An energy storage battery health online evaluation method based on attenuation trajectory matching

Abstract

This invention discloses an online health assessment method for energy storage batteries based on degradation trajectory matching, relating to the field of energy storage battery technology. The method includes the following steps: S1, conducting multi-condition accelerated aging cycle experiments on the target model energy storage battery, extracting the charging voltage curve morphology feature sequence, constant current charging time sequence, and corresponding capacity retention rate, and forming a reference trajectory template library containing degradation mode labels through cluster analysis. The online health assessment method for energy storage batteries proposed in this invention, by constructing a similarity matching mechanism between the reference trajectory template library and online observation features, transforms SOH assessment from traditional modeling extrapolation prediction to empirical trajectory matching and localization, avoiding the problem of accumulated assessment errors caused by deviations between model assumptions and actual degradation paths in predictive methods. Utilizing the degradation mode labels inherent in the reference templates, simultaneous diagnosis of degradation modes is completed while outputting the SOH assessment value, providing a basis for differentiated operation and maintenance decisions.

Description

Technical Field

[0001] This invention relates to the field of energy storage battery technology, specifically to an online health assessment method for energy storage batteries based on degradation trajectory matching. Background Technology

[0002] During long-term operation, energy storage batteries are affected by factors such as charge-discharge cycles, temperature, and depth of discharge, leading to a gradual degradation in capacity and internal resistance. Accurately assessing their health status is a prerequisite for ensuring the safe operation of energy storage systems and optimizing maintenance decisions. Currently, online assessment methods for the state of health (SOH) of energy storage batteries are mainly divided into two categories: model-driven and data-driven. Model-driven methods describe the aging process by establishing electrochemical models or equivalent circuit models and estimate SOH using parameter identification techniques such as Kalman filtering. Data-driven methods extract health features from charge-discharge data and use machine learning models to map the correlation between these features and SOH.

[0003] The aforementioned methods have achieved good evaluation accuracy under standard laboratory conditions, but they still face the following specific problems in actual operation scenarios of energy storage power stations: First, when energy storage power stations participate in peak shaving and frequency regulation, the depth of charge and discharge rate changes frequently with dispatch instructions, and a single cycle rarely covers the complete full charge and discharge interval, making it difficult for feature extraction methods relying on complete charge and discharge curves to continuously obtain effective input; Second, model-driven methods rely on extrapolation prediction based on mathematical assumptions about the degradation trend. When the actual degradation path of the battery deviates from the preset model, the evaluation error gradually accumulates, especially when a nonlinear acceleration stage occurs in the middle and late stages of degradation; Third, existing online evaluation methods usually output a single SOH value without distinguishing the degradation mode of the battery. Different degradation modes correspond to different safety risks and remaining life expectations, and the lack of mode differentiation capability will limit the refinement of operation and maintenance decisions. To address these issues, we propose an online health evaluation method for energy storage batteries based on degradation trajectory matching. Summary of the Invention

[0004] To address the aforementioned technical issues, an online health assessment method for energy storage batteries based on degradation trajectory matching is provided, comprising the following steps:

[0005] S1. Conduct accelerated aging cycle tests under multiple operating conditions on the target model energy storage battery, extract the charging voltage curve morphology feature sequence, constant current charging time sequence and corresponding capacity retention rate, and form a reference trajectory template library containing decay mode labels through cluster analysis.

[0006] S2. Collect voltage and current data during the operation of the energy storage power station, segment the data according to charging events, extract the voltage change and constant current charging time in the specified voltage range during the charging process, and form an online observation feature sequence that allows for missing points in time sequence;

[0007] S3. Align and compare the end segments of the online observed feature sequences with the reference templates in the template library, calculate the similarity using dynamic time warping distance, select the reference template with the smallest matching distance as the optimal matching template, and read the corresponding capacity retention rate based on the position of the optimal matching segment in the template as the health status assessment result.

[0008] S4. Output the current capacity retention rate, decay mode type and relative position of the trajectory. Repeat the above matching and update steps when new online feature data is accumulated. Trigger a decay anomaly warning when the matching distance exceeds the preset threshold.

[0009] Preferably, for the target model of energy storage battery, multiple combinations of operating conditions, including temperature conditions, depth of discharge, and charge / discharge rate, are set, and accelerated aging cycle tests are conducted respectively.

[0010] Perform standard capacity calibration on the experimental battery at each set period and record the current capacity retention rate;

[0011] Extract the voltage change and constant current charging time within a specified voltage range during each calibration cycle, and construct the attenuation trajectory dataset for each operating condition together with the corresponding capacity retention rate.

[0012] Preferably, the trajectory morphology feature distance between the attenuation trajectory datasets under each working condition is calculated, and the distance is used as a similarity measure for hierarchical clustering, so that trajectories with consistent morphology features are grouped into the same cluster.

[0013] The central trajectory of each cluster is used as a reference template for that type of decay mode, and each reference template is labeled with a decay mode type label and an aging mechanism label.

[0014] All reference templates are stored in a uniform format to form a reference trajectory template library; each reference template contains a reference sequence of charging voltage curve morphology characteristics, a reference sequence of constant current charging time, and a capacity retention health scale.

[0015] Preferably, the terminal voltage and current data of the energy storage battery are collected at a set sampling frequency. When the charging current is detected to jump from zero to a value greater than a set current threshold and continue for more than a set time, a charging event is marked as starting. When the charging current drops below the set current threshold, the charging event is marked as ending.

[0016] For each charging event, determine whether its charging process covers a specified voltage range. If it does, extract the voltage change and constant current charging time within that range as online feature data points for that charging event. If it does not cover the range, mark that charging event as feature missing.

[0017] Preferably, the online feature data points extracted from each charging event are arranged in chronological order to form an online observation feature sequence;

[0018] When a charging event is marked as a missing feature, a gap is reserved at the corresponding position in the online observation feature sequence, thus forming an online observation feature sequence that allows for missing points.

[0019] Preferably, the matching window length is set, and the most recent set number of valid feature data points in the online observation feature sequence are taken as the matching window;

[0020] For each reference template in the reference trajectory template library, align all consecutive data segments of a set length in the feature reference sequence of the template with the window to be matched in sequence;

[0021] During alignment, missing points in the online observation feature sequence are skipped. The dynamic time warping algorithm is used to calculate the distance between each continuous data segment and the window to be matched, and the data segment with the smallest distance in each reference template and its distance value are recorded.

[0022] Preferably, the minimum distance value of all reference templates is compared, and the reference template corresponding to the minimum distance value is selected as the optimal matching template;

[0023] Determine the start and end positions of the optimal matching segment in the optimal matching template, and read the capacity retention rate corresponding to the end position of the optimal matching segment from the health scale of the optimal matching template as the current health status assessment value of the energy storage battery.

[0024] The degradation mode type label of the optimal matching template is used as the degradation mode diagnosis result of the current energy storage battery.

[0025] Preferably, the output includes the current health status assessment value of the energy storage battery, the degradation mode diagnosis result, and the position and proportion of the optimal matching segment in the reference template;

[0026] When a new charging event generates a valid feature data point, the data point is added to the online observation feature sequence, the similarity distance calculation and optimal matching template selection steps are re-executed, and the health status assessment value and decay mode diagnosis results are updated.

[0027] Preferably, the minimum distance value corresponding to the optimal matching template is recorded at each matching update to form a historical sequence of matching residuals;

[0028] Based on the statistical distribution of the historical sequence of the matching residuals, a matching residual early warning threshold is set;

[0029] When the minimum distance value exceeds the matching residual warning threshold in a set number of consecutive matching updates, an attenuation anomaly warning signal is triggered.

[0030] Preferably, after triggering the attenuation anomaly warning signal, the current best matching template is excluded from the reference trajectory template library, and the similarity distance is recalculated for the remaining reference templates;

[0031] If there is a template among the other reference templates whose minimum distance value is lower than the matching residual warning threshold, then the current best matching template is replaced with that template, the attenuation mode diagnosis result is updated, and it is marked as an attenuation mode switch.

[0032] If the minimum distance value of all other reference templates exceeds the matching residual warning threshold, it is marked as an abnormal uncovered attenuation mode.

[0033] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0034] The proposed online health assessment method for energy storage batteries transforms SOH assessment from traditional modeling and extrapolation prediction to empirical trajectory matching and localization by constructing a reference trajectory template library and a similarity matching mechanism for online observation features. This avoids the problem of accumulated assessment errors caused by deviations between model assumptions and actual degradation paths in predictive methods. By skipping missing data points in the feature extraction and matching stages, the method can continuously obtain assessment results even under fragmented conditions of incomplete charging and discharging in energy storage power stations. Utilizing the degradation mode labels inherent in the reference templates, the method simultaneously diagnoses degradation modes while outputting SOH assessment values, providing a basis for differentiated operation and maintenance decisions. Through continuous monitoring of matching residuals, abnormal deviations in the degradation path can be detected before significant capacity decline, providing a scalable technical path for early warning of anomalies in energy storage batteries. Attached Figure Description

[0035] Figure 1 This is a flowchart of the online health assessment method for energy storage batteries according to the present invention. Detailed Implementation

[0036] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0037] Reference Figure 1 As shown, an online health assessment method for energy storage batteries based on degradation trajectory matching includes the following steps:

[0038] S1. Conduct accelerated aging cycle tests under multiple operating conditions on the target model energy storage battery, extract the charging voltage curve morphology feature sequence, constant current charging time sequence and corresponding capacity retention rate, and form a reference trajectory template library containing decay mode labels through cluster analysis.

[0039] S1 includes:

[0040] For the target model of energy storage battery, multiple combinations of operating conditions, including temperature conditions, depth of discharge, and charge / discharge rate, were set. The temperature conditions ranged from -20℃ to 60℃, the depth of discharge ranged from 20% to 100%, and the charge / discharge rate ranged from 0.2C to 2C. All operating condition combinations were generated using a full-factor orthogonal combination method. Three parallel batteries were set up for each operating condition to conduct accelerated aging cycle experiments.

[0041] Standard capacity calibration was performed on the experimental batteries at each set cycle. The standard capacity calibration was carried out in a constant temperature environment of 25℃±2℃. The batteries were discharged at a constant current rate of 0.2C to the nominal discharge cutoff voltage of the target model battery. The nominal discharge cutoff voltage of lithium iron phosphate batteries was 2.5V and that of ternary lithium batteries was 3.0V. The current capacity retention rate was recorded. Standard capacity calibration was performed once every 50 accelerated aging cycles. When the battery capacity retention rate was lower than 80%, the calibration interval was shortened to 20 accelerated aging cycles.

[0042] The voltage change and constant current charging time within a specified voltage range during each calibration cycle are extracted. The specified voltage range is selected based on the constant current charging plateau segment of the battery corresponding to that range. The voltage change in this stage is mainly affected by the battery's internal resistance and the available capacity of the active material, and it has the highest sensitivity to battery aging. The specified voltage range for lithium iron phosphate batteries is 3.2V to 3.4V, and for ternary lithium batteries it is 3.6V to 3.9V. In the early stage of aging, the voltage change in this range increases slowly with capacity decay, and the constant current charging time decreases linearly. In the middle stage of aging, the rate of increase in voltage change accelerates, and the rate of decrease in constant current charging time increases. In the late stage of aging, the voltage change rises sharply, and the constant current charging time decreases rapidly. This method can clearly distinguish different aging stages and, together with the corresponding capacity retention rate, construct a degradation trajectory dataset under various operating conditions.

[0043] The extraction of charging voltage curve morphology and constant current charging time is carried out using the following method: sampling is performed at 1-second intervals within a specified voltage range, and at least 100 sampling points are collected for each charging event within this range. The charging voltage curve morphology sequence is a sequence formed by arranging the voltage values ​​of each sampling point in chronological order. The constant current charging time is the length of time from when the charging current reaches the set constant current value and the voltage enters the specified voltage range until the voltage leaves the specified voltage range or the charging current begins to decrease. The relationship between the capacity retention rate and the constant current charging time is constructed by correlating the capacity retention rate obtained from each calibration with the feature sequence and constant current charging time extracted from the most recent complete charging event within the calibration period to form a feature-capacity mapping pair.

[0044] S1 further includes:

[0045] The trajectory morphology feature distance between the attenuation trajectory datasets under each working condition is calculated. The trajectory morphology feature distance is calculated using a modified Frescher distance, and the formula is as follows:

[0046]

[0047] in and These are two trajectories to be compared. and These are the time parameters for the two trajectories. The Euclidean distance is used as a similarity measure for hierarchical clustering. The Ward method is used as the hierarchical clustering algorithm, and the Euclidean distance is used as the distance measure. The cluster merging criterion is to minimize the sum of squares within groups. The elbow rule is used to determine the number of clusters. A curve showing the relationship between the number of clusters and the sum of squares within groups is plotted. The number of clusters at the inflection point of the curve is selected as the final number of clusters. Trajectories with consistent morphological features are grouped into the same cluster.

[0048] The central trajectory of each cluster is used as a reference template for that type of decay mode. The central trajectory is calculated by taking the arithmetic mean of the feature values ​​of all trajectories within the cluster at the corresponding time points. Each reference template is labeled with a decay mode type label and an aging mechanism label.

[0049] All reference templates are stored in a uniform format to form a reference trajectory template library; the charging voltage curve morphology feature reference sequence of each reference template is processed using minimum-maximum normalization, with the following formula:

[0050]

[0051] in This is the original voltage value. It is the minimum value of all voltage values ​​in this reference template. This is the maximum value. The constant current charging time reference sequence also uses minimum-maximum normalization, and the formula is:

[0052]

[0053] in This is the original constant current charging time. It is the minimum constant current charging time among all constant current charging times in this reference template. It is the maximum value. The capacity retention health scale directly uses the original percentage value without normalization. Each reference template includes a reference sequence of charging voltage curve morphology characteristics, a reference sequence of constant current charging time, and a capacity retention health scale.

[0054] S2. Collect voltage and current data during the operation of the energy storage power station, segment the data according to charging events, extract the voltage change and constant current charging time in the specified voltage range during the charging process, and form an online observation feature sequence that allows for missing points in time sequence;

[0055] S2 includes:

[0056] The terminal voltage and current data of the energy storage battery are collected at a set sampling frequency of 1Hz. The noise filtering of the terminal voltage and current data adopts the moving average filter with a window length of 5 sampling points. The outlier removal adopts the 3σ criterion. Data points that exceed the mean ± 3 times the standard deviation are considered outliers and replaced with the arithmetic mean of the two valid data points before and after. When the charging current jumps from zero to a set current threshold greater than 0.05C and lasts for more than 30 seconds, a charging event is marked as the start of a charging event. When the charging current drops to below the set current threshold of 0.05C, the charging event is marked as the end of the charging event.

[0057] For each charging event, determine whether its charging process covers a specified voltage range. If it does, extract the voltage change and constant current charging time within that range as online feature data points for that charging event. If it does not cover the range, mark that charging event as feature missing.

[0058] S2 further includes:

[0059] According to the time sequence of charging events, the online feature data points extracted from each charging event are arranged. The online observation feature sequence adopts a dual-channel feature organization method. The charging voltage curve morphology feature sequence and constant current charging time extracted from each charging event are used as a two-dimensional feature vector and arranged in time sequence to form a dual-channel sequence, thus forming the online observation feature sequence.

[0060] When a charging event is marked as a feature missing, a vacancy is reserved at the corresponding position in the online observation feature sequence. Missing points are marked with the null value symbol NaN in the sequence. When the number of consecutive missing points does not exceed 3, the vacancy is reserved and automatically skipped in the subsequent matching process. When the number of consecutive missing points exceeds 3, the online observation feature sequence is cleared and the accumulation of valid feature data points starts again to form an online observation feature sequence that allows for missing points.

[0061] S3. Align and compare the end segments of the online observed feature sequences with the reference templates in the template library, calculate the similarity using dynamic time warping distance, select the reference template with the smallest matching distance as the optimal matching template, and read the corresponding capacity retention rate based on the position of the optimal matching segment in the template as the health status assessment result.

[0062] S3 includes:

[0063] Set the matching window length, which is based on 10% to 20% of the average length of the feature sequences in the reference template. It is usually set to 10 effective feature data points. The longer the window length, the higher the evaluation accuracy, but the greater the computational load. The shorter the window length, the smaller the computational load, but the lower the evaluation accuracy. A window length of 10 data points can achieve a good balance between evaluation accuracy and computational load. Take the most recent set number of effective feature data points in the online observed feature sequence as the matching window.

[0064] For each reference template in the reference trajectory template library, align all consecutive data segments of a set length in the feature reference sequence of the template with the window to be matched in sequence;

[0065] During alignment, missing points in the online observation feature sequence are skipped. A dynamic time warping algorithm is used to calculate the distance between each consecutive data segment and the window to be matched. The specific calculation steps of the dynamic time warping algorithm are as follows: First, a distance matrix is ​​constructed, where each element represents the Euclidean distance between the i-th valid data point in the window to be matched and the j-th data point in the reference template data segment. Then, the cumulative distance matrix is ​​calculated, with the following elements:

[0066]

[0067] in These are elements of the distance matrix. The bottom right element of the cumulative distance matrix is ​​the dynamic time-normalized distance between the two sequences. The weighted distance calculation method for dual-channel features is as follows: the weight of the charging voltage curve shape feature is 0.7, the weight of the constant current charging time is 0.3, and the total distance is the weighted sum of the distances of the two channels. The method for skipping missing points during alignment is as follows: when a missing point is encountered in the window to be matched, the point is skipped and does not participate in the distance calculation. The cumulative distance is directly passed to the next valid data point. The data segment with the smallest distance in each reference template and its distance value are recorded.

[0068] S3 further includes:

[0069] Compare the minimum distance values ​​of all reference templates and select the reference template corresponding to the minimum distance value as the optimal matching template;

[0070] Determine the start and end positions of the optimal matching segment within the optimal matching template. Read the capacity retention rate corresponding to the end position of the optimal matching segment from the health scale of the optimal matching template. When the end position of the optimal matching segment exactly corresponds to a sampling point of a certain capacity retention rate in the reference template, directly read the capacity retention rate of that sampling point. When the end position of the optimal matching segment lies between two adjacent sampling points, calculate the corresponding capacity retention rate using linear interpolation. The formula is:

[0071]

[0072] in It is the calculated capacity retention rate. and It is the capacity retention rate between two adjacent sampling points. It is the end position of the optimal matching segment. and It is the location of two adjacent sampling points, which serves as the current health status assessment value of the energy storage battery;

[0073] The attenuation mode type label of the optimal matching template is used as the attenuation mode diagnosis result of the current energy storage battery. The relative position of the trajectory is calculated by dividing the end position of the optimal matching segment by the total length of the reference template, and the percentage obtained is the relative position of the trajectory.

[0074] S4. Output the current capacity retention rate, decay mode type and relative position of the trajectory. Repeat the above matching and update steps when new online feature data is accumulated. Trigger a decay anomaly warning when the matching distance exceeds the preset threshold.

[0075] S4 includes:

[0076] Output the current health status assessment value of the energy storage battery, the degradation mode diagnosis result, and the position and proportion of the optimal matching segment in the reference template;

[0077] When a new charging event generates valid feature data points, these data points are added to the online observation feature sequence. The similarity distance calculation and optimal matching template selection steps are then re-executed to update the health status assessment value and degradation mode diagnosis results. The computational resource requirement for online matching updates is approximately [amount missing] per battery per update. The execution time of the floating-point operation on a typical embedded processor is no more than 100 milliseconds. The data caching strategy is to cache the feature data of the most recent 100 charging events and the results of the most recent 10 matching updates. The cached data is updated in a first-in-first-out manner. The real-time guarantee measures are to set the online matching update task as a high-priority task and use a time-sharing scheduling algorithm to ensure that each matching update is completed within 1 second, which meets the real-time requirements of the energy storage power station.

[0078] S4 further includes:

[0079] Record the minimum distance value corresponding to the optimal matching template at each matching update to form a historical sequence of matching residuals;

[0080] Based on the statistical distribution of the historical sequence of matched residuals, a warning threshold for matched residuals is set. The statistical distribution characteristics of the historical sequence of matched residuals approximately follow a normal distribution. The warning threshold is calculated by adding three times the standard deviation to the mean of the historical sequence of matched residuals. The formula is as follows:

[0081]

[0082] in It is the mean of the matching residual history sequence. It is the standard deviation. The higher the threshold, the lower the false alarm rate, but the higher the false alarm rate. The lower the threshold, the lower the false alarm rate, but the higher the false alarm rate. The 3σ threshold can control the false alarm rate below 0.3% while ensuring a low false alarm rate.

[0083] When the minimum distance value exceeds the matching residual warning threshold in three consecutive matching updates, an attenuation anomaly warning signal is triggered. The specific format of the warning signal includes the warning time, battery number, current matching residual value, warning threshold, and number of consecutive exceedances. The review and manual intervention process after the warning is as follows: after the warning is triggered, the battery management system automatically collects the complete voltage and current data of the battery's last 10 charging events, re-extracts features and performs template matching. If the review result is still abnormal, a manual intervention work order is generated to notify the maintenance personnel to perform offline testing and fault diagnosis on the battery.

[0084] S4 further includes:

[0085] After triggering the attenuation anomaly warning signal, the current best matching template is excluded from the reference trajectory template library, and the similarity distance is recalculated for the remaining reference templates;

[0086] If there is a template among the other reference templates whose minimum distance value is lower than the matching residual warning threshold, then the current best matching template is replaced with that template, the attenuation mode diagnosis result is updated, and it is marked as an attenuation mode switch.

[0087] If the minimum distance value of all other reference templates exceeds the matching residual warning threshold, it is marked as an uncovered degradation mode anomaly. The template library is stored in a relational database, with each reference template corresponding to a database record. The record fields include template ID, degradation mode type label, aging mechanism label, charging voltage curve morphology characteristic reference sequence, constant current charging time reference sequence, capacity retention rate health scale, creation time, and version number. The template library is updated quarterly with a comprehensive update. Based on historical data of online operating batteries and new accelerated aging test data, new reference templates are added. The addition of new battery models or new degradation mode templates is done by generating new reference templates according to step S1 after completing the accelerated aging test of the new battery model or discovering a new degradation mode. After expert review, the templates are added to the template library. The version management strategy uses an incremental version number. Each time the template library is updated, the version number is incremented by 1, and historical versions are retained for easy backtracking and comparison.

[0088] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for online health assessment of energy storage batteries based on degradation trajectory matching, characterized in that, Includes the following steps: S1. Conduct accelerated aging cycle tests under multiple operating conditions on the target model energy storage battery, extract the charging voltage curve morphology feature sequence, constant current charging time sequence and corresponding capacity retention rate, and form a reference trajectory template library containing decay mode labels through cluster analysis. S2. Collect voltage and current data during the operation of the energy storage power station, segment the data according to charging events, extract the voltage change and constant current charging time in the specified voltage range during the charging process, and form an online observation feature sequence that allows for missing points in time sequence; S3. Align and compare the end segments of the online observed feature sequences with the reference templates in the template library, calculate the similarity using dynamic time warping distance, select the reference template with the smallest matching distance as the optimal matching template, and read the corresponding capacity retention rate based on the position of the optimal matching segment in the template as the health status assessment result. S4. Output the current capacity retention rate, decay mode type and relative position of the trajectory. Repeat the above matching and update steps when new online feature data is accumulated. Trigger a decay anomaly warning when the matching distance exceeds the preset threshold.

2. The online health assessment method for energy storage batteries based on attenuation trajectory matching according to claim 1, characterized in that, S1 includes: For the target model of energy storage battery, multiple combinations of operating conditions, including temperature conditions, depth of discharge, and charge / discharge rate, were set, and accelerated aging cycle tests were conducted. Perform standard capacity calibration on the experimental battery at each set period and record the current capacity retention rate; Extract the voltage change and constant current charging time within a specified voltage range during each calibration cycle, and construct the attenuation trajectory dataset for each operating condition together with the corresponding capacity retention rate.

3. The online health assessment method for energy storage batteries based on attenuation trajectory matching according to claim 2, characterized in that, S1 further includes: Calculate the trajectory morphology feature distance between attenuation trajectory datasets under each working condition, and use this distance as a similarity measure to perform hierarchical clustering, grouping trajectories with consistent morphology features into the same cluster. The central trajectory of each cluster is used as a reference template for that type of decay mode, and each reference template is labeled with a decay mode type label and an aging mechanism label. All reference templates are stored in a uniform format to form a reference trajectory template library; each reference template contains a reference sequence of charging voltage curve morphology characteristics, a reference sequence of constant current charging time, and a capacity retention health scale.

4. The online health assessment method for energy storage batteries based on attenuation trajectory matching according to claim 3, characterized in that, S2 includes: The terminal voltage and current data of the energy storage battery are collected at a set sampling frequency. When the charging current jumps from zero to a value greater than the set current threshold and continues for a set duration, a charging event is marked as starting. When the charging current drops below the set current threshold, the charging event is marked as ending. For each charging event, determine whether its charging process covers a specified voltage range. If it does, extract the voltage change and constant current charging time within that range as online feature data points for that charging event. If it does not cover the range, mark that charging event as feature missing.

5. The online health assessment method for energy storage batteries based on attenuation trajectory matching according to claim 4, characterized in that, S2 further includes: The online feature data points extracted from each charging event are arranged in chronological order to form an online observation feature sequence. When a charging event is marked as a missing feature, a gap is reserved at the corresponding position in the online observation feature sequence, thus forming an online observation feature sequence that allows for missing points.

6. The online health assessment method for energy storage batteries based on attenuation trajectory matching according to claim 5, characterized in that, S3 includes: Set the matching window length, and take the most recent set number of valid feature data points in the online observed feature sequence as the window to be matched; For each reference template in the reference trajectory template library, align all consecutive data segments of a set length in the feature reference sequence of the template with the window to be matched in sequence; During alignment, missing points in the online observation feature sequence are skipped. The dynamic time warping algorithm is used to calculate the distance between each continuous data segment and the window to be matched, and the data segment with the smallest distance in each reference template and its distance value are recorded.

7. The online health assessment method for energy storage batteries based on attenuation trajectory matching according to claim 6, characterized in that, S3 further includes: Compare the minimum distance values ​​of all reference templates and select the reference template corresponding to the minimum distance value as the optimal matching template; Determine the start and end positions of the optimal matching segment in the optimal matching template, and read the capacity retention rate corresponding to the end position of the optimal matching segment from the health scale of the optimal matching template as the current health status assessment value of the energy storage battery. The degradation mode type label of the optimal matching template is used as the degradation mode diagnosis result of the current energy storage battery.

8. The online health assessment method for energy storage batteries based on attenuation trajectory matching according to claim 7, characterized in that, S4 includes: Output the current health status assessment value of the energy storage battery, the degradation mode diagnosis result, and the position and proportion of the optimal matching segment in the reference template; When a new charging event generates a valid feature data point, the data point is added to the online observation feature sequence, the similarity distance calculation and optimal matching template selection steps are re-executed, and the health status assessment value and decay mode diagnosis results are updated.

9. The online health assessment method for energy storage batteries based on attenuation trajectory matching according to claim 8, characterized in that, S4 further includes: Record the minimum distance value corresponding to the optimal matching template at each matching update to form a historical sequence of matching residuals; Based on the statistical distribution of the historical sequence of the matching residuals, a matching residual early warning threshold is set; When the minimum distance value exceeds the matching residual warning threshold in a set number of consecutive matching updates, an attenuation anomaly warning signal is triggered.

10. The online health assessment method for energy storage batteries based on attenuation trajectory matching according to claim 9, characterized in that, S4 further includes: After triggering the attenuation anomaly warning signal, the current best matching template is excluded from the reference trajectory template library, and the similarity distance is recalculated for the remaining reference templates; If there is a template among the other reference templates whose minimum distance value is lower than the matching residual warning threshold, then the current best matching template is replaced with that template, the attenuation mode diagnosis result is updated, and it is marked as an attenuation mode switch. If the minimum distance value of all other reference templates exceeds the matching residual warning threshold, it is marked as an abnormal uncovered attenuation mode.

Images