A method for evaluating dynamic inconsistency and state of health of lithium-ion battery pack
By employing an inconsistency-based SOH assessment method, utilizing DTW and DBSCAN algorithms to evaluate differences between individual battery cells, and combining kernel functions to assess similarity distribution, the accuracy of battery capacity assessment and external environmental interference issues in vehicle power battery systems are resolved, achieving efficient battery health status assessment under low sampling rate data.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HARBIN INST OF TECH AT WEIHAI
- Filing Date
- 2023-11-09
- Publication Date
- 2026-05-29
AI Technical Summary
Existing technologies struggle to accurately and quickly assess battery capacity changes in vehicle power battery systems, and traditional algorithms are ill-suited to meet the cloud-based assessment requirements for low-sampling-rate data, while external environmental factors have a significant impact.
A non-inconsistency-based SOH evaluation method is adopted. The differences between battery cells are evaluated by dynamic time warping (DTW) and density-based clustering (DBSCAN) algorithm. The probability density function of similarity distribution is evaluated by kernel function, and density clustering algorithm is introduced to remove outliers, reduce external interference and data sampling accuracy requirements.
It improves the accuracy and robustness of the algorithm, reduces computational costs, and can accurately assess battery health status in real vehicle and cloud data, adapting to abnormal data and variable sampling frequencies.
Smart Images
Figure CN117471326B_ABST
Abstract
Description
Technical Field
[0001] This invention mainly relates to the field of power battery management technology, specifically a method for assessing the dynamic inconsistency and health status of lithium-ion battery packs. Background Technology
[0002] For power batteries, especially vehicle-mounted power battery systems, most mainstream SOH estimation algorithms currently assess battery capacity changes as an indicator of SOH change. However, in actual vehicle use, battery capacity is difficult to assess accurately and quickly, and is affected by external environmental factors (such as temperature), making it difficult to meet the application requirements of real-world conditions. Meanwhile, with the establishment of new energy vehicle monitoring platforms, a large amount of real-time data is uploaded to the cloud; however, traditional algorithms struggle to estimate SOH based on low-sampling-rate data, failing to meet the needs of cloud-based assessment. Summary of the Invention
[0003] To address the shortcomings of current technologies, this invention, based on existing technologies and practical applications, provides a method for assessing the dynamic inconsistency and state of health (SOH) of lithium-ion battery packs. Compared to capacity-based SOH assessments, which are susceptible to interference from external environmental factors and require high data sampling rates, inconsistency-based SOH evaluations determine the degradation trend of the battery pack's SOH by assessing changes in differences between individual battery cells. This method is less affected by external interference, has lower requirements for sampling accuracy, is more robust to abnormal data, and is more adaptable to both real-vehicle data and cloud data.
[0004] Meanwhile, in response to the characteristics of numerous outliers and variable sampling frequencies in real vehicle data and cloud data, a comprehensive outlier processing mechanism was established. The dynamic time warping algorithm was used to avoid errors caused by data sampling misalignment, and the similarity was accurately assessed. The kernel function was used to evaluate the probability density function of the similarity distribution, and the density clustering algorithm was introduced to remove outliers, which greatly improved the accuracy and reliability of the algorithm while reducing the computational cost.
[0005] The specific technical solution of the present invention is as follows:
[0006] A method for assessing the dynamic inconsistency and health status of lithium-ion battery packs, wherein SOH represents the battery health status, DTW represents the dynamic time warping algorithm, and DBSCAN represents the density-based clustering algorithm, characterized by the following steps:
[0007] Step 1: Data Preprocessing
[0008] The abnormal data collected is processed. The data includes individual cell voltage, SOC and mileage. Then the data is segmented. The continuously collected data is divided into segments. After segmentation, the segments of each charge and discharge of the vehicle are obtained. The average value of the individual cell voltage curve of each segment is taken to obtain the average voltage curve.
[0009] Step 2: Similarity Assessment
[0010] After obtaining the segmented individual voltage data and average voltage data, DTW is used to evaluate the similarity between the individual voltage and the average voltage to obtain the similarity between each individual and the average voltage of the segment.
[0011] Step 3: Probability density estimation
[0012] After obtaining the similarity, the probability density distribution is estimated from the similarity of each individual unit in each segment, and the probability density is estimated using a kernel function to obtain the probability density function.
[0013] Step 4: Obtaining the SOH index
[0014] After obtaining the probability density function, the inconsistency index is calculated for each segment. After removing outliers using DBSCAN, the data is fitted to obtain the final SOH change curve.
[0015] Furthermore, in step 1, abnormal individual voltage data includes two categories: data transmission failure and sensor failure.
[0016] The transmission fault detection method first calculates the standard deviation of the individual voltage data for all individual cells. Data that has changed from the previous time step and whose standard deviation is less than 1 / 10 of the average value is considered abnormal data, and linear interpolation is used to handle it.
[0017] The sensor fault detection method involves identifying voltage spikes greater than 0.3V and voltage differences greater than 0.2V between the spike and the average voltage of the individual sensor at that moment. The processing method is linear interpolation.
[0018] The SOC data anomaly detection method is to identify an outlier if the SOC jump value is greater than 3, and the processing method is linear interpolation;
[0019] The main manifestation of mileage data anomalies is SOC jumps. The judgment method is that an SOC jump greater than 3 is considered an outlier, and the handling method is linear interpolation.
[0020] Furthermore, in step 1, after abnormal data processing, data segmentation is performed. The continuously collected data is segmented into segments, and each use is recorded as a segment. Based on the charging status indicator, the data is first divided into charging data and discharging data. For charging data, data with the same mileage are identified as the same charging segment and segmented accordingly.
[0021] For discharge data, the SOC abrupt change point is used as the dividing point to segment the data, and the segmentation yields the segments of each charge and discharge cycle of the vehicle.
[0022] Finally, the average voltage curve is obtained by averaging the individual voltage curves of each segment.
[0023] Furthermore, in step 2, the similarity evaluation method specifically includes:
[0024] Step 21: Initialize the distance matrix D. First, select a set of voltage segments, and then select the voltage curve of a single unit (containing x data points) V. c =[V c0 V c1 ,…,V cn (n = 0, 1, ..., x-1) represents the voltage value of this single cell at each point in this segment. The average voltage curve of this segment (containing y data points) is V. m =[V m0 V m1 ,…,V mn (n = 0, 1, ..., y-1) represents the average voltage of each individual point in this segment. Then, the distance matrix D is initialized:
[0025]
[0026] In the formula: the number of rows and columns of matrix D is x and y (usually equal), and the initial value of the matrix is infinity;
[0027] Step 22: Calculate local distance
[0028] First, let matrix D[0,0] = 0, then calculate the remaining contents of the matrix using the following formula:
[0029] The formula for calculating the leftmost column of matrix D is:
[0030] D[i,0]=|V ci -V m0 |+D[i-1,0] (2)
[0031] Where: i = 1, 2, ..., x.
[0032] The formula for calculating the bottom row of matrix D is:
[0033] D[0,j]=|V c0 -V mj |+D[0,j-1] (3)
[0034] Where: j = 1, 2, ..., y.
[0035] The formulas for calculating the values of other positions in the matrix are as follows:
[0036] D[i,j]=|V ci -V mj |+min(D[i-1,j],D[i,j-1],D[i-1,j-1]) (4)
[0037] Where: i = 1, 2, ..., x, j = 1, 2, ..., y.
[0038] Finally, the local distance matrix is obtained;
[0039] Step 23: Calculate the globally optimal path
[0040] Starting from the top right corner of matrix D, search for the optimal path towards the bottom left corner. Let the top right corner be D[x,y]. Starting from this point, select the minimum value among the three nearest neighbors to the bottom left as the starting point for the next step, and continue backtracking until the bottom left corner of the matrix ends. The line connecting all the selected values in this process is V. c With V m Align the path;
[0041] Step 24: Align the curves
[0042] Obtain the aligned V based on the alignment path obtained in step 23. c_new and V m_new ;
[0043] Step 25: Calculate similarity
[0044] V c With V m The similarity of the aligned V c_new and V m_new The sum of Euclidean distances is used for evaluation, and the calculation formula is as shown in equation (5):
[0045]
[0046] The similarity evaluation of the remaining segments and individual units is carried out in the same way as above, and finally the similarity set S between each individual unit and the average voltage of the segment is obtained. line .
[0047] Furthermore, in step 3, the method for obtaining the probability density function is as follows:
[0048] Step 31: Determine the bandwidth
[0049] Using the Scott criterion for bandwidth estimation, the standard deviation of similarity S is first calculated, as shown in equation (6):
[0050]
[0051] Where σ is the standard deviation, and N represents S. line The number of data points in S, μ represents the number of data points in S. line The mean,
[0052] The bandwidth is calculated using the Scott criterion, as shown in equation (7):
[0053]
[0054] Step 32: Select kernel function
[0055] The Gaussian kernel function is chosen as the kernel function for probability density estimation. The formula for the Gaussian kernel function is shown in equation (8):
[0056]
[0057] Step 33: Estimate kernel function density
[0058] First, apply a kernel function to each data point for a given dataset S. line Density estimation in the vicinity is achieved by applying the selected kernel function to the point.
[0059] Step 33: Obtain the probability density function
[0060] After obtaining the kernel function results for each point, the results are weighted and summed to estimate the probability density function, as shown in equation (9):
[0061]
[0062] Furthermore, in step 4, the area under the probability density curve within 10% to the left and right of the highest point of the probability density distribution function is selected as the probability P of the high-density interval. The quotient of the probability P of the high-density interval and the distribution range S of the probability density function yields the SOH index. By calculating the inconsistency index for each segment, the trend of SOH variation can be obtained.
[0063] Furthermore, in step 4, spline fitting is selected to obtain the SOH index curve. Spline fitting generates a smooth overall curve by connecting multiple local interpolation parts, and finally obtains the SOH curve based on inconsistency.
[0064] The beneficial effects of this invention are:
[0065] 1. This invention uses a dynamic time warping algorithm to evaluate the similarity of individual voltage curves, reducing the impact of data misalignment caused by acquisition equipment errors and improving the accuracy of the algorithm when mixed with misaligned data.
[0066] 2. Compared with traditional algorithms that can only use charging data to assess capacity and obtain health status, the proposed algorithm uses inconsistency as an indicator, which can make full use of charging and discharging data, greatly improve data utilization, and improve algorithm reliability.
[0067] 3. Compared with traditional algorithms, which have high requirements for data accuracy, sampling precision and sampling frequency, the proposed algorithm has lower data requirements and has good accuracy and robustness for both real vehicle data and cloud data. Attached Figure Description
[0068] Figure 1Flowchart of a battery health status assessment method based on inconsistencies;
[0069] Figure 2 Diagram of outlier handling methods;
[0070] Figure 3 Flowchart of Dynamic Time Warping (DTW) algorithm;
[0071] Figure 4 Flowchart for extracting inconsistency SOH assessment indicators;
[0072] Figure 5 Original results based on SOH assessment results due to inconsistencies;
[0073] Figure 6 SOH assessment results after removing outliers;
[0074] Figure 7 SOH curve after spline fitting;
[0075] Figure 8 Result comparison chart. Detailed Implementation
[0076] The present invention will be further described in conjunction with the accompanying drawings and specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined in this application.
[0077] This embodiment provides a method for assessing the dynamic inconsistency and health status of lithium-ion battery packs. The power battery system mainly includes known power battery cells, power battery packs, or assembled power batteries.
[0078] In this embodiment, SOH represents the battery health state, DTW represents the dynamic time warping algorithm, DBSCAN represents the density-based clustering algorithm, and the inconsistency-based battery health state assessment method is attached. Figure 1 As shown, this evaluation method is also applicable to nonlinear systems. The method mainly includes four aspects: data preprocessing, similarity evaluation, probability density estimation, and SOH index acquisition. The above four aspects will be described in detail below.
[0079] 1. Data Preprocessing
[0080] Compared to laboratory data, real-vehicle data and cloud data often contain a large amount of abnormal and missing data, and the data sampling frequency is uncertain, requiring data preprocessing. Taking cloud data as an example, its sampling rate is approximately 0.1Hz, with some data sampling frequencies fluctuating, and there are issues such as data anomalies and missing data. The data format conforms to the "Technical Specification for Remote Service and Management System of Electric Vehicles" (GB / T 32960-2016). The specific processing procedure is as follows:
[0081] First, abnormal data processing is performed. The data involved in this embodiment mainly includes individual cell voltage, SOC, and mileage. Therefore, abnormal data processing is performed on these three items.
[0082] Data anomaly handling process and examples are as follows: Figure 2 The abnormal individual voltage data shown can be categorized into two types: data transmission failures and sensor failures. Transmission failures manifest as abnormal data changes with individual voltages converging at a single point. The method for identifying these failures is to first calculate the standard deviation of all individual voltage data. Data that shows a change from the previous time point and whose standard deviation is less than 1 / 10 of the average is considered abnormal. The mean of the data before and after the abnormal point is used as the value at that point, i.e., linear interpolation. Sensor failures mainly manifest as abnormal jumps in individual individual voltages. The method for identifying these failures is to identify voltage jumps greater than 0.3V with a difference greater than 0.2V from the average voltage of the individual cells at that time. The processing method is linear interpolation.
[0083] The main manifestation of SOC data anomalies is SOC jumps. The judgment method is that an SOC jump greater than 3 is considered an outlier, and the handling method is linear interpolation.
[0084] The main manifestation of mileage data anomalies is SOC jumps. The judgment method is that an SOC jump greater than 3 is considered an outlier, and the handling method is linear interpolation.
[0085] Next, data segmentation is performed, dividing continuously collected data into segments based on usage, with each single use counted as a segment. Based on the charging status indicator, the data is first divided into charging data and discharging data. For charging data, data with the same mileage are considered as part of the same charging segment and segmented accordingly; for discharging data, the point of SOC abrupt change is used as the dividing point for segmentation. After segmentation, segments representing each charge and discharge cycle of the vehicle are obtained.
[0086] Finally, the average voltage curve is obtained by averaging the individual voltage curves of each segment.
[0087] 2. Similarity assessment
[0088] After obtaining the segmented individual voltage data and average voltage data V m Then, DTW was used to evaluate the similarity between the individual cell voltage and the average voltage, taking the voltage V of a certain individual cell in a certain segment as an example. c For example, the process is as follows: Figure 3 As shown, the detailed steps are as follows:
[0089] Step ①: Initialize the distance matrix D
[0090] First, select a set of voltage segments, and then select the voltage curve of a single unit within that segment (containing x data points). c =[V c0 V c1 ,…,V cn (n = 0, 1, ..., x-1) represents the voltage value of this single cell at each point in this segment. The average voltage curve of this segment (containing y data points) is V. m =[V m0 V m1 ,…,V mn (n = 0, 1, ..., y-1) represents the average voltage of each individual point in this segment. Then, the distance matrix D is initialized:
[0091]
[0092] In the formula: the number of rows and columns of matrix D are x and y (usually equal), and the initial value of the matrix is infinity.
[0093] Step 2: Calculate local distance
[0094] First, let matrix D[0,0] = 0, then calculate the remaining contents of the matrix using the following formula:
[0095] The formula for calculating the leftmost column of matrix D is:
[0096] D[i,0]=|V ci -V m0 |+D[i-1,0] (2)
[0097] Where: i = 1, 2, ..., x.
[0098] The formula for calculating the bottom row of matrix D is:
[0099] D[0,j]=|V c0 -V mj |+D[0,j-1] (3)
[0100] Where: j = 1, 2, ..., y.
[0101] The formulas for calculating the values of other positions in the matrix are as follows:
[0102] D[i,j]=|V ci -V mj |+min(D[i-1,j],D[i,j-1],D[i-1,j-1]) (4)
[0103] Where: i = 1, 2, ..., x, j = 1, 2, ..., y.
[0104] Finally, the local distance matrix is obtained.
[0105] Step 3: Calculate the globally optimal path
[0106] Starting from the top right corner of matrix D, search for the optimal path towards the bottom left corner. Let the top right corner be D[x,y]. Starting from this point, select the minimum value among the three nearest values to the bottom left (i.e., D[x-1,y], D[x,y-1], and D[x-1,y-1]) as the starting point for the next step, and continue backtracking until the bottom left corner of the matrix ends. The line connecting all the selected values in this process is V. c With V m Alignment path.
[0107] Step 4: Align the curves
[0108] Obtain the aligned V based on the alignment path obtained in step ③. c_new and V m_new .
[0109] Step 5: Calculate similarity
[0110] V c With V m Similarity in V c_new and V m_new The sum of Euclidean distances is used for evaluation, and its calculation formula is shown in equation (5):
[0111]
[0112] The similarity evaluation of the remaining segments and individual units is carried out in the same manner as above, and finally the similarity between each individual unit and the average voltage of the segment is obtained. If the data contains 96 individual units, then S line =[S line1 ,S line2 ,S line3 ,…,S line96 ].
[0113] 3. Probability density estimation
[0114] After obtaining the similarity, it is necessary to estimate the probability density distribution from the similarity of each individual element in each segment in order to extract the inconsistency features. This embodiment chooses to use a kernel function for probability density estimation to obtain the probability density function. The detailed steps are as follows:
[0115] Step 1: Determine the bandwidth
[0116] To ensure that the probability density function evaluated by the kernel function retains sufficient detail and is relatively smooth, while also removing most noise and fluctuations, an appropriate bandwidth needs to be selected. This embodiment uses Scott's rule for bandwidth estimation, which has the advantages of being simple, universal, and interpretable.
[0117] First, calculate the similarity S. line The standard deviation is shown in equation (6):
[0118]
[0119] Where σ is the standard deviation, and N represents S. line The number of data points in S, μ represents the number of data points in S. line The mean.
[0120] Then the bandwidth is calculated using the Scott criterion, as shown in equation (7):
[0121]
[0122] Step 2: Select the kernel function
[0123] The Gaussian kernel function was chosen as the kernel function for probability density estimation. The Gaussian kernel function performs well in nonparametric estimation, is suitable for various data distributions, and has good smoothness and bandwidth adjustment capabilities. The formula for the Gaussian kernel function is shown in equation (8):
[0124]
[0125] The kernel function takes a given point S as its base. line (i) Taking the center as the center, calculate S line (i) The contribution of nearby points, used to evaluate the contribution of nearby points to S line (i) The importance of the probability density at point S, usually due to the distance from S line (i) Points that are closer are of greater importance.
[0126] Step 3: Estimate kernel function density
[0127] First, apply a kernel function to each data point. For a given data point S... line (i) The density estimate of the vicinity is achieved by applying the selected kernel function to the point.
[0128] Step 4: Obtain the probability density function
[0129] After obtaining the kernel function results for each point, the results are weighted and summed to estimate the probability density function, as shown in equation (9):
[0130]
[0131] Finally, the probability density function of the segment can be obtained.
[0132] 4. Obtaining the SOH index
[0133] After obtaining the probability density function, this invention extracts features from it to characterize SOH, the process of which is as follows: Figure 4 As shown.
[0134] The probability density function distribution range S represents the size of the distribution interval of the similarity of each individual in the segment. The smaller this index, the denser the distribution. The area under the probability density curve within 10% to the left and right of the highest point of the probability density distribution function (i.e., the probability) is selected as the probability P of the high-density interval. The larger this index, the denser the distribution. The former is affected by the overall sample population, while the latter focuses on the curve of dense distribution. The SOH index is obtained by dividing the probability P of the high-density interval by the distribution range S. The trend of SOH can be obtained by calculating the inconsistency index for each segment.
[0135] The assessed SoH (Solar OH) still contains some outliers, which are scattered and the number of clusters is uncertain. To remove the outliers generated during the assessment process, DBSCAN is applied. This algorithm can discover clusters from datasets of arbitrary shapes and sizes, identifying clusters of arbitrary shapes more accurately and efficiently. After removing outliers, the data can be fitted to obtain the final SoH variation curve.
[0136] This embodiment selects spline fitting to obtain the SOH index curve. Spline fitting generates a smooth overall curve by connecting multiple local interpolation parts (usually low-order polynomials). Compared to conventional polynomial fitting, it is more suitable for data without clear patterns of change and can fully reflect the changes in data trends. The interpolation function is a low-order polynomial in each local interval to ensure smoothness, continuity, and local approximation. Connecting the interpolation functions obtained in each local interval constructs a smooth spline-fitted curve. The final SOH curve based on inconsistency is obtained. In practical applications, since the index obtained by direct calculation is relatively small and easily affected by insufficient computer storage precision, the quotient of S and P can be calculated first when calculating the SOH index, outlier handling can be performed, spline fitting can be performed, and then the reciprocal can be taken to avoid errors caused by insufficient computer storage precision.
[0137] This invention also validates the evaluation method to ensure the effectiveness of the method proposed in this embodiment. Taking cloud data of nine new energy vehicles over the past year as an example, the data collection fields are based on the requirements of the national standard GB / T 32960, the vehicle models cover two typical regions, Northeast and South China, and the vehicle characteristics cover typical user usage characteristics such as long-term storage followed by driving, continuous driving, and driving followed by storage.
[0138] Figure 5The original results of the SOH assessment based on inconsistencies are presented. Figure 6 The results after outlier handling are shown. Figure 7 The final SOH curve obtained after spline fitting is shown. Figure 8 The results show a comparison between the SOH assessment results based on inconsistency and the capacity assessment results, where the results of the inconsistency assessment were subjected to a simple linear transformation to better compare with the capacity values.
[0139] from Figure 8 It can be seen that the SOH assessment method based on inconsistency achieves results similar to capacity assessment without extracting capacity information, proving that it has a good assessment capability for battery SOH. The more complete the data, the better the assessment effect.
[0140] As can be seen from the above embodiments and results, the SOH estimation method based on inconsistency proposed in this invention has the following advantages compared with traditional methods:
[0141] (1) It does not rely on high-precision, fixed-sampling-rate charging data, so it has good effect on data under real vehicle and cloud conditions. At the same time, the use of discharge data improves data utilization and adaptability.
[0142] (2) Through multiple measures such as data cleaning, application of dynamic time warping algorithm, and kernel function probability density evaluation, the interference of abnormal data on the algorithm is reduced, and the algorithm has strong robustness.
[0143] (3) By assessing the inconsistency of the battery pack, the SOH is assessed, which not only reflects the decrease in SOH caused by the decrease in battery capacity, but also significantly reduces the required data accuracy and requirements while achieving an effect similar to the SOH assessment method based on capacity estimation. It also reflects the decrease in battery electrical performance caused by the change in inconsistency.
[0144] (4) Compared with capacity assessment, the computational cost of this algorithm is relatively low, which greatly reduces the computational cost and computation time.
Claims
1. A method for assessing the dynamic inconsistency and health status of a lithium-ion battery pack, wherein the method uses SOH to represent the battery health status, DTW to represent the dynamic time warping algorithm, and DBSCAN to represent the density-based clustering algorithm, characterized in that... Includes the following steps: Step 1: Data Preprocessing The collected abnormal data, including cell voltage, SOC, and mileage, is processed. The continuously collected data is then segmented into segments, resulting in segments for each charge / discharge cycle of the vehicle. The average voltage curve of each segment is calculated by averaging the cell voltage curves. Abnormal cell voltage data includes two categories: data transmission failures and sensor failures. After processing, the continuously collected data is segmented into segments, with each use counted as one segment. Based on the charging status indicator, the data is first divided into charging data and discharging data. For charging data, data with the same mileage are considered as belonging to the same charging segment and segmented accordingly. For discharge data, the SOC abrupt change point is used as the dividing point to segment the data, and the segmentation yields the segments of each charge and discharge cycle of the vehicle. Finally, the average voltage curve of each segment is obtained by averaging the individual voltage curves. Step 2: Similarity Assessment After obtaining the segmented individual voltage data and average voltage data, DTW is used to evaluate the similarity between the individual voltage and the average voltage to obtain the similarity between each individual and the average voltage of the segment. Step 3: Probability density estimation After obtaining the similarity, the probability density distribution is estimated from the similarity of each individual unit in each segment. The kernel function is used to estimate the probability density, and the probability density function is obtained. The distribution range of the probability density function represents the size of the distribution interval of the similarity of each individual unit in the segment. Step 4: Obtaining the SOH index After obtaining the probability density function, an inconsistency index is calculated for each segment. After removing outliers using DBSCAN, the data is fitted to obtain the final SOH change curve. The area under the probability density curve within 10% to the left and right of the highest point of the probability density function distribution is selected as the probability P of the high-density interval. The SOH index is obtained by dividing the probability P of the high-density interval by the distribution range S of the probability density function. By calculating the inconsistency index for each segment, the trend of SOH change can be obtained.
2. The method for assessing dynamic inconsistency and health status of a lithium-ion battery pack according to claim 1, characterized in that, In step 1, the transmission fault detection method is to first calculate the standard deviation of the individual voltage data for all individual cells. Data that has changed compared to the previous time step and whose standard deviation is less than 1 / 10 of the average value is considered abnormal data, and the processing method is linear interpolation. The sensor fault detection method involves identifying voltage spikes greater than 0.3V and voltage differences greater than 0.2V between the spike and the average voltage of the individual sensor at that moment. The processing method is linear interpolation. The SOC data anomaly detection method is to identify an outlier if the SOC jump value is greater than 3, and the processing method is linear interpolation; The main manifestation of mileage data anomalies is SOC jumps. The judgment method is that an SOC jump greater than 3 is considered an outlier, and the handling method is linear interpolation.
3. The method for assessing dynamic inconsistency and health status of a lithium-ion battery pack according to claim 1, characterized in that, In step 2, the similarity evaluation method specifically includes: Step 21: Initialize the distance matrix D First, select a set of voltage segments, and then select the voltage curve V of a single unit within that segment. c =[V c0 V c1 ,…,V cn [n=0,1,…,x-1] represents the voltage value of the single cell at each point in this segment, and the average voltage curve of this segment is V. m =[V m0 V m1 ,…,V mn [n=0,1,…y-1 represents the average unit voltage at each point in this segment, and then the distance matrix D is initialized:] （1） In the formula: the number of rows and columns of matrix D are x and y, respectively, and Inf represents infinity. ， The initial values of the matrices are all infinity; Step 22: Calculate local distance First, define the matrix, then calculate the remaining contents of the matrix using the following formula: The formula for calculating the leftmost column of matrix D is: （2） Where: i = 1, 2, ..., x, The formula for calculating the bottom row of matrix D is: （3） Where: j = 1, 2, ..., y, The formulas for calculating the values of other positions in the matrix are as follows: （4） Where: i = 1, 2, ..., x, j = 1, 2, ..., y, Finally, the local distance matrix is obtained; Step 23: Calculate the globally optimal path Starting from the top right corner of matrix D, search for the optimal path towards the bottom left corner. Let the top right corner be D[x,y]. Starting from this point, select the minimum value among the three nearest neighbors to the bottom left as the starting point for the next step, and continue backtracking until the bottom left corner of the matrix ends. The line connecting all the selected values in this process is V. c With V m Align the path; Step 24: Align the curves Obtain the aligned V based on the alignment path obtained in step 23. c_new and V m_new ; Step 25: Calculate similarity V c With V m The similarity of the aligned V c_new and V m_new The sum of Euclidean distances is used for evaluation, and the calculation formula is as shown in equation (5): （5） The similarity evaluation of the remaining segments and individual units is carried out in the same way as above, and finally the set S of similarity between each individual unit and the average voltage of the segment is obtained. line .
4. The method for assessing dynamic inconsistency and health status of a lithium-ion battery pack according to claim 1, characterized in that, In step 3, the method for obtaining the probability density function is as follows: Step 31: Determine the bandwidth Bandwidth estimation is performed using the Scott criterion, first calculating the similarity S. line The standard deviation is shown in equation (6): （6） Where σ is the standard deviation, and N represents S line The number of data points in S, μ represents the number of data points in S. line The mean, The bandwidth is calculated using the Scott criterion, as shown in equation (7): （7） Step 32: Select kernel function The Gaussian kernel function is chosen as the kernel function for probability density estimation. The formula for the Gaussian kernel function is shown in equation (8): （8） Step 33: Estimate kernel function density First, apply a kernel function to each data point for a given dataset S. line For each point in the matrix, the density estimate of its vicinity is achieved by applying the selected kernel function to that point; Step 33: Obtain the probability density function After obtaining the kernel function results for each point, the results are weighted and summed to estimate the probability density function, as shown in equation (9): （9）。 5. The method for assessing dynamic inconsistency and health status of a lithium-ion battery pack according to claim 1, characterized in that, In step 4, spline fitting is selected to obtain the SOH index curve. Spline fitting generates a smooth overall curve by connecting multiple local interpolation parts, and finally obtains the SOH curve based on inconsistency.