Charging segment screening method for battery state of health assessment model training

Abstract

The application provides a charging segment screening method for battery state of health evaluation model training. The screening method comprises: collecting the voltage and current of the battery within a sampling period, extracting a plurality of charging periods from the sampling period based on the change trend of the current; dividing the charging periods into a plurality of voltage intervals based on the voltage change range, and generating a cumulative charging capacity sequence and an average charging capacity sequence of the charging periods; extracting a health parameter from the cumulative charging capacity sequence and the average charging capacity sequence of the charging periods; obtaining the corresponding total charging capacity based on the cumulative charging capacity sequence of the charging periods, calculating the correlation coefficient of the health parameter of the charging period and the corresponding total charging capacity, and screening the target charging period from the charging period set according to the correlation coefficient. The application can screen the target charging period which is strongly related to the battery state of health, which is used to construct the source domain training set to train the battery state of health evaluation model, and the trained model is migrated to the target domain.

Description

Technical Field

[0001] This invention relates to the field of battery technology, and in particular to a method for selecting charging segments for training a battery health status assessment model. Background Technology

[0002] With the widespread application of electrified transportation vehicles, batteries, as core energy storage devices, play a crucial role in ensuring the safe operation of these vehicles by accurately estimating their State of Health (SOH). Electrified transportation vehicles include, but are not limited to, electric vehicles, rail transit, electric ships, and electric aircraft, and battery types include, but are not limited to, lithium-ion batteries, sodium-ion batteries, quasi-solid-state batteries, and all-solid-state batteries. In actual operation, these batteries gradually degrade due to complex operating conditions. If the battery's health status cannot be accurately monitored, it may lead to abnormal battery performance or even safety risks.

[0003] Traditional battery health state estimation methods are typically based on test data under constant laboratory conditions, which are difficult to adapt to the complex and variable operating conditions in real-world applications. Furthermore, real-world data often suffers from high noise levels and inaccurate battery health state labels. Directly using this data to train a battery health state prediction model can easily lead to large prediction errors and poor generalization ability, failing to meet the needs of engineering practice. Therefore, a charging segment selection method is needed for training a battery health state assessment model. Summary of the Invention

[0004] This invention provides a method for selecting charging segments for training a battery health status assessment model, in order to improve the technical problem that the existing technology, which collects health status training data based on constant laboratory operating conditions, results in poor generalization performance of the health status estimation model.

[0005] This invention provides a charging segment selection method applied to electrified vehicles. The method includes: collecting the voltage and current of the battery under test within a preset sampling period, and extracting multiple charging periods from the sampling period based on the current change trend to form a charging period set; dividing each charging period into multiple continuous voltage intervals based on the voltage change range of each charging period, and calculating the cumulative charging capacity and average charging capacity of each voltage interval to generate a cumulative charging capacity sequence and an average charging capacity sequence for the corresponding charging period; extracting a preset type of health parameter from the cumulative charging capacity sequence and the average charging capacity sequence of each charging period; obtaining the corresponding total charging capacity based on the cumulative charging capacity sequence of each charging period, calculating the correlation coefficient between the health parameter of each charging period and the corresponding total charging capacity, and selecting the target charging period from the charging period set based on the correlation coefficient.

[0006] In one embodiment of the present invention, for each charging period, based on the voltage variation range of the charging period, the charging period is divided into multiple consecutive voltage intervals, and the cumulative charging capacity and average charging capacity of each voltage interval are calculated to generate a cumulative charging capacity sequence and an average charging capacity sequence for the corresponding charging period. This includes: dividing the voltage range corresponding to the charging period into multiple consecutive voltage intervals; for each voltage interval, calculating the cumulative charging capacity and average charging capacity of the voltage interval based on the current corresponding to the voltage interval using the Coulomb counting method; and summing the cumulative charging capacity and average charging capacity of each voltage interval to obtain the cumulative charging capacity sequence and average charging capacity sequence for the charging period.

[0007] In one embodiment of the present invention, the cumulative charging capacity of the j-th voltage range during the charging period is... Where I(i) is the current sampled for the i-th time during the charging period, Δt is the sampling interval, and S j Let be the set of all sampling points within the j-th voltage interval.

[0008] In one embodiment of the present invention, the average charging capacity of the j-th voltage range during the charging period is... in, Let Q be the average voltage value of the j-th voltage interval. sum (j) represents the cumulative charging capacity of the j-th voltage range during the charging period.

[0009] In one embodiment of the present invention, all voltage ranges have the same length.

[0010] In one embodiment of the present invention, the total charging capacity corresponding to each charging period is obtained based on the cumulative charging capacity sequence of each charging period, the correlation coefficient between the health parameters of each charging period and the corresponding total charging capacity is calculated, and the target charging period is selected from the charging period set based on the correlation coefficient. The process includes: obtaining the total charging capacity corresponding to each charging period based on the cumulative charging capacity sequence of each charging period; selecting the second charging period as the charging period to be tested; for each type of health parameter of the charging period to be tested, merging the health parameter with the health parameter corresponding to the selected period, and calculating the Pearson correlation coefficient and grey relational degree between the merged health parameter and the corresponding total charging capacity respectively; wherein, the initial selected period includes the first charging period; selecting the largest Pearson correlation coefficient and grey relational degree from all the Pearson correlation coefficients and grey relational degrees corresponding to all health parameters; determining whether the largest Pearson correlation coefficient is greater than or equal to a preset first threshold, and whether the largest grey relational degree is greater than or equal to a second threshold: if yes, the charging period to be tested is selected as the target charging period, and the next charging period is selected as the charging period to be tested for analysis until all charging periods have been analyzed; if no, the next charging period is selected as the charging period to be tested for analysis until all charging periods have been analyzed.

[0011] In one embodiment of the present invention, the health parameter is merged with the health parameters corresponding to the selected time period, and the Pearson correlation coefficient and grey relational degree between the merged health parameter and the corresponding total charging capacity are calculated respectively. This includes: merging the health parameter of the charging time period to be tested with the health parameter of the corresponding type in the selected time period set; merging the cumulative charging capacity of the charging time period to be tested with the cumulative charging capacity of the selected time period set; and calculating the Pearson correlation coefficient and grey relational degree corresponding to the type of health parameter based on the merged health parameter and the cumulative charging capacity of the corresponding charging time period.

[0012] This invention also provides a training method for a battery health status assessment model. The method includes: acquiring source domain training samples and corresponding source domain sample labels, and target domain training samples and corresponding target domain sample labels; wherein, both the source domain training samples and the target domain training samples include cumulative charging capacity sequences, average charging capacity sequences, and voltage sequences corresponding to different target charging periods, and the target charging periods are obtained through any of the above-mentioned charging segment selection methods; pre-training the battery health status assessment model based on the source domain training samples and corresponding source domain sample labels to obtain a pre-trained battery health status assessment model; and performing transfer training on the pre-trained battery health status assessment model based on the target domain training samples and corresponding target domain sample labels to obtain a finally trained battery health status assessment model.

[0013] In one embodiment of the present invention, a pre-trained battery health status assessment model is transferred to obtain a final trained battery health status assessment model based on target domain training samples and corresponding target domain sample labels. This includes: calculating the cumulative charging similarity between the cumulative charging capacity of the source domain training samples and the cumulative charging capacity of the target domain training samples, and the average charging similarity between the average charging capacity of the source domain training samples and the average charging capacity of the target domain training samples; determining whether the cumulative charging similarity is less than a preset cumulative charging similarity threshold, and whether the average charging similarity is less than a preset average charging similarity threshold; if so, freezing the parameters of the feature extraction network of the pre-trained battery health status assessment model, and fine-tuning the parameters of the prediction network of the pre-trained battery health status assessment model based on the target domain training samples and corresponding target domain sample labels to obtain the final trained battery health status assessment model; otherwise, performing adversarial training on the pre-trained battery health status assessment model based on the domain alignment method, using the source domain training samples and the target domain training samples, to obtain the final trained battery health status assessment model.

[0014] In one embodiment of the present invention, the domain alignment method is the maximum mean difference.

[0015] The beneficial effects of this invention are as follows: This invention proposes a method for selecting charging segments for training a battery health status assessment model. It collects voltage and current data within a preset sampling period and extracts multiple charging segments based on current changes. These charging segments are divided into multiple voltage intervals, and the cumulative charging capacity sequence and average charging capacity sequence for each charging segment are obtained through these voltage intervals. Furthermore, health parameters are extracted, and correlation coefficients are calculated based on the total charging capacity of the corresponding charging segments, enabling quantitative evaluation of different charging segments. This invention can filter target charging segments highly correlated with battery health status from a large amount of complex and noisy operational data, ensuring not only the reliability and stability of health parameter extraction but also effectively improving the accuracy and robustness of the subsequent battery health status assessment model. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0017] In the attached diagram:

[0018] Figure 1 This is a schematic flowchart of a charging segment selection method provided in an embodiment of the present invention;

[0019] Figure 2 This is a flowchart illustrating a training method for a battery health status assessment model provided in one embodiment of the present invention. Detailed Implementation

[0020] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0021] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. The drawings only show components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0022] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0023] This invention provides a method for selecting battery charging segments. Voltage and current are collected within a preset sampling period, and based on current changes, multiple charging segments are extracted from the sampling period. These charging segments are divided into multiple voltage intervals, and the cumulative charging capacity sequence and average charging capacity sequence for each charging segment are obtained through these voltage intervals. Furthermore, health parameters are extracted, and correlation coefficients are calculated based on the total charging capacity of the corresponding charging segment, enabling quantitative evaluation of different charging segments. This invention can filter target charging segments highly correlated with battery health status from a large amount of complex and noisy operational data, ensuring not only the reliability and stability of health parameter extraction but also effectively improving the accuracy and robustness of subsequent battery health status assessment models. The selected target charging segments are used to construct a source domain training set to train the battery health status assessment model, and the trained model is then transferred to the target domain.

[0024] like Figure 1 As shown, the battery charging segment screening method includes the following steps:

[0025] S11. Collect the voltage and current of the battery under test within the preset sampling period, and extract multiple charging periods from the sampling period based on the current change trend to form a charging period set.

[0026] Within a preset sampling period, the voltage and current of the battery under test are continuously collected at preset sampling intervals. Since the sampling period may cover multiple charge-discharge cycles of the battery, all charging processes need to be separated for easier subsequent analysis. Specifically, all collected currents are arranged according to the acquisition time sequence, and trend analysis is performed on the sorted current sequence: if multiple consecutive sampling points show negative current values, the time interval corresponding to these sampling points is taken as an initial charging period. Furthermore, to improve the accuracy of identification, this invention also introduces a State of Charge (SOC) verification mechanism. Specifically, for the aforementioned initial charging period, it is determined whether the battery's SOC shows a continuously monotonically increasing state within that period: if so, the period is confirmed as a valid charging period; otherwise, it is discarded. By combining the current change trend with the SOC change trend, misjudgments caused by measurement noise can be effectively eliminated, ensuring that the extracted charging periods are more accurate and reliable.

[0027] S12. Based on the voltage change range of each charging period, each charging period is divided into multiple consecutive voltage intervals, and the cumulative charging capacity and average charging capacity of each voltage interval are calculated to generate the cumulative charging capacity sequence and average charging capacity sequence of the corresponding charging period.

[0028] For each charging period, based on its voltage variation range, this range is divided into multiple continuous and non-overlapping voltage intervals according to a fixed voltage step size. Each voltage interval corresponds to a defined voltage range. All current sampling points within each voltage interval are extracted, and the cumulative charging capacity and average charging capacity of that interval are calculated using the Coulomb counting method. The cumulative charging capacity sequence and average charging capacity sequence for each voltage interval are then sequentially arranged to obtain the cumulative charging capacity sequence and average charging capacity sequence for that charging period. The cumulative charging capacity sequence includes the cumulative charging capacity of multiple voltage intervals, representing the cumulative characteristic of charging capacity variation with voltage within the current charging period. The average charging capacity sequence includes the average charging capacity of multiple voltage intervals, representing the variation of charging capacity under different voltages within the charging period.

[0029] In an optional embodiment of the present invention, for each charging period, step S12 includes steps S121 to S123:

[0030] S121. Divide the voltage range corresponding to the charging period into multiple continuous voltage intervals.

[0031] The voltage variation range corresponding to the charging period is divided into multiple continuous and non-overlapping voltage intervals. To ensure consistent data resolution across different voltage intervals, in an optional embodiment of the invention, all voltage intervals have the same length. For example, each voltage interval can be set to two adjacent integer voltage ranges [V]. i V i+1 Since the sampling accuracy of battery voltage is 0.1V in practical applications, in order to reduce the impact of the sampling interval (usually 10s) on the calculation results, the collected voltage values ​​can be rounded to fall within the corresponding integer voltage range, thereby ensuring the consistency and comparability of capacity calculations in different ranges.

[0032] S122. For each voltage range, calculate the cumulative charging capacity and average charging capacity of that voltage range based on the current corresponding to that voltage range using the Coulomb counting method.

[0033] For each voltage range during the charging period, the following processing is performed: extract all currents within the voltage range, and calculate the cumulative charging capacity and average charging capacity of the voltage range according to formulas (1) and (2) based on the Coulomb counting method:

[0034]

[0035] Among them, Q sum (j) represents the cumulative charging capacity of the j-th voltage range during the current charging period, I(i) represents the current sampled at the ith time during the current charging period (unit: A), Δt represents the sampling interval (unit: s), and S j Let Q be the set of all sampling points within the j-th voltage interval. avg (j) represents the average charging capacity of the j-th voltage range during the current charging period. Let be the average voltage value of the j-th voltage range.

[0036] S123. Summarize the cumulative charging capacity and average charging capacity of each voltage range to obtain the cumulative charging capacity sequence and average charging capacity sequence for that charging period.

[0037] The aforementioned cumulative charging capacity and average charging capacity are summarized in order of their corresponding voltage ranges to form a cumulative charging capacity sequence and an average charging capacity sequence covering the charging period. With voltage as the x-axis and the corresponding cumulative charging capacity as the y-axis, the cumulative charging capacity curve for that charging period is obtained (denoted as Q). sum (V)) and the average charging capacity curve (denoted as Q) ave (V)). Through extensive experiments, the applicant discovered that, regarding the cumulative charging capacity curve, the peak value of the cumulative charging capacity gradually decreases as the battery ages, and the cumulative charging capacity changes significantly with changes in battery health within the 350-370V voltage range. Furthermore, regarding the average charging capacity curve, the peak value decreases accordingly with battery aging, and the overall shape of the curve becomes smoother. The average charging capacity also changes significantly with changes in battery health within the 350–370V voltage range.

[0038] S13. Extract preset type health parameters from the cumulative charging capacity sequence and average charging capacity sequence of each charging period.

[0039] For a sequence of cumulative charging capacity, the maximum cumulative charging capacity (i.e., Q) can be extracted. sum (V) peak value), voltage corresponding to the maximum cumulative charging capacity, and minimum cumulative charging capacity (i.e., Q) sum The values ​​of (V) include the valley value, the voltage corresponding to the minimum cumulative charging capacity, the rate of change of the maximum and minimum cumulative charging capacity (i.e., the slope of the line connecting the peak and valley values), and the area of ​​the region where the maximum cumulative charging capacity is located.

[0040] The rate of change k of the maximum cumulative charging capacity and the minimum cumulative charging capacity is shown in formula (3):

[0041]

[0042] Among them, V min V max These are the voltages corresponding to the maximum cumulative charging capacity and the minimum cumulative charging capacity, respectively, Q. sum (V min ), Q sum (V max These represent the maximum cumulative charging capacity and the minimum cumulative charging capacity, respectively.

[0043] The area S of the region containing the maximum cumulative charging capacity is shown in formula (4):

[0044]

[0045] Wherein, ΔV is the preset voltage integral half-width (e.g., 1V).

[0046] For an average charging capacity sequence, the maximum average charging capacity (i.e., Q) can be extracted. ave The parameters include the peak value of (V), the voltage corresponding to the maximum average charging capacity, and the average value within a preset voltage range (e.g., 360-380V). Through this method, nine types of health parameters can be obtained for each charging period to characterize the battery's health status. It is understood that this invention does not limit the specific types of health parameters. In addition to the health parameters exemplified above, other statistical quantities or characteristics can be combined in practical applications, such as the integral value of the average charging capacity curve, the inflection point of the average / cumulative charging capacity curve, etc., which will not be detailed here.

[0047] S14. Obtain the corresponding total charging capacity based on the cumulative charging capacity sequence of each charging period, calculate the correlation coefficient between the health parameters of each charging period and the corresponding total charging capacity, and select the target charging period from the charging period set based on the correlation coefficient.

[0048] Specifically, in an optional embodiment of the present invention, step S14 includes steps S141 to S145:

[0049] S141. Obtain the corresponding total charging capacity based on the cumulative charging capacity of each charging period.

[0050] For each charging period, the following processing is performed: the cumulative charging capacity of each voltage range during the charging period is accumulated sequentially to obtain the total charging capacity of the charging period.

[0051] S142. Select the second charging period as the charging period to be tested.

[0052] S143. For each type of health parameter during the charging period to be tested, merge the health parameter with the health parameter sequence corresponding to the selected period, and calculate the Pearson correlation coefficient and grey relational degree between the health parameter and the corresponding total charging capacity respectively; wherein, the initial selected period includes the first charging period.

[0053] For each type of health parameter in the charging period to be tested, the following processing is performed: the health parameter is concatenated with the corresponding type of health parameter sequence in the selected period set to obtain a merged health parameter sequence. Furthermore, the total charging capacity of the charging period to be tested is concatenated with the total charging capacity sequence of the selected period set to obtain a merged total charging capacity sequence. The merged parameter sequence and the merged total charging capacity sequence obtained in the above manner not only include relevant information of the confirmed target charging period but also incorporate information of the current charging period to be tested, thus providing a more comprehensive reflection of the correlation between health parameters and total charging capacity. The initial selected period set only includes the first charging period, and its corresponding Pearson correlation coefficient and grey relational degree are used as reference benchmarks. Specifically, let X be the merged health parameter sequence corresponding to the k-th type of health parameter. k ={x k,1 ,x k,2 ,...,x k,n The total charging capacity sequence after merging is Y = {y1, y2, ..., y}. n}, where n is the number of charging time periods included in the calculation after merging (i.e., the number of selected time periods + the current time period to be tested), x k,i For the k-th type of health parameter in the i-th charging period after merging, y i Let be the total charging capacity for the i-th charging period after merging. The Pearson correlation coefficient between the health parameters of the tested charging period and the total charging capacity is shown in formula (5):

[0054]

[0055] in, The mean of the merged health parameter sequence corresponding to the i-th charging period (i.e., X) k (mean) Let Y be the mean of the combined total charging capacity sequence corresponding to the i-th charging period (i.e., the mean of Y).

[0056] Furthermore, to measure the trend consistency between the health parameters of the tested charging period and the corresponding total charging capacity, it is also necessary to calculate the grey relational degree between the health parameters of the tested charging period and the corresponding total charging capacity. Specifically, let X be the merged health parameter sequence corresponding to the k-th type of health parameter. k ={x k,1 ,x k,2 ,...,xk,n The total charging capacity sequence after merging is Y = {y1, y2, ..., y}. n}, as shown in formula (6), calculate the difference between the k-th type of health parameter and the total charging capacity during the i-th charging period:

[0057] Δ k,i =|x k,i -y i |,i=1,2,..,n (6)

[0058] Where, Δ k,i For the k-th type of health parameter x in the i-th charging period k,i The difference between the total charging capacity yi corresponding to this charging period and the total charging capacity yi. Based on the difference Δ calculated above... k,i The correlation coefficient ξ between the k-th type of health parameter and the total charging capacity during the i-th charging period is calculated according to formula (7). k,i :

[0059]

[0060] Wherein, min(Δ k,i ) and max(Δ k,i ) represent the minimum and maximum differences among all charging periods involved in the calculation after merging, respectively, and μ is the preset resolution coefficient, usually taken as 0.5. The average of the correlation coefficients of all charging periods involved in the calculation after merging is taken, as shown in formula (8), to obtain the gray correlation degree r between the k-th type of health parameter and the total charging capacity. k :

[0061]

[0062] Where, r k The closer the value is to 1, the more consistent the trend of change between this type of health parameter and the total charging capacity, and the higher the correlation between the two.

[0063] In an optional embodiment of the present invention, step S143 includes the following data processing procedure: merging the health parameters of the charging period to be tested with the corresponding type of health parameters of the selected period set; merging the cumulative charging capacity of the charging period to be tested with the cumulative charging capacity of the selected period set; and calculating the Pearson correlation coefficient and grey relational degree corresponding to the type of health parameter based on the merged health parameters and the cumulative charging capacity of the corresponding charging period.

[0064] Since the amount of data for a single charging period is limited, its correlation coefficient or grey relational degree lacks statistical significance. Therefore, it is necessary to construct a dataset containing health parameters and total charging capacity based on multiple charging periods to achieve a global correlation assessment between the health parameter sequence and the total charging capacity sequence. Specifically, the following processing is performed on the current charging period to be tested: the extracted health parameters of the charging period to be tested are concatenated with the corresponding type of health parameter sequence in the selected period set to form a merged health parameter sequence covering the historical selected period and the current charging period to be tested. In addition, the cumulative charging capacity of the charging period to be tested is merged with the cumulative charging capacity sequence of the selected period set to obtain a merged cumulative charging capacity sequence. Using the above formula (5), the Pearson correlation coefficient between the merged health parameter sequence and the cumulative charging capacity sequence is calculated to quantify the linear synchronous change relationship between the two. The grey relational degree between the two is calculated using formulas (6)–(8) to assess the overall similarity between the health parameters and the total charging capacity. By combining historical and current data in this calculation method, the present invention not only preserves the statistical characteristics of historical high-quality periods, but also dynamically introduces new data for correlation verification, effectively improving the robustness and accuracy of charging segment selection, and reducing the risk of misjudgment caused by data fluctuations in a single period.

[0065] S144. From all the Pearson correlation coefficients and grey relational degrees corresponding to the health parameters, select the largest Pearson correlation coefficient and grey relational degree.

[0066] For each type of health parameter, the Pearson correlation coefficient and grey relational degree between it and the total charging capacity are calculated. From the Pearson correlation coefficients corresponding to all health parameters, the largest value is selected as the target Pearson correlation coefficient to measure the optimal result of the linear correlation between the health parameter and the total charging capacity. Similarly, from the grey relational degrees corresponding to all health parameters, the largest value is selected as the target grey relational degree to measure the optimal result of the correlation between the health parameter and the trend of change in the total charging capacity. This maximum value selection strategy ensures that subsequent correlation measurements are always based on the most representative indicators, thereby improving the reliability and accuracy of the charging period selection results.

[0067] S145. Determine whether the largest Pearson correlation coefficient is greater than or equal to the preset first threshold, and whether the largest grey relational degree is greater than or equal to the second threshold: If yes, then take the charging period to be tested as the target charging period, and continue to select the next charging period as the charging period to be tested for analysis, until all charging periods have been analyzed; if no, then continue to select the next charging period as the charging period to be tested for analysis, until all charging periods have been analyzed.

[0068] If the target Pearson correlation coefficient is greater than or equal to the first threshold (e.g., 0.95), and the target grey relational degree is greater than or equal to the second threshold (e.g., 0.95), it indicates a strong correlation between the health parameters and cumulative charging capacity of the current charging period. This charging period can then be selected as the target charging period, and the analysis can continue until all charging periods have been analyzed. Conversely, if the correlation coefficient is not found, the next charging period can be directly selected for analysis. This dual-threshold determination method not only avoids misjudgments caused by fluctuations in a single indicator but also considers the linear correlation and trend consistency of both indicators, thereby improving the stability and accuracy of the target charging period selection results.

[0069] like Figure 2 As shown, in an optional embodiment of the present invention, a training method for a battery health status assessment model is also provided, comprising the following processing steps:

[0070] S21. Obtain source domain training samples and corresponding source domain sample labels, target domain training samples and corresponding target domain sample labels; wherein, both source domain training samples and target domain training samples include the cumulative charging capacity, average charging capacity and voltage corresponding to different target charging periods, and the target charging periods are obtained by any of the above battery charging segment screening methods.

[0071] The source and target domains can correspond to charging data collected under different battery objects, operating conditions, or usage scenarios. Specifically, regardless of the source or target domain, each training sample includes the voltage sequence, cumulative charging capacity sequence, and average charging capacity sequence extracted during the target charging period. Correspondingly, the source domain sample label refers to the battery health status corresponding to the target charging period in the source domain training samples, and the target domain sample label refers to the battery health status corresponding to the target charging period in the target domain training samples. The target charging periods in both the source and target domains are obtained through the aforementioned battery charging segment selection method. This ensures that the battery health status assessment model receives more robust and reliable training data, resulting in a more robust model.

[0072] S22. Based on the source domain training samples and the corresponding source domain sample labels, the battery health status assessment model is pre-trained to obtain the pre-trained battery health status assessment model.

[0073] Specifically, the voltage sequence, cumulative charging capacity sequence, and average charging capacity sequence corresponding to the target charging period in the source domain training samples are input into the battery health status assessment model to generate a predicted battery health status value for the corresponding target charging period. The difference between this predicted battery health status value and the source domain sample label (i.e., the true battery health status value) is calculated, and the parameters of the health status assessment model are updated based on the difference. This process is iterated multiple times to obtain a pre-trained battery health status assessment model. The difference can be mean squared error or cross-entropy, etc., without specific limitations. It is understood that the battery health status assessment model can be any model capable of establishing a mapping relationship between charging characteristics and battery health status, including but not limited to Convolutional Neural Networks (CNN), Deep Neural Networks (DNN), Recurrent Neural Networks (RNN), etc., without specific limitations.

[0074] S23. Based on the target domain training samples and the corresponding target domain sample labels, perform transfer training on the pre-trained battery health status assessment model to obtain the final trained battery health status assessment model.

[0075] Considering the differences between the source domain and the target domain in terms of battery type, operating conditions, or usage scenarios, directly applying the source domain model to the target domain would be difficult to guarantee prediction accuracy. Therefore, after completing the pre-training of the source domain, this invention further performs transfer training on the pre-trained battery health status assessment model based on the target domain training samples and their corresponding target domain sample labels to obtain the final trained battery health status assessment model. Specifically, the voltage sequence, cumulative charging capacity sequence, and average charging capacity sequence extracted from the target charging period in the target domain training samples are input into the pre-trained battery health status assessment model to generate a predicted battery health status value for the target charging period. The difference between this predicted value and the target domain sample labels is calculated, and the model parameters are iteratively updated based on the difference to enable the model to gradually transfer and adapt to the characteristics of the target domain, thus obtaining the final trained battery health status assessment model.

[0076] In an optional embodiment of the present invention, step S23 includes the following process: calculating the cumulative charging similarity between the cumulative charging capacity of the source domain training samples and the cumulative charging capacity of the target domain training samples, and the average charging similarity between the average charging capacity of the source domain training samples and the average charging capacity of the target domain training samples; determining whether the cumulative charging similarity is less than a preset cumulative charging similarity threshold, and whether the average charging similarity is less than a preset average charging similarity threshold; if so, freezing the parameters of the feature extraction network of the pre-trained battery health status assessment model, and fine-tuning the parameters of the prediction network of the pre-trained battery health status assessment model based on the target domain training samples and the corresponding target domain sample labels, to obtain the final trained battery health status assessment model; otherwise, performing adversarial training on the pre-trained battery health status assessment model based on the domain alignment method, based on the source domain training samples and the target domain training samples, to obtain the final trained battery health status assessment model.

[0077] Specifically, the cumulative charging similarity between the cumulative charging capacity of the source domain training samples and the cumulative charging capacity of the target domain training samples, as well as the average charging similarity between the average charging capacity of the source domain training samples and the average charging capacity of the target domain training samples, are calculated. It is then determined whether the cumulative charging similarity is less than a preset cumulative charging similarity threshold, and whether the average charging similarity is less than a preset average charging similarity threshold. If both are less than the threshold, it indicates a significant difference in charging characteristics between the source and target domains. In this case, the parameters of the feature extraction network in the pre-trained battery health status assessment model are frozen, and the parameters of the prediction network are fine-tuned only based on the target domain training samples and their corresponding target domain sample labels, thus obtaining the final trained battery health status assessment model. Conversely, if the similarity is not less than the threshold, it indicates that the charging characteristics of the source and target domains are relatively similar. In this case, a domain alignment method is used to perform adversarial training on the pre-trained battery health status assessment model based on the source and target domain training samples to achieve consistency in cross-domain feature distribution, thereby obtaining the final trained battery health status assessment model. In order to reduce the inconsistency of feature distribution between two domains and thus improve the effectiveness of cross-domain migration, in an optional embodiment of the present invention, the domain alignment method is the maximum mean difference.

[0078] As a specific example, the training data comes from the Shanghai New Energy Vehicle Public Data Collection and Monitoring Center, which provides operational data for 50 private cars from June 2022 to June 2023, with a sampling interval of 10 seconds. These vehicles include three different models, all using ternary lithium battery cells, but with differences in rated capacity and operating voltage. This invention selects data from one model to construct a source domain dataset, and then uses the proposed battery health status assessment model to train the model, transferring it to the other two models.

[0079] Specifically, charging segments are extracted from the measured data of the source domain to construct the cumulative charging capacity curve Q. sum (V) and average charging capacity curve Q ave (V), extract health parameters highly correlated with battery capacity from the two curves, specifically including: Q sum (V) Peak value (H1) and its corresponding voltage (H2) in the 350-370V voltage range, valley value (H3) and its corresponding voltage (H4) in the 370-390V voltage range, slope of the line connecting the peak value and the estimated value (H5), area of ​​the region corresponding to the peak value (H6); Q ave (V) Peak value (H7) and its corresponding voltage (H8) in the 350-370V voltage range, and mean value (H9) in the 360-380V voltage range. For each charging segment, the Pearson correlation coefficient and grey relational degree between each type of health parameter and the total charging capacity are calculated, and the maximum Pearson correlation coefficient ρ for each charging segment is statistically analyzed. max And the maximum grey relational degree r max If ρ max Greater than or equal to the preset first threshold and r max If the value is greater than or equal to a preset second threshold, then the charging segment is considered the target charging segment. This dual-threshold determination method allows for the selection of high-quality charging segments from chosen vehicle models, enabling the construction of source or target domain training samples.

[0080] Furthermore, the battery health status assessment model in this embodiment includes a cascaded feature extraction network and a prediction network, wherein the feature extraction network is a convolutional neural network and the prediction network is a bidirectional long short-term memory network. The convolutional neural network receives three input channels, namely the voltage sequence, cumulative charging capacity sequence, and average charging capacity sequence for each charging segment, and outputs the extracted latent vector. The latent vector is input into the bidirectional long short-term memory network to obtain the predicted battery health status value for the input charging segment. This embodiment uses a high-quality charging segment of one vehicle model to train the battery health status assessment model in the source domain, and then transfers the trained model to two other vehicle models. Validation results on two target domain datasets show that the model obtained by the proposed battery health status assessment model training method has an estimation error (maximum root mean square error) of no more than 3%.

[0081] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A method for screening charging segments, characterized in that, Applied to electrified vehicles, the method includes: The voltage and current of the battery under test are collected within a preset sampling period, and multiple charging periods are extracted from the sampling period based on the current change trend to form a charging period set. Based on the voltage variation range of each charging period, each charging period is divided into multiple consecutive voltage intervals, and the cumulative charging capacity and average charging capacity of each voltage interval are calculated to generate the cumulative charging capacity sequence and average charging capacity sequence of the corresponding charging period. Extract preset types of health parameters from the cumulative charging capacity sequence and average charging capacity sequence for each charging period; The total charging capacity is obtained based on the cumulative charging capacity sequence of each charging period. The correlation coefficient between the health parameters of each charging period and the corresponding total charging capacity is calculated. The target charging period is then selected from the set of charging periods based on the correlation coefficient. The process involves obtaining the corresponding total charging capacity based on the cumulative charging capacity sequence of each charging period, calculating the correlation coefficient between the health parameters of each charging period and the corresponding total charging capacity, and selecting target charging periods from the set of charging periods based on the correlation coefficient. The total charging capacity is obtained based on the cumulative charging capacity sequence of each charging period. Select the second charging period as the charging period to be tested; For each type of health parameter during the charging period to be tested, the health parameter is merged with the health parameters corresponding to the selected period, and the Pearson correlation coefficient and grey relational degree between the merged health parameter and the corresponding total charging capacity are calculated respectively; where the initial selected period includes the first charging period. From all the Pearson correlation coefficients and grey relational coefficients corresponding to the health parameters, select the largest Pearson correlation coefficient and grey relational coefficient. Determine whether the largest Pearson correlation coefficient is greater than or equal to a preset first threshold, and whether the largest grey relational degree is greater than or equal to a second threshold: If so, the charging period to be tested will be taken as the target charging period, and the next charging period will be selected as the charging period to be tested for analysis, until all charging periods have been analyzed. If not, continue to select the next charging period as the charging period to be analyzed, until all charging periods have been analyzed.

2. The charging segment screening method according to claim 1, characterized in that, For each charging period, based on the voltage variation range of the charging period, the charging period is divided into multiple consecutive voltage intervals, and the cumulative charging capacity and average charging capacity of each voltage interval are calculated to generate the corresponding cumulative charging capacity sequence and average charging capacity sequence for the charging period, including: The voltage range corresponding to the charging period is divided into multiple consecutive voltage intervals; For each voltage range, the cumulative charging capacity and average charging capacity of that voltage range are calculated based on the coulomb counting method according to the current corresponding to that voltage range. By summing the cumulative charging capacity and average charging capacity of each voltage range, the cumulative charging capacity sequence and average charging capacity sequence for that charging period are obtained.

3. The charging segment screening method according to claim 1, characterized in that, During the charging period Cumulative charging capacity of each voltage range = ,in, During the charging period The current of the next sample, The sampling interval is... For the first The set of all sampling points within a voltage range.

4. The charging segment screening method according to claim 1, characterized in that, During the charging period Average charging capacity of each voltage range ,in, For the first The average voltage across a voltage range. During the charging period The cumulative charging capacity of each voltage range.

5. The charging segment screening method according to claim 1, characterized in that, All voltage ranges have the same length.

6. The charging segment screening method according to claim 1, characterized in that, The process involves merging the health parameter with the health parameters corresponding to the selected time period, and calculating the Pearson correlation coefficient and grey relational degree between the merged health parameter and the corresponding total charging capacity, including: The health parameters of the charging period to be tested are merged with the corresponding health parameters of the selected period set, and the cumulative charging capacity of the charging period to be tested is merged with the cumulative charging capacity of the selected period set. Based on the merged health parameters and the cumulative charging capacity of the corresponding charging period, the Pearson correlation coefficient and grey relational degree corresponding to this type of health parameter are calculated.

7. A training method for a battery health status assessment model, characterized in that, The method includes: Obtain source domain training samples and corresponding source domain sample labels, target domain training samples and corresponding target domain sample labels; wherein, both the source domain training samples and the target domain training samples include cumulative charging capacity sequences, average charging capacity sequences and voltage sequences corresponding to different target charging periods, and the target charging periods are obtained by the charging segment filtering method described in any one of claims 1 to 6; The battery health status assessment model is pre-trained based on the source domain training samples and the corresponding source domain sample labels to obtain the pre-trained battery health status assessment model. The pre-trained battery health status assessment model is transferred to train based on the target domain training samples and the corresponding target domain sample labels to obtain the final trained battery health status assessment model.

8. The training method for the battery health status assessment model according to claim 7, characterized in that, The pre-trained battery health status assessment model is transferred to another training model based on the target domain training samples and corresponding target domain sample labels to obtain the final trained battery health status assessment model, including: Calculate the cumulative charging similarity between the cumulative charging capacity of the source domain training samples and the cumulative charging capacity of the target domain training samples, as well as the average charging similarity between the average charging capacity of the source domain training samples and the average charging capacity of the target domain training samples. Determine whether the cumulative charging similarity is less than a preset cumulative charging similarity threshold, and whether the average charging similarity is less than a preset average charging similarity threshold; If so, freeze the parameters of the feature extraction network of the pre-trained battery health status assessment model, and fine-tune the parameters of the prediction network of the pre-trained battery health status assessment model based on the target domain training samples and the corresponding target domain sample labels to obtain the final trained battery health status assessment model. Otherwise, based on the domain alignment method, adversarial training is performed on the pre-trained battery health status assessment model using training samples from the source domain and the target domain to obtain the final trained battery health status assessment model.

9. The training method for the battery health status assessment model according to claim 8, characterized in that, The domain alignment method is the maximum mean difference.

Images