window at a certain moment.

[0026] Furthermore, in step 4), the mean and standard deviation of L data in the historical impact window are calculated according to the following formula:

[0027]

[0028]

[0029] Where L represents the length of the historical impact window, Avg m×n represents the mean of L data in the historical impact window, S m×n Indicates the standard deviation of L data within the historical impact window.

[0030] Further, in step 5), the safety risk characteristics of the i-th battery cell at the j-th moment include mean, standard deviation, kurtosis, rectified mean, form factor, peak factor, and information entropy, which are calculated according to the following formula:

[0031]

[0032]

[0033]

[0034]

[0035]

[0036]

[0037]

[0038] In the formula, D′ represents the difference matrix, a ij Represents the safety risk signal of the i-th battery cell at the j-th moment.

[0039] Beneficial effects of adopting the above technical solution:

[0040] 1. This method has low data requirements: only national standard data can be used for calculation, and it does not require high data collection frequency, strict data continuity, or single cell voltage values ​​to achieve accurate safety warnings to the battery cell.

[0041] 2. This method can quantify the differences of battery cells: the process of constructing the difference matrix realizes the quantification of the inconsistency within the battery pack, and the size of each element in the matrix is ​​the quantified size of the inconsistency of each battery cell.

[0042] 3. This method is based on a statistical model and is simple to implement: no large amount of machine learning is required in the feature extraction stage. After mathematical calculations are performed on the original data in the production environment, it can be brought into the model for risk identification.

[0043] 4. This method can perform real-time calculations: when calculating the difference value at each moment, only the historical data of the historical observation window l and the historical influence window L before that moment are required.

[0044] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A broken line analysis diagram of a faulty vehicle in an embodiment of the method;

[0046] Figure 2This is a line analysis diagram of a normal vehicle in the embodiment of the method;

[0047] Figure 3 Flow chart of the steps for implementing this method. DETAILED DESCRIPTION

[0048] The inherent physical and chemical properties of battery packs produce stable inconsistencies in battery cells. The data characteristics of the parameters are usually manifested as voltage extremes appearing on a fixed number of battery cells. The magnitude of the range generally conforms to the Gaussian distribution. Analyzing the abnormal state inside the battery pack is to analyze the abnormal changes in the inconsistency of the battery cells in the battery pack, that is, to identify the new extreme value battery cells and the abnormal increase of the range that appear over time.

[0049] See also Figures 1 to 3 , a method for intelligent identification of single-cell difference risks in new energy vehicle battery systems, based on the six data items of SOC, charging state, highest voltage battery cell code, highest value of battery cell voltage, lowest voltage battery cell code, and lowest value of battery cell voltage specified in the national standard GB / T32960, by statistically analyzing the extreme differences and the battery cells where the extreme differences occur, a difference matrix is ​​constructed to structure the difference characteristics; the extreme difference matrix is ​​accumulated within the observation window to eliminate the influence of outliers; the average value within the historical influence window is used to square it to highlight the abnormal changes in consistency; a safety risk signal is constructed to calculate the safety risk feature vector; a logistic regression model is constructed to realize intelligent identification of safety risks.

[0050] The specific steps are as follows:

[0051] 1) Data collection: real-time collection of six operating parameters of the vehicle i (battery level at time i), status i (charging and discharging status at time i), umax i (the highest voltage value at time i), umin i (the lowest voltage value at time i), smax i (the highest voltage cell number at time i), smin i (the lowest voltage cell number at time i). The above data are all time-varying data with time stamps.

[0052] 2) Structuring of battery cell inconsistency characteristics: defining the range matrix A m×n :

[0053] A m×n =(a ij )∈R m×n

[0054] in

[0055] In the formula, m represents the number of battery cells in the battery pack, n represents the number of valid data, and a ij represents the element in the i-th row and j-th column of the range matrix, R m×n Represents a real number.

[0056] 3) Quantification of battery cell inconsistency:

[0057] 3-1) Set the historical observation window length l, which is used to observe the historical data in the observation window when the observation mode is used. Construct a first-order half-Gaussian filter with a length of 2l and a standard deviation of σ=2l / 3.75. The calculation formula is as follows:

[0058] f=(f1,f2,…,f l ),in

[0059] Because it is necessary to refer to the values ​​of each moment and the l-1 moments before it to construct the difference matrix, and the closer to the current moment, the higher the weight, so a Gaussian filter with a length of 2l is constructed and its left half is taken as the weight;

[0060] 3-2) Convolve the range matrix with the filter f to obtain the difference matrix D that reflects the current inconsistency of the battery pack m×n =cov(A,f), the calculation formula is as follows,

[0061]

[0062] Where, d ij represents the element in the i-th row and j-th column of the difference matrix, f i represents the i-th element of the filter, a i+1-k,j represents the element in the i+1-kth row and jth column of the range matrix, and l represents the length of the historical observation window.

[0063] 4) Identification of abnormal changes in battery cell inconsistency:

[0064] 4-1) Set the length of the historical impact window L, and calculate the average and standard deviation of L valid data in the window. The calculation formula is as follows:

[0065]

[0066]

[0067] Where L represents the length of the historical impact window, Avg m×n represents the mean of L data in the historical impact window, d i+1-k,j represents the element in the i+1-kth row and jth column of the difference matrix, S m×n represents the standard deviation of L data in the historical impact window, σ ijThe element in row i and column j represents the standard deviation;

[0068] 4-2) Based on the mean and standard deviation of L data in the historical impact window, the difference matrix is ​​scaled. The corresponding mean is subtracted from the value of each difference matrix, and then 3 times the standard deviation is added to obtain the relative difference matrix D′. The calculation formula is as follows:

[0069]

[0070] Where d′ ij represents the element in the i-th row and j-th column of the relative difference matrix, d ij represents the difference matrix, x ij represents the mean of L data in the historical impact window, S ij Indicates the standard deviation of L data within the historical impact window.

[0071] 5) Construction of characteristic signal of inconsistent safety risk: For the ith battery cell, the safety risk signal at the jth moment is set as α according to the data sampling frequency. ij , calculate the signal characteristics of the security risk signal;

[0072] According to statistical laws, the cause of most fire accidents is the internal short circuit of individual battery cells, which is easy to occur under the two working conditions of discharging at the SOC and charging at a high SOC. Therefore, combining the signal characteristics of the safety risk signal of the battery cell at this time with the SOC and the charging and discharging state, its safety risk feature vector α is obtained. ij .

[0073] 6) Intelligent risk identification of battery system safety risks: Use the safety risk characteristics of the problematic battery cells of the accident vehicle 3 hours before the accident as the risk training set, and use the safety risk feature vectors of the battery cells of normal vehicles and the normal battery cells of the accident vehicle as the safety training set to train the logistic regression classification model, thereby realizing intelligent risk identification of battery system safety risks.

[0074] Example 1: Intelligent identification and analysis of cell difference risks in new energy vehicle battery systems is as follows:

[0075] 1. Obtain the data of the whole life cycle of a certain type of new energy vehicle that has caught fire and a normal vehicle. It is known that the vehicle has 120 battery cells and the data acquisition frequency Fs is 0.1Hz. The operating parameters obtained include: the battery power soc at time i i , the charging and discharging status at time i i , the highest voltage value umax at time i i , the lowest voltage value umin at time i i , the highest voltage cell number at time i is smax i, the lowest voltage cell number smin at time i i ;

[0076] 2. Construct the range matrix A m×n ;

[0077] 3. According to experience, the observation window is set to 10 minutes, so l = 10*60*Fs = 60, and the filter f is obtained;

[0078] 4. Based on A m×n and Fs to get the difference matrix D m×n ;

[0079] 5. Based on experience, the historical impact window is set to 7 days. The vehicle is used for about 2 hours a day, so L = 7*2*60*60*Fs = 5040 is set, and the mean matrix Avg is calculated;

[0080] 6. The relative difference matrix D′ is calculated according to the formula, and a line chart is drawn based on D′. The horizontal axis of the line chart represents time (Unix timestamp), and the vertical axis is the value of the difference matrix D′. Each line corresponds to a battery cell in the battery pack. The accident vehicle and the normal vehicle are analyzed separately.

[0081] like Figure 1 As shown: The difference between the No. 45 and No. 74 battery cells of the vehicle increased abnormally before the accident, such as Figure 2 As shown, the inconsistency of the car is relatively stable, there is no sudden change, and the difference value is small. Such a difference can be expressed by calculating the statistical characteristics and informatics characteristics of the signal;

[0082] 7. Based on 30*60*Fs=180 relative difference values ​​of each battery cell within 30 minutes at each moment, a safety risk signal is constructed, and the mean, standard deviation, kurtosis, rectified mean, waveform factor, peak factor, and information entropy of the signal are calculated according to the formula, and the safety risk feature vector is obtained by combining the SOC and charge and discharge status at the corresponding moment;

[0083] 8. Use the safety risk characteristics of problematic cells of multiple accident vehicles of the same model 3 hours before the accident as the risk training set, and use the safety risk feature vectors of cells of multiple normal vehicles of the same model and the normal cells of the accident vehicle as the safety training set to train the logistic regression classification model, and further realize the intelligent risk identification of battery system safety risks based on the obtained model.

Claims

1. A method for intelligently identifying the risk of cell differences in a new energy vehicle battery system, characterized in that The specific steps are as follows: 1) Data collection: real-time collection of six operating parameters of the vehicle, battery power soc at time i i , the charging and discharging status at time i i , the highest voltage value umax at time i i , the lowest voltage value umin at time i i , the highest voltage cell number at time i is smax i , the lowest voltage cell number smin at time i i ,The above data are all time-varying data with time scale; 2) Structuring of battery cell inconsistency characteristics: defining the range matrix A m×n : A m×n =(a ij )∈R m×n in In the formula, m represents the number of battery cells in the battery pack, n represents the number of valid data, and a ij represents the element in the i-th row and j-th column of the range matrix, R m×n Represents a real matrix with m rows and n columns; 3) Quantification of battery cell inconsistency: 3-1) According to the historical observation window l, set the filter f; 3-2) Convolve the range matrix with the filter f to obtain the difference matrix D that reflects the current inconsistency of the battery pack m×n =cov(A,f), the calculation formula is as follows, Where, d ij represents the element in the i-th row and j-th column of the difference matrix, f i represents the i-th element of the filter, a i+1-k,j represents the element in the i+1-kth row and jth column of the range matrix, and l represents the length of the historical observation window; 4) Identification of abnormal changes in battery cell inconsistency: Based on the mean and standard deviation of L data in the historical impact window, the difference matrix is ​​scaled to obtain the relative difference matrix D′. The calculation formula is as follows: Where, d i ' j represents the element in the i-th row and j-th column of the relative difference matrix, D ij represents the difference matrix, x ij represents the mean of L data in the historical impact window, S ij Indicates the standard deviation of L data within the historical impact window; 5) Construction of characteristic signal of inconsistent safety risk: According to the sampling frequency Fs, the safety risk signal of the i-th battery cell at the j-th moment is set to a ij , extract the mean μ of the safety risk signal ij , standard deviation σ ij , Kurtosis K ij , rectified average value X arvij , Form Factor Peak factor k aij 、Information entropy H(a ij ), combined with the SOC and charge / discharge state at this time, the safety risk characteristic vector α is obtained ij ,in: a ij =(D′ i,j-600*Fs+1 ,D′ i,j-600*Fs+2 ,…,D′ i,j ), 6) Intelligent identification of battery system safety risks: Use the safety risk characteristics of the problematic battery cells of the accident vehicle 3 hours before the accident as the risk training set, and use the safety risk characteristics of the battery cells of a normal vehicle and the normal battery cells of the accident vehicle as the safety training set to train a logistic regression classification model, thereby realizing intelligent identification of battery system safety risks.

2. The method for intelligently identifying the risk of cell differences in a new energy vehicle battery system according to claim 1 is characterized in that: In step 3-1), the filter f is a first-order half-Gaussian filter with a length of 2l and a standard deviation of σ=2l / 3.

75. The calculation formula is as follows: f=(f1,f2,…,f l ),in Where l is the set historical observation window length, which is used to observe the historical data within the observation window at a certain moment.

3. The method for intelligently identifying monomer difference risks of a new energy vehicle battery system according to claim 1 is characterized in that: In step 4), the mean and standard deviation of L data in the historical impact window are calculated according to the following formula: Where L represents the length of the historical impact window, Avg m×n represents the mean of L data in the historical impact window, S m×n Indicates the standard deviation of L data within the historical impact window.

Images