Battery state of health estimation method, apparatus, device, and medium

Abstract

The application relates to the technical field of batteries, and discloses a battery health state estimation method, device, equipment and medium; wherein the method first collects time sequence data of battery operation and divides the time sequence data into multiple continuous time segments; then, a target parameter to be estimated is determined, and a battery state space model is constructed; subsequently, sensitivity calculation is performed on each continuous time segment based on the battery state space model, and sensitive data segments sensitive to the target parameter are screened; then, a proxy model is used to perform optimization search on the target parameter based on the sensitive data segments, and a candidate parameter combination is generated; finally, the candidate parameter combination is substituted into the battery state space model for error calculation, the proxy model is iteratively updated according to the error result, and a battery health state estimation value is obtained. The method can eliminate low-sensitivity data to avoid interference and improve SOH estimation accuracy, can reduce the consumption of calculation resources, can accelerate the convergence speed, can guarantee estimation stability, and can meet the dual requirements of real-time performance and accuracy of a vehicle-mounted BMS.

Description

Technical Field

[0001] This application relates to the field of battery technology, and more specifically, to a method, apparatus, device, and medium for estimating battery health status. Background Technology

[0002] Accurate estimation of the State of Health (SOH) of lithium-ion batteries is a core prerequisite for ensuring the safe and reliable operation of electric vehicle batteries and is crucial to the performance of the Battery Management System (BMS). Current research focuses on designing precise battery models and advanced estimation algorithms to improve estimation accuracy, but generally neglects the critical impact of data quality. Traditional estimation algorithms treat all measured data, such as current, voltage, and temperature, equally, while in reality only a small portion of the data is sensitive to SOH estimation. Low-sensitivity data is easily affected by measurement noise and model uncertainty, leading to significant estimation errors. At the same time, mainstream estimation algorithms such as differential evolution and particle swarm optimization require a large amount of computational resources, have large parameter search spaces, and slow convergence, making it difficult to meet the real-time requirements of automotive BMS and severely limiting their feasibility for application in practical scenarios. Summary of the Invention

[0003] The main purpose of this application is to provide a battery health state estimation method, which aims to solve the technical problems in battery SOH estimation, such as low accuracy due to neglecting data quality and low-sensitivity data interference, and the large computational resource requirements and slow convergence of mainstream algorithms, which make it difficult to meet the real-time requirements of vehicle BMS.

[0004] The first aspect of this application proposes a battery health state estimation method, including:

[0005] Collect timing data of battery operation and divide it into multiple continuous time segments;

[0006] Define the target parameters to be estimated and construct a battery state-space model;

[0007] Sensitivity calculations are performed on each continuous time segment based on the battery state-space model, and sensitive data segments that are sensitive to the target parameters are selected.

[0008] Using a proxy model, the target parameters are optimized and searched based on the sensitive data fragments to generate candidate parameter combinations;

[0009] The candidate parameter combination is substituted into the battery state-space model for error calculation. The surrogate model is iteratively updated based on the error results to obtain the battery health state estimate.

[0010] Furthermore, the step of collecting time-series data of battery operation and dividing it into multiple consecutive time segments includes:

[0011] Collect raw operating parameter data during battery operation;

[0012] Based on a fixed time window or battery charging / discharging phase, the original operating parameter data is divided into multiple continuous time segments.

[0013] Furthermore, the step of defining the target parameters to be estimated and constructing a battery state-space model includes:

[0014] Define the target parameters to be estimated, including the initial charge state and the ohmic internal resistance;

[0015] A battery state-space model is established based on the target parameters.

[0016] Furthermore, the step of performing sensitivity calculations on each continuous time segment based on the battery state-space model and filtering out sensitive data segments that are sensitive to the target parameter includes:

[0017] Sensitivity calculations are performed on each consecutive time segment using the battery state-space model to obtain sensitivity indices.

[0018] The sensitivity level of each consecutive time segment is determined based on the aforementioned sensitivity index, and the initial data segments are selected.

[0019] Sensitive data segments are selected from the initial data segments based on their sensitivity level.

[0020] Furthermore, before the step of using a proxy model to optimize the search for the target parameters based on the sensitive data fragment and generate candidate parameter combinations, the following steps are included:

[0021] Build the initial agent model;

[0022] Generate initial parameter combinations of multiple target parameters, substitute them into the battery state space model, simulate the predicted voltage based on the current data of the sensitive data segment, and calculate the root mean square error between the predicted voltage and the actual voltage of the sensitive data segment.

[0023] The initial proxy model is initialized and trained using the initial parameter combination and the corresponding error to obtain a target proxy model that has the ability to map between target parameters and error.

[0024] Furthermore, the step of using a proxy model to optimize the search for the target parameters based on the sensitive data fragment and generate candidate parameter combinations further includes:

[0025] Define the acquisition function for the target agent model;

[0026] The potential combinations of target parameters are evaluated using the acquisition function.

[0027] Select parameter combinations whose evaluation results of the acquisition function meet the preset conditions as candidate parameter combinations.

[0028] Further, the step of substituting the candidate parameter combination into the battery state-space model for error calculation, and iteratively updating the surrogate model based on the error result to obtain the battery health state estimate includes:

[0029] The candidate parameter combination is substituted into the battery state space model, and the current data of the sensitive data segment is input to calculate the predicted voltage.

[0030] Calculate the root mean square error between the predicted voltage and the actual voltage of the sensitive data segment, and use it as the error value;

[0031] The candidate parameter combinations and their corresponding error values ​​are added to the training set.

[0032] The agent model is retrained using the updated training set, and the iteration is repeated until a preset termination condition is met, and the battery health status estimate is output.

[0033] A second aspect of this application also provides a battery health state estimation device, comprising:

[0034] The data acquisition and segmentation module is used to collect time-series data of battery operation and divide it into multiple continuous time segments;

[0035] The parameter model building module is used to define the target parameters to be estimated and build a battery state-space model.

[0036] The sensitive segment filtering module is used to perform sensitivity calculations on each continuous time segment based on the battery state space model, and filter out sensitive data segments that are sensitive to the target parameters.

[0037] The parameter combination generation module is used to optimize the search for the target parameters based on the sensitive data fragment using a proxy model, and generate candidate parameter combinations.

[0038] The health status estimation module is used to substitute the candidate parameter combination into the battery state space model to calculate the error, and iteratively update the surrogate model based on the error result to obtain the battery health status estimate.

[0039] A third aspect of this application provides a battery health state estimation device, comprising:

[0040] processor;

[0041] Memory, used to store computer programs;

[0042] Wherein, when the processor executes the computer program, it implements the battery health status estimation method described in any of the above embodiments.

[0043] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the battery health state estimation method described in any of the above embodiments.

[0044] The first aspect of this plan brings the following benefits:

[0045] This application collects time-series data and filters data segments that are highly sensitive to parameters while eliminating data segments that are not sensitive to parameters. This avoids noise and model uncertainty interference from the source, significantly improving estimation accuracy. At the same time, it uses a surrogate model to replace the large number of parameter traversals in traditional algorithms, and combines optimization search to generate candidate parameters, reducing computational resource consumption and accelerating convergence. Furthermore, it ensures stability by iteratively updating the surrogate model, adapting to the dual requirements of real-time performance and accuracy of in-vehicle BMS, effectively overcoming the limitations of traditional methods. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating a battery health state estimation method according to an embodiment of this application;

[0047] Figure 2 This is a schematic diagram of the structure of a battery health status estimation device according to an embodiment of this application;

[0048] Figure 3 This is a schematic diagram of the structure of a computer device according to an embodiment of this application;

[0049] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0051] Those skilled in the art will understand that, unless explicitly stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in the specification of this application means the presence of features, integers, steps, operations, elements, modules, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, modules, components, and / or groups thereof. It should be understood that when an element is “connected” or “coupled” to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein may include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any modules and all combinations of one or more associated listed items.

[0052] Those skilled in the art will understand that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0053] Reference Figure 1 This application provides a battery health state estimation method, including:

[0054] S1: Collect timing data of battery operation and divide it into multiple continuous time segments;

[0055] S2: Define the target parameters to be estimated and construct a battery state-space model;

[0056] S3: Based on the battery state-space model, perform sensitivity calculations on each continuous time segment and filter out sensitive data segments that are sensitive to the target parameter;

[0057] S4: Using a proxy model, optimize the search for the target parameters based on the sensitive data fragments to generate candidate parameter combinations;

[0058] S5: Substitute the candidate parameter combination into the battery state space model to calculate the error, and iteratively update the surrogate model based on the error result to obtain the battery health state estimate.

[0059] In step S1, "collecting time-series data of battery operation and dividing it into multiple continuous time segments" is the fundamental input step of the entire battery health state estimation method. It requires first acquiring core data that reflects the battery's true operating characteristics, and then rationally dividing it into the smallest analyzable data units to provide accurate input for subsequent sensitive data screening and parameter optimization. Taking electric vehicle on-board lithium-ion batteries (such as lithium iron phosphate batteries) as an example, the battery management system (BMS) first collects real-time time-series data of the battery during actual driving through current sensors, voltage sensors, and temperature sensors—specifically, charging and discharging current (e.g., -200A~300A, negative values ​​for charging and positive values ​​for discharging), terminal voltage (e.g., 2.5V~3.65V), temperature, and SOC 10 times per second. The collection period is approximately three months, resulting in a raw dataset of over 311,000 frames of current-voltage-temperature data containing timestamps and battery usage status. Next, considering the characteristics of battery charging and discharging conditions, a "battery charging stage" division method (which better reflects battery characteristic changes compared to a fixed SOC window) is adopted to divide the raw data into multiple continuous SOC segments: These include data segments corresponding to the low SOC (0-45%), medium SOC (45%-85%), and high SOC (85-100%) charging segments (stable current); data segments corresponding to the pre-charging rest segment (stable voltage, absolute current ≤ 5A); data segments corresponding to the completed charging rest segment (absolute current ≤ 2A); and data segments corresponding to the pulse discharge segment (SOC in the 0-45%, 45%-85%, and 85-100% ranges, with a current difference ≥ 100A between adjacent frames). This results in 8+ continuous time segments, each containing complete current-voltage and temperature change characteristics for that stage, providing targeted data units for the sensitivity calculation in subsequent step S3. This step ensures that the data accurately reflects the battery's true operating state by precisely collecting core time-series data, avoiding estimation bias caused by missing data. At the same time, it divides the data into segments according to charging, discharging, and storage stages, allowing each segment to focus on the battery characteristics under specific operating conditions. This lays the foundation for subsequent screening of highly sensitive data, effectively reducing interference from low-sensitivity redundant data in subsequent steps and improving the efficiency and accuracy of the overall estimation process.

[0060] In step S2, combining the current-voltage time-series data of the lithium iron phosphate battery collected in step S1 above, the target parameters to be estimated are first defined: the core parameters are the initial state of charge (SOC0, SOC, State of Charge) and internal resistance (R0). Among them, the initial SOC0 reflects the proportion of the battery's remaining capacity at the beginning of the segment (e.g., the initial SOC0 of the constant current charging segment is 60%), and the internal resistance R0 is directly related to the degree of battery aging (R0 of a new battery is about 50mΩ, which increases to more than 65mΩ after aging). The two together determine the estimation result of SOH (State of Health). Next, based on the model selection principle of "adapting to the lightweight requirements of automotive applications," a Rint (Resistance Internal) model was constructed as the battery state-space model. This model has a simple structure (composed of an ideal voltage source U0 and an ohmic internal resistance R0 connected in series) and can accurately describe the "current input-voltage output" relationship. Its state equation is: U_meas = U0 - I×R0 (U_meas is the actual terminal voltage collected in step S1, I is the collected current, and U0 is obtained from the table by looking up SOC0, such as U0=3.2V when SOC0=60%). Through this model, the target parameters (SOC0, R0) can be directly correlated with the current-voltage data in step S1. For example, by substituting the current data of the charging segment (I=20A) in the 0-45% SOC range into the model, the corresponding predicted voltage can be calculated, providing model support for the sensitivity calculation in the subsequent step S3. This step focuses on core target parameters to avoid irrelevant parameters interfering with estimation accuracy; it selects a lightweight Rint model to ensure the accuracy of the mapping between data and parameters while reducing the computational burden in the vehicle scenario, laying an efficient and reliable foundation for subsequent sensitive data screening and proxy model optimization.

[0061] In step S3, based on the above, and using the Rint model constructed in step S2 (U_meas = U0 - I×R0, where U0 is determined by SOC0), sensitivity calculations are performed on the 32 lithium iron phosphate battery time segments (including charging in the 0-45%, 45%-85%, and 85-100% SOC ranges, pulse discharge, and rest before and after charging) divided in step S1. Referring to the "algebraic sensitivity calculation method," the partial derivative of the output voltage with respect to the target parameters (SOC0, R0) is used as the sensitivity index: the "voltage change / parameter change" in each segment is calculated. For example, in the 0-45% SOC range charging segment (I=20A), when R0 increases from 50mΩ to 51mΩ, the voltage decreases by 0.02V, and the sensitivity index is 0.02V / mΩ; while in the 45%-85% charging segment (I decreases from 20A to 5A), the same change in R0 only causes the voltage to decrease by 0.005V, and the sensitivity index is 0.005V / mΩ. Based on the index ranking, 20 initial data segments with sensitivity higher than 0.01V / mΩ were first selected. Then, the Top 10 segments (all charging / discharging segments) were chosen as sensitive data segments—these segments respond more significantly to parameter changes. For example, the sensitivity index of a certain discharge segment (I=30A) reaches 0.03V / mΩ, which can more accurately reflect the true changes in SOC0 and R0, providing a high-quality data foundation for the surrogate model optimization in subsequent step S4. This step quantifies the data value through sensitivity calculation, accurately selecting sensitive segments with high information density and eliminating low-sensitivity SOC segments and other data, avoiding noise and model uncertainty interference, providing accurate input for subsequent parameter optimization, and improving the accuracy of SOH estimation from the source.

[0062] In step S4, based on the above, and using the 10 charge / discharge sensitive data segments selected in step S3 (e.g., a certain SOC discharge segment: I=30A, actual voltage 2.8V-3.6V), a Gaussian Process Regression (GPR) model is first constructed as an initial surrogate model using the "small sample learning adaptation" principle. Next, 20 initial combinations of target parameters (SOC0: 40%-90%, R0: 40mΩ-90mΩ) are generated and substituted into the Rint model from step S2. The voltage is then simulated and predicted based on the current data of the sensitive segments, and the root mean square error (RMSE) between the predicted and actual voltages is calculated. For example, when SOC0=60% and R0=50mΩ, RMSE=0.02V; when SOC0=70% and R0=60mΩ, RMSE=0.08V. The GPR model is trained using the "initial parameter combination + corresponding RMSE" to obtain a target surrogate model that maps the relationship between parameters and error. Subsequently, a "balanced exploration and development" acquisition function was selected, with the maximum expected improvement (EI) set as the acquisition function. Potential parameter combinations were evaluated—the higher the EI value, the greater the parameter improvement potential. Ultimately, the combination with the highest EI value, "SOC0=62%, R0=52mΩ," was selected as the candidate parameter combination. This combination predicted an RMSE of only 0.015V in the sensitive segment, providing the optimal candidate for subsequent error verification. This step reduces the number of complex simulations through the GPR surrogate model, accurately locates high-quality parameters using the EI acquisition function, significantly reduces computational resource consumption, accelerates parameter search speed, and ensures parameter accuracy based on sensitive data, laying an efficient foundation for subsequent iterative optimization.

[0063] In step S5, step S5, "substitute the candidate parameter combination into the battery state-space model to calculate the error, iteratively update the surrogate model based on the error result, and obtain the battery health state estimate", is a closed-loop iterative link to achieve accurate estimation - through error feedback, the model is continuously optimized to ensure that the parameters converge to the true value, and finally the SOH is derived. Based on the above, firstly, the candidate parameter combination "SOC0=62%, R0=52mΩ" generated in step S4 is substituted into the Rint model constructed in step S2, and the discharge sensitive segment current data (I=30A) selected in step S3 is input to calculate the predicted voltage: According to the Rint model formula U_pred=U0-I×R0, from SOC0=62%, U0=3.22V is obtained from the table. Substituting I=-30A and R0=52mΩ (0.052Ω), U_pred=3.22+ 30×0.052 / 100=3.22+0.0156=3.2356V (100 is the number of series-connected single-cell battery modules). Next, the root mean square error (RMSE) between the predicted voltage and the actual voltage of the sensitive segment (e.g., the average of 3.4V in the 3.2V-3.6V range) is calculated, and the current error value is 0.018V. The parameter combination and corresponding RMSE were then added to the training set. Following the requirement of "iterative updates to improve stability," the GPR surrogate model from step S4 was retrained using the updated training set. The process of "selecting optimal parameters for the acquisition function → error calculation → model update" was repeated, with the iteration termination condition set as "RMSE difference less than 0.002V for 3 consecutive rounds." After 5 iterations, the optimal parameter combination "SOC0=61.5%, R0=51.8mΩ" was obtained. Combining this with the mapping relationship between SOH and R0 in lithium iron phosphate batteries (e.g., R0=51.8mΩ corresponds to SOH=92%), the battery health state estimate was output as 92%. This step forms a closed-loop iteration through error feedback, continuously optimizing the surrogate model and parameter accuracy, ensuring that the final SOH estimate converges to the true level. Simultaneously, the termination condition balances accuracy and efficiency, meeting the dual requirements of real-time performance and accuracy for automotive BMS.

[0064] In one embodiment, the step of acquiring time-series data of battery operation and dividing it into multiple consecutive time segments includes:

[0065] S10. Collect raw operating parameter data during battery operation;

[0066] S11. Based on a fixed time window or battery charging / discharging phase, the original operating parameter data is divided into multiple continuous time segments.

[0067] In this embodiment, it is necessary to first obtain the original parameters that truly reflect the battery's operating condition, and then select a reasonable division method based on the battery characteristics to provide accurate data units for subsequent sensitive data screening. This is implemented in two steps: The first step is to collect the original operating parameter data during battery operation. Taking the lithium iron phosphate battery for electric vehicles mentioned above as an example, relying on the high-precision sensors mounted on the vehicle battery management system (BMS), the core parameters of the battery during actual driving are collected in real time—current data is obtained through a Hall current sensor connected in series in the battery circuit, with a sampling frequency of 1Hz (i.e., 1 time per second), covering a dynamic range from 300A (maximum discharge current, corresponding to high-speed driving or acceleration conditions) to -200A (maximum charging current, corresponding to fast charging or energy recovery conditions); voltage data is obtained through a voltage acquisition chip connected in parallel across the battery cells, also sampled at a 1Hz frequency, covering the range from 2.5V (discharge cutoff voltage) to 3.65V (charging cutoff voltage); temperature data (collected through an NTC temperature sensor, ranging from -20℃ to 60℃) is supplemented as auxiliary parameters. Data was collected continuously for about three months, resulting in a raw dataset containing approximately 311,000 sets of "current-voltage-temperature" data. Each set of data was timestamped to ensure the integrity of the time sequence and to provide a continuous data source for subsequent segmentation.

[0068] The second step is to segment the raw data based on a fixed time window or battery charging / discharging stages. Considering the significant fluctuations in battery operating conditions in the scenario described above (different states during vehicle operation such as acceleration, deceleration, and idling), the "battery charging / discharging stage" segmentation method is prioritized, while "fixed time window" is used as supplementary verification: For the charging process, based on voltage and current change characteristics, the data is divided into charging segments with different SOC ranges (low SOC (0-45%), medium SOC (45%-85%), high SOC range (85-100%, voltage rises from 2.5V to 3.6V), and rest time segments before and after charging (current is small and stable). The charging phase corresponds to three sets of data: low SOC (0-45%), medium SOC (45%-85%), and high SOC (85-100%). The time intervals before and after charging correspond to two sets of data. For the discharge process, it is divided into pulse discharge segments with different SOC intervals (SOC in the 0-45%, 45%-85%, and 85-100% ranges, with a current difference ≥100A between adjacent frames), corresponding to 3+ sets of data. If the operating conditions are stable during a certain discharge period (e.g., vehicle idling, battery stationary), a fixed 5-minute time window is used, with each segment corresponding to 30 sets of data. Ultimately, 38+ continuous time segments are obtained, each focusing on battery parameter changes under a single operating condition. This highly matches the sensitivity calculation requirements based on the Rint model in subsequent steps, ensuring that the data segments accurately reflect the changing characteristics of the target parameters (initial SOC, ohmic internal resistance). This embodiment ensures that the data can accurately reflect the battery's operating status by acquiring core parameters with high precision and full time sequence, avoiding estimation deviations caused by missing or incorrect data. It also selects the charging and discharging stage division method based on the operating conditions, allowing the segments to focus on a single characteristic, providing high-quality units for subsequent sensitive data screening, while supplementing with a fixed window to ensure coverage of all scenarios and improve the overall process applicability.

[0069] In one embodiment, the step of specifying the target parameter to be estimated and constructing a battery state-space model includes:

[0070] S20. Define the target parameters to be estimated, including the initial charge state and the ohmic internal resistance;

[0071] S21. Establish a battery state-space model based on the target parameters.

[0072] In this embodiment, it is necessary to first identify the key parameters that are strongly correlated with the state of battery health (SOH), and then establish a model that can accurately describe the mapping relationship between the parameters and the data, so as to provide a benchmark for subsequent sensitivity calculation and error analysis. Specifically, it is implemented in two steps: The first step is to define the target parameters to be estimated, the core of which are the initial state of charge (SOC0) and the ohmic internal resistance (R0). Combining the time-series data of electric vehicle lithium iron phosphate batteries collected in the above scenario (including segments of charging / discharging at different SOC intervals, and resting before and after charging), these two parameters directly determine the estimation accuracy of SOH: Initial SOC0 reflects the proportion of the battery's remaining capacity at the beginning of each data segment. Taking the charging segment above as an example, at the beginning of this segment, the remaining capacity of the battery after discharge corresponds to approximately 60% SOC0. This parameter determines the benchmark for voltage changes within the segment (e.g., when SOC0=60%, the open-circuit voltage U0 of the lithium iron phosphate battery is approximately 3.2V); Internal resistance R0 is the core indicator of battery aging. A new battery has an R0 of approximately 50mΩ. As the number of cycles increases, electrode material wear causes R0 to gradually increase. When R0 increases to above 65mΩ, the battery SOH will drop below 80% (a warning is needed). Furthermore, R0 directly affects the voltage drop calculation (e.g., when the discharge current is 30A, R0=50mΩ will produce a voltage drop of 0.015V, with 100 series-connected single-cell modules). These two parameters can transform the abstract SOH into a quantifiable and computable physical quantity, providing a clear direction for subsequent model construction.

[0073] The second step is to establish a battery state-space model based on the target parameters. Referring to the requirement of "lightweight computing for the vehicle BMS" in the scenario above, and combining the model selection principle of "adapting to real-time performance," the Rint model (ideal voltage source + internal resistance series structure) is prioritized as the state-space model. The core equation of this model is: U_meas = U0 - I×R0, where U_meas is the actual terminal voltage collected in the previous step (e.g., 3.6V~2.8V during the discharge phase), I is the collected current data (e.g., 30A during the discharge phase), U0 is the open-circuit voltage obtained from the table by looking up the initial SOC0 (e.g., U0=3.2V when SOC0=60%), and R0 is the internal resistance to be estimated. Taking the discharge stage as an example, substituting I=30A and SOC0=60% (corresponding to U0=3.2V) into the model, and assuming R0=50mΩ, the predicted voltage U_pred=3.2 +30×0.05 / 100=3.2+0.015=3.215V (example data needs to be calibrated in conjunction with actual conditions; the core is to demonstrate the mapping between parameters and voltage, and the number of series-connected modules is 100). This predicted voltage can be compared with the actual collected U_meas to calculate the error. Meanwhile, the Rint model has a simple structure, requires no complex RC network, and its computation time is only 1 / 5 of the Thevenin model, fully meeting the real-time requirements of automotive BMS. It can also accurately correlate the target parameters (SOC0, R0) with the current-voltage data collected above, providing reliable model support for the sensitivity calculation in the subsequent step S3. This embodiment focuses on the two core parameters, initial SOC0 and internal resistance, to avoid irrelevant parameters interfering with the estimation accuracy. It selects the lightweight Rint model, which not only ensures the accuracy of the mapping between parameters and data, but also reduces the computational burden on the vehicle, laying an efficient and reliable foundation for subsequent sensitive data screening and proxy model optimization.

[0074] In one embodiment, the step of performing sensitivity calculations on each consecutive time segment based on the battery state-space model and filtering out sensitive data segments sensitive to the target parameter includes:

[0075] S30. Perform sensitivity calculations on each continuous time segment in conjunction with the battery state-space model to obtain sensitivity indices;

[0076] S31. Determine the sensitivity of each consecutive time segment based on the sensitivity index, and select the initial data segment;

[0077] S32. Select sensitive data segments from the initial data segments based on their sensitivity level.

[0078] In this embodiment, high information density segments are accurately located by quantifying the response of data to target parameters. This is implemented in three steps: The first step is to perform sensitivity calculations on each continuous time segment using the battery state-space model to obtain sensitivity indices. Taking the Rint model (U_measU0-I×R0) and 38 time segments (including constant current charging, constant voltage charging, and constant current discharging) constructed above as an example, algebraic sensitivity analysis is used to calculate the partial derivative of the output voltage with respect to the target parameters (initial SOC0, internal resistance R0) in each segment as the sensitivity index. For the constant current discharge segment (I=30A), when R0 increases from 50mΩ to 51mΩ, the voltage change is 0.0003V (due to the increased voltage drop), so the sensitivity index of this segment to R0 is 0.03V / mΩ; while in the constant voltage charging segment (I decreases from 20A to 5A), the same change in R0 only results in a voltage change of 0.005V, and the sensitivity index is 0.005V / mΩ. For an initial SOC0, during the constant current charging segment, for every 1% increase in SOC0, U0 increases by 0.01V, corresponding to a voltage change of 0.01V, with a sensitivity index of 0.01V / %. However, during the resting segment, changes in SOC0 have minimal impact on the voltage, and the sensitivity index is close to 0. Through this calculation, a two-dimensional index of "sensitivity to SOC0 + sensitivity to R0" is generated for each segment.

[0079] The second step is to determine the sensitivity of each segment based on sensitivity indicators and screen the initial data segments. A dual threshold is set: sensitivity to R0 ≥ 0.01V / mΩ, and sensitivity to SOC0 ≥ 0.005V / %. The indicators of 38 segments are compared with the thresholds. Constant current charge / discharge segments (e.g., 15-minute constant current charging segment, 25-minute constant current discharging segment) all meet the threshold requirements. Only two segments in the constant voltage charging segment barely meet the requirements, and all segments in the resting segment fail to meet the requirements. Finally, 22 initial data segments are selected, initially eliminating the low-sensitivity resting segment and most of the constant voltage segment data. The third step is to select sensitive data segments from the initial data segments based on their sensitivity. The 22 initial segments are weighted and sorted according to their sensitivity indicators (R0 weight 0.6, SOC0 weight 0.4), and the top 10 segments with the highest comprehensive scores are selected: 6 segments are in the constant current discharging segment (the highest-scoring segment has a sensitivity to R0 of 0.035V / mΩ), and 4 segments are in the constant current charging segment (a sensitivity to SOC0 of 0.012V / %). These sensitive segments accurately reflect changes in target parameters. For example, in a constant current discharge segment, the voltage fluctuation reaches 0.017V when R0 changes by 0.5mΩ, which is much higher than the 0.002V in the constant voltage segment. This provides high-information-density input for the surrogate model optimization in subsequent step S4, forming a logical closed loop with the Rint model and target parameters constructed earlier. This embodiment accurately selects segments that significantly respond to target parameters by quantifying sensitivity indicators, eliminating low-sensitivity redundant data, reducing noise interference, and providing high-quality input for subsequent parameter optimization. This improves the accuracy of SOH estimation from the source and adapts to the data efficiency utilization requirements of automotive scenarios.

[0080] In one embodiment, before the step of using a proxy model to optimize the search for the target parameters based on the sensitive data fragment and generate candidate parameter combinations, the following steps are included:

[0081] S410, Construct the initial agent model;

[0082] S420. Generate initial parameter combinations of multiple target parameters, substitute them into the battery state space model, simulate the predicted voltage based on the current data of the sensitive data segment, and calculate the root mean square error between the predicted voltage and the actual voltage of the sensitive data segment.

[0083] S430. Initialize and train the initial agent model using the initial parameter combination and the corresponding error to obtain a target agent model with mapping capability between target parameters and error.

[0084] In this embodiment, considering the "efficient learning with small sample sizes" requirement in the scenario described above, a Gaussian Process Regression (GPR) model is selected as the initial surrogate model. The GPR model has advantages in handling nonlinear relationships and output prediction confidence intervals, especially in adapting to the complex mapping between target parameters (initial SOC0, ohmic internal resistance R0) and error values—SOC0 ranges from 40% to 90% (corresponding to the initial capacity of the sensitive segment mentioned above), and R0 ranges from 40mΩ to 90mΩ (covering the internal resistance change from new to aged batteries). The relationship between these two parameters and the root mean square error (RMSE) exhibits nonlinear characteristics. The GPR kernel function (such as the quadratic exponential kernel) can effectively capture this relationship, and the model has few parameters (only noise variance and length scale need to be optimized), resulting in short training time, which aligns with the lightweight requirements of automotive BMS. The second step is to generate the initial parameter combination and calculate the error. Based on the 10 sensitive data segments (6 constant current discharge segments and 4 constant current charging segments) selected above, 30 initial parameter combinations were generated using the Latin hypercube sampling method to ensure uniform coverage of the parameter space: for example, combination 1 is "SOC0=60%, R0=50mΩ", combination 2 is "SOC0=65%, R0=55mΩ", etc. Substitute each combination into the Rint model constructed in step S2, input the current data of the sensitive segment (e.g., constant current discharge segment I=30A, constant current charging segment I=-20A), and simulate to obtain the predicted voltage: Taking combination 1 as an example, in the constant current discharge segment, U_pred=U0 (SOC0=60% corresponds to 3.2V)I×R0=3.2+30×0.05 / 100=3.215V, compared with the actual average voltage of 3.4V in this segment, the RMSE is calculated to be 0.0185V; the predicted voltage of the combination "SOC0=70%, R0=60mΩ" in the same segment is 3.3V (U0=3.3V)30×0.06 / 100=0.051V, RMSE=0.08V. Finally, 30 sets of "parameter combination-RMSE" sample pairs are obtained, forming the initial training set.

[0085] The third step is to initialize and train the target surrogate model. Thirty sets of sample pairs are input into the initial GPR model, and the kernel function parameters are optimized using maximum likelihood estimation: the length scale is set to 5 (to control the range of parameter influence on the error), and the noise variance is set to 0.001 (to filter measurement noise interference). After training, the target surrogate model has accurate mapping capabilities—inputting any combination of parameters (e.g., "SOC0=62%, R0=52mΩ"), it can output a predicted RMSE of 0.016V, with a deviation of only 0.001V from the actual simulation value of 0.015V. This model replaces the time-consuming Rint model simulation (reducing the single prediction time from 20ms to 1ms) and can reflect the trend of parameter changes on the error (e.g., RMSE increases as R0 increases), providing an efficient tool for the subsequent optimization search in step S4, forming a logical closed loop with the sensitive data fragments and target parameters mentioned earlier. This embodiment constructs a surrogate model using GPR and generates uniform samples using Latin hypercube sampling. After training, it achieves an accurate mapping between parameters and errors, significantly reducing computational costs while ensuring mapping accuracy, thus laying an efficient and reliable foundation for subsequent parameter optimization searches.

[0086] In one embodiment, the step of using a proxy model to optimize the search for the target parameters based on the sensitive data fragment and generate candidate parameter combinations further includes:

[0087] S420, Set the acquisition function for the target agent model;

[0088] S421. Evaluate the potential combinations of target parameters using the acquisition function;

[0089] S422. Select parameter combinations whose evaluation results of the acquisition function meet the preset conditions as candidate parameter combinations.

[0090] In this embodiment, the first step is to set the acquisition function of the target surrogate model. Combining the Gaussian Process Regression (GPR) target surrogate model trained above (which can map the relationship between the initial SOC0, ohmic internal resistance R0, and root mean square error RMSE), the Expected Improvement (EI) is selected as the acquisition function. The core advantage of the EI function is that it balances "exploration" and "development": it focuses on the currently known low-error region (development) while also taking into account the unknown region where there may be better solutions (exploration). The formula is EI(x)=E[max(f(x*)-f(x),0)], where x* is the current optimal parameter combination and f(x) is the prediction error. For the target parameter range (SOC0: 40%-90%, R0: 40mΩ-90mΩ) in the scenario above, the EI function can quantify the probability and magnitude of a parameter combination "improving the current optimal solution". For example, for combinations at the edge of the known low error zone, the EI value will increase due to the "high probability of small improvement"; for combinations at the edge of the parameter space, the EI value will remain at a certain value due to the "possibility of discovering a completely new optimal solution", avoiding getting trapped in local optima. The second step is to evaluate the potential combinations of target parameters through a data collection function. Based on the 10 sensitive data segments mentioned above, 500 sets of potential parameter combinations are randomly generated in the parameter space (such as "SOC0=61%, R0=51mΩ", "SOC0=58%, R0=53mΩ", etc.). Each combination is input into the GPR model to obtain the predicted RMSE, and then substituted into the EI function to calculate the evaluation value. Using the current optimal combination "SOC0=60%, R0=50mΩ" (predicted RMSE=0.02V) as a benchmark, a potential combination "SOC0=62%, R0=52mΩ" has a predicted RMSE of 0.015V and an EI value of 0.004 (indicating a high probability of improving the current optimal combination). Another combination "SOC0=85%, R0=80mΩ" (at the edge of the parameter space) has a predicted RMSE of 0.05V, but due to the high uncertainty of the model's prediction for this region, the EI value still reaches 0.002 (retaining exploratory value). Through batch evaluation, the EI value distribution of 500 potential combinations was obtained.

[0091] The third step is to select parameter combinations that meet preset conditions as candidate parameter combinations. The preset conditions are set as "EI value ranking in the top 5% and predicted RMSE < 0.02V". From 500 groups, 25 candidate groups are selected, and then the combination with the highest EI value, "SOC0 = 61.5%, R0 = 51.8mΩ", is selected as the final candidate. This combination has a predicted RMSE of 0.014V and an EI value of 0.005 in the GPR model, which is both within the known low error range (development) and has significant improvement potential compared to the current optimal solution. This candidate parameter combination is highly compatible with the sensitive data segments (such as the constant current discharge segment) and the Rint model mentioned above. Substituting it into the model allows for accurate voltage prediction, providing an optimal starting point for error calculation and iterative updates in subsequent step S5, forming a logical closed loop. This embodiment explores and develops the EI acquisition function in a balanced manner, accurately evaluates potential parameter combinations, and selects candidate parameters that combine improvement potential and reliability. This avoids local optima and improves search efficiency, providing a high-quality starting point for subsequent iterative optimization.

[0092] In one embodiment, the step of substituting the candidate parameter combination into the battery state-space model for error calculation, and iteratively updating the surrogate model based on the error results to obtain the battery health state estimate includes:

[0093] S50. Substitute the candidate parameter combination into the battery state space model, and calculate the predicted voltage by inputting the current data of the sensitive data segment.

[0094] S51. Calculate the root mean square error of the actual voltage of the sensitive data segment predicted voltage, and use it as the error value;

[0095] S52. Add the candidate parameter combinations and corresponding error values ​​to the training set.

[0096] S53. Retrain the agent model using the updated training set, repeat the iteration until the preset termination condition is met, and output the battery health status estimate.

[0097] In this embodiment, "substituting candidate parameter combinations into the battery state-space model to calculate the error, iteratively updating the surrogate model and obtaining the battery health state estimate" is the closed-loop link to achieve accurate estimation. The parameters and model are continuously optimized through error feedback, and it is implemented in four steps: The first step is to substituting candidate parameter combinations to calculate the predicted voltage. Taking the candidate parameter combination "SOC0=61.5%, R0=51.8mΩ" selected above as an example, substitute it into the Rint model constructed in step S2 (U_meas=U0-I×R0), and input the current data of the 10 sensitive data segments selected in step S3: For the constant current discharge segment (I=30A), from the table, U0=3.21V is obtained from SOC0=61.5%, and the predicted voltage U_pred=3.21+30×0.0518 / 100=3.21+0.01554=3.22554V is obtained by substituting into the formula; for the constant current charging segment (I=20A), U_pred=3.21 - 20×0.0518 / 100=3.21-0.01036=3.19964V (the example data needs to be calibrated in conjunction with actual conditions, and the core reflects the mapping between parameters and voltage). The calculation process covers all sensitive segments to ensure that the predicted voltage reflects the battery characteristics under various operating conditions. The second step is to calculate the root mean square error (RMSE) as the error value. The predicted voltage is then compared with the actual voltage of the sensitive data segments: the average actual voltage of the constant current discharge segment is 3.4V, and the sum of squares of the errors for each segment is Σ(3.22554-3.4). 2 =18.6 (simplified calculation); the actual average voltage of the constant current charging section is 3.3V, and the sum of squares of the errors in a single section is Σ(3.19964-3.3). 2 =12.6. Taking the square root of the mean of the sum of squared errors of the 10 segments, we get the overall RMSE = 0.016V. This value serves as the error index for the current parameter combination, reflecting the degree of deviation between the parameters and the true values.

[0098] The third step is to update the training set and retrain the surrogate model. "SOC0=61.5%, R0=51.8mΩ" and the corresponding RMSE=0.016V are added to the original training set (containing 30 samples), forming an updated training set of 31 samples. The GPR surrogate model from step S4 is retrained using this dataset, optimizing the kernel function parameters (length scale adjusted to 4.8, noise variance maintained at 0.001), enabling the model to more accurately map the relationship between parameters and error—the new model's prediction RMSE for the parameter combination "SOC0=61%, R0=51mΩ" is 0.015V, reducing the deviation from the actual simulated value of 0.014V to 0.001V, thus improving mapping accuracy. The fourth step is to repeat the iteration until the termination condition is met, outputting the SOH estimate. The termination condition is set as "the difference in RMSE over three consecutive iterations is <0.002V". Candidate parameter combinations are regenerated using the method in step S4 (e.g., the optimal combination for the second round is "SOC0=61.2%, R0=51.5mΩ", RMSE=0.015V; the optimal combination for the third round is "SOC0=61.3%, R0=51.6mΩ", RMSE=0.0145V). Since the difference in all three rounds is <0.002V, the termination condition is met. Based on the final parameter R0=51.6mΩ, and considering the mapping relationship between SOH and internal resistance of the lithium iron phosphate battery (R0=50mΩ corresponds to SOH=100%, and SOH decreases by 0.5% for every 1mΩ increase), SOH=100%-(51.6-50)×0.5%=99.2%. This value is output as the battery health status estimation result, forming a complete logical closed loop with the sensitive data, model, and parameters mentioned earlier. This embodiment optimizes the surrogate model through error feedback iteration, making the parameters continuously approach the true value. It balances accuracy and efficiency by combining termination conditions, and the final output SOH estimate has small error and high reliability, which is fully adapted to the real-time monitoring requirements of vehicle BMS.

[0099] refer to Figure 2 A battery health status estimation device, comprising:

[0100] The data acquisition and segmentation module 100 is used to acquire the timing data of battery operation and divide it into multiple continuous time segments;

[0101] The parameter model construction module 200 is used to define the target parameters to be estimated and construct the battery state-space model.

[0102] The sensitive segment filtering module 300 is used to perform sensitivity calculations on each continuous time segment based on the battery state space model, and filter out sensitive data segments that are sensitive to the target parameters.

[0103] The parameter combination generation module 400 is used to optimize the search for the target parameters based on the sensitive data fragment using a proxy model, and generate candidate parameter combinations.

[0104] The health status estimation module 500 is used to substitute the candidate parameter combination into the battery state space model to calculate the error, and iteratively update the surrogate model according to the error result to obtain the battery health status estimate.

[0105] Furthermore, the aforementioned data acquisition and segmentation module 100 includes:

[0106] The key parameter acquisition unit is used to collect raw operating parameter data during battery operation.

[0107] The data segmentation unit is used to divide the original operating parameter data into multiple continuous time segments based on a fixed time window or battery charging and discharging phase.

[0108] Furthermore, the aforementioned parameter model construction module 200 includes:

[0109] The target parameter definition unit is used to define the target parameters to be estimated, including the initial charge state and the ohmic internal resistance.

[0110] A state-space modeling unit is used to establish a battery state-space model based on the target parameters.

[0111] Furthermore, the aforementioned sensitive segment filtering module 300 includes:

[0112] The sensitivity calculation unit is used to perform sensitivity calculations on each continuous time segment in conjunction with the battery state-space model to obtain sensitivity indices.

[0113] A high-sensitivity segment screening unit is used to determine the sensitivity level of each consecutive time segment based on the sensitivity index and to screen out the initial data segments.

[0114] The sensitive data selection unit is used to select sensitive data segments from the initial data segments based on their sensitivity level.

[0115] Furthermore, the above parameter combination generation module 400 includes:

[0116] The proxy model building unit is used to build the initial proxy model;

[0117] The initial parameter simulation unit is used to generate multiple sets of initial parameter combinations of target parameters, substitute them into the battery state space model, simulate the predicted voltage based on the current data of the sensitive data segment, and calculate the root mean square error between the predicted voltage and the actual voltage of the sensitive data segment.

[0118] The surrogate model training unit is used to initialize and train the initial surrogate model using the initial parameter combination and the corresponding error, so as to obtain a target surrogate model with mapping capability between target parameters and error.

[0119] Furthermore, the parameter combination generation module 400 also includes:

[0120] The data acquisition function setting unit is used to set the data acquisition function of the proxy model;

[0121] A parameter combination evaluation unit is used to evaluate potential combinations of target parameters through the acquisition function;

[0122] The candidate parameter selection unit is used to select parameter combinations whose evaluation results of the acquisition function meet preset conditions as candidate parameter combinations.

[0123] Furthermore, the aforementioned health status estimation module 500 includes:

[0124] The predicted voltage calculation unit is used to substitute the candidate parameter combination into the battery state space model and input the current data of the sensitive data segment to calculate the predicted voltage.

[0125] An error calculation unit is used to calculate the root mean square error of the actual voltage of the sensitive data segment predicted voltage, as the error value.

[0126] The dataset update unit is used to add the candidate parameter combinations and their corresponding error values ​​to the training set.

[0127] The surrogate model iterative update unit is used to retrain the surrogate model using the updated training set, repeating the iteration until a preset termination condition is met, and outputting a battery health status estimate.

[0128] Reference Figure 3 This application also provides a computer device, which may be a server, and its internal structure may be as follows: Figure 3As shown. The computer device includes a processor, memory, network interface, and database connected via a bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores operations, computer programs, and the database. The internal memory provides an environment for the operation and execution of the computer programs stored in the non-volatile storage media. The database stores data such as battery health state estimation methods. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements a battery health state estimation method, including the following steps: collecting time-series data of battery operation and dividing it into multiple continuous time segments; identifying the target parameter to be estimated and constructing a battery state-space model; performing sensitivity calculations on each continuous time segment based on the battery state-space model, and filtering out sensitive data segments sensitive to the target parameter; using a surrogate model, optimizing the search for the target parameter based on the sensitive data segments to generate candidate parameter combinations; substituting the candidate parameter combinations into the battery state-space model for error calculation, iteratively updating the surrogate model based on the error results, and obtaining the battery health state estimate.

[0129] One embodiment of this application also provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements a battery health state estimation method, including the following steps: collecting time-series data of battery operation and dividing it into multiple continuous time segments; identifying the target parameter to be estimated and constructing a battery state-space model; performing sensitivity calculations on each continuous time segment based on the battery state-space model, and filtering out sensitive data segments that are sensitive to the target parameter; using a surrogate model, optimizing and searching for the target parameter based on the sensitive data segments to generate candidate parameter combinations; substituting the candidate parameter combinations into the battery state-space model for error calculation, iteratively updating the surrogate model based on the error results, and obtaining an estimated battery health state value.

[0130] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media provided in this application and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0131] The above description is only a preferred embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural changes made based on the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for estimating battery health status, characterized in that, include: Collect timing data of battery operation and divide it into multiple continuous time segments; Define the target parameters to be estimated and construct a battery state-space model; Sensitivity calculations are performed on each continuous time segment based on the battery state-space model. An algebraic sensitivity calculation method is used, with the partial derivative of the output voltage with respect to the target parameter as the sensitivity index. The sensitivity index is obtained by calculating the ratio of the voltage change to the parameter change in each continuous time segment, and sensitive data segments that are sensitive to the target parameter are selected. Using a proxy model, the target parameters are optimized and searched based on the sensitive data fragments to generate candidate parameter combinations; The candidate parameter combination is substituted into the battery state space model to calculate the error. The surrogate model is iteratively updated based on the error result to obtain the battery health state estimate. Before the step of using a proxy model to optimize the search for the target parameters based on the sensitive data fragment and generate candidate parameter combinations, the following steps are included: Build the initial agent model; Multiple sets of initial parameter combinations of target parameters are generated and substituted into the battery state-space model. The predicted voltage is obtained by simulation based on the current data of the sensitive data segment. The root mean square error between the predicted voltage and the actual voltage of the sensitive data segment is calculated. The Latin hypercube sampling method is used to generate multiple sets of initial parameter combinations. The value ranges of the initial charge state and ohmic internal resistance in the target parameters cover the starting capacity range corresponding to the sensitive data segment and the internal resistance change range of the battery throughout its entire life cycle, respectively. The initial proxy model is initialized and trained using the initial parameter combination and the corresponding error to obtain a target proxy model that has the ability to map between target parameters and error.

2. The battery health status estimation method according to claim 1, characterized in that, The step of collecting time-series data of battery operation and dividing it into multiple consecutive time segments includes: Collect raw operating parameter data during battery operation; Based on a fixed time window or battery charging / discharging phase, the original operating parameter data is divided into multiple continuous time segments.

3. The battery health status estimation method according to claim 1, characterized in that, The steps of identifying the target parameters to be estimated and constructing a battery state-space model include: Define the target parameters to be estimated, including the initial charge state and the ohmic internal resistance; A battery state-space model is established based on the target parameters.

4. The battery health status estimation method according to claim 1, characterized in that, The step of performing sensitivity calculations on each continuous time segment based on the battery state-space model and filtering out sensitive data segments that are sensitive to the target parameter includes: Sensitivity calculations are performed on each consecutive time segment using the battery state-space model to obtain sensitivity indices. The sensitivity level of each consecutive time segment is determined based on the aforementioned sensitivity index, and the initial data segments are selected. Sensitive data segments are selected from the initial data segments based on their sensitivity level.

5. The battery health status estimation method according to claim 1, characterized in that, The step of using a proxy model to optimize the search for the target parameters based on the sensitive data fragment and generate candidate parameter combinations includes: Define the acquisition function for the target agent model; The potential combinations of target parameters are evaluated using the acquisition function. Select parameter combinations whose evaluation results of the acquisition function meet the preset conditions as candidate parameter combinations.

6. The battery health state estimation method according to claim 1, characterized in that, The step of substituting the candidate parameter combination into the battery state-space model for error calculation, and iteratively updating the surrogate model based on the error results to obtain the battery health state estimate includes: The candidate parameter combination is substituted into the battery state space model, and the current data of the sensitive data segment is input to calculate the predicted voltage. Calculate the root mean square error between the predicted voltage and the actual voltage of the sensitive data segment, and use it as the error value; The candidate parameter combinations and their corresponding error values ​​are added to the training set. The agent model is retrained using the updated training set, and the iteration is repeated until a preset termination condition is met, and the battery health status estimate is output.

7. A battery health status estimation device, used to implement the method according to any one of claims 1-6, characterized in that, include: The data acquisition and segmentation module is used to collect time-series data of battery operation and divide it into multiple continuous time segments; The parameter model building module is used to define the target parameters to be estimated and build a battery state-space model. The sensitive segment filtering module is used to perform sensitivity calculations on each continuous time segment based on the battery state space model, and filter out sensitive data segments that are sensitive to the target parameters. The parameter combination generation module is used to optimize the search for the target parameters based on the sensitive data fragment using a proxy model, and generate candidate parameter combinations. The health status estimation module is used to substitute the candidate parameter combination into the battery state space model to calculate the error, and iteratively update the surrogate model based on the error result to obtain the battery health status estimate.

8. A battery health status estimation device, characterized in that, include: processor; Memory, used to store computer programs; The processor, when executing the computer program, implements the battery health state estimation method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the battery health state estimation method according to any one of claims 1 to 6.

Images