SOH online estimation method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention discloses an SOH online estimation method and device, electronic equipment and a storage medium, and relates to the technical field of battery SOH online estimation, and the method comprises the steps: collecting battery operation data in real time, carrying out the preprocessing, intelligently detecting a high / low SOC standing point through a multi-mode standing point recognition mechanism, and carrying out the recognition of the high / low SOC standing point; and dynamically calibrating the SOC value in combination with the OCV-SOC relation curve. And matching and screening a standing point combination through a cross-day state continuation mechanism, calculating SOC variation, and deriving capacity attenuation based on a physical rule (such as ampere-hour integral) to generate an SOH estimated value and confidence evaluation. Statistical features are extracted from historical data to train a random forest model to predict an SOH value, and uncertainty is quantified through out-of-bag estimation. And according to a physical rule and confidence and uncertainty output by machine learning, carrying out moving average processing on historical data by adopting adaptive weighted fusion or attenuation weight filtering to obtain an SOH value. The problems of low data utilization rate and poor reliability in the prior art are effectively solved.

Description

Technical Field

[0001] This invention relates to the field of online battery health state (SOH) estimation technology, and particularly to an online SOH estimation method, apparatus, electronic device, and storage medium. Background Technology

[0002] With the large-scale participation of energy storage power stations in grid frequency regulation services, batteries are in a state of frequent and irregular interleaved charging and discharging for a long time. The complexity of frequency regulation conditions poses a severe challenge to the accurate estimation of battery state of health (SOH).

[0003] However, traditional SOH estimation methods have significant limitations when dealing with such complex scenarios: physical model-based methods (such as the ampere-hour integral method and the OCV-SOC curve method) have clear physical meanings, but in frequency modulation conditions with discontinuous charging and discharging, interleaved charging and discharging, and scarce complete cycles, it is difficult to capture effective data segments, resulting in a surge in estimation failure rate and low data utilization; while data-driven methods (such as neural networks and random forests) can mine nonlinear relationships in massive amounts of data, but due to the lack of physical constraints, they have poor interpretability and are prone to deviating from physical common sense when the data quality is poor or new operating conditions are encountered, raising doubts about their reliability.

[0004] Furthermore, traditional full-capacity calibration methods fail because full charge and discharge tests are difficult to perform at FM stations; conventional ampere-hour integration methods exhibit rapidly diverging errors with high-frequency cycles; single-cycle OCV methods lack sufficient accuracy; ICA / DVA analysis cannot be applied online; and Kalman filtering joint estimation methods fail due to high-frequency drift of model parameters. Existing technologies either rely on ideal segments or are limited by static models, failing to simultaneously address the dynamic and fragmented characteristics of FM operating conditions.

[0005] Therefore, there is an urgent need for a highly reliable online SOH estimation method that can integrate the advantages of physical rules and data-driven approaches and adapt to complex working conditions. Summary of the Invention

[0006] The embodiments of this invention provide an online SOH estimation method to address the problems of low data utilization, high estimation failure rate, and inability to adapt to high-frequency dynamic operating conditions in existing technologies under fragmented frequency modulation scenarios. The technical solution is as follows: According to one aspect of the present invention, an online SOH estimation method is provided, the method comprising: real-time acquisition and preprocessing of battery operating data; intelligent detection of SOC resting points based on the operating data using a multi-mode resting point identification mechanism to obtain a set of resting points; the operating data including voltage, current, SOC, temperature, and timestamp; the preprocessing including denoising, interpolation, alignment, and filtering; dynamic calibration of the set of resting points using a pre-established OCV-SOC relationship curve; and matching and filtering of the resting points using a set cross-day state continuation mechanism to obtain paired resting point groups. The system combines the change in SOC with the settling point information, including type, time, and SOC. It calculates capacity decay based on the change in SOC using physical rules to obtain an estimated SOH value and assess its confidence level. Simultaneously, it predicts the SOH value by training a random forest model and quantifies the uncertainty index predicted by the model through out-of-bag estimation. The physical rules include ampere-hour integrals. It obtains the SOH value by weighting the estimated and predicted SOH values ​​using adaptive weighted fusion based on the confidence level and the uncertainty index, or by performing a moving average processing on historical data using decay weighted filtering.

[0007] In one embodiment, real-time acquisition and preprocessing of battery operating data are achieved through the following steps: using high-precision sensors to continuously monitor various key operating parameters of the battery in real time, and using filtering techniques to denoise the operating parameters; the filtering techniques include moving average filtering or Kalman filtering; using interpolation methods to fill in missing values ​​in the operating parameters, aligning the data according to the timestamps of the operating parameters, and smoothing the operating parameters through low-pass filtering.

[0008] In one embodiment, a multi-mode resting point identification mechanism intelligently detects SOC resting points based on the operating data to obtain a set of resting points through the following steps: The operating data is traversed; when the battery abruptly changes from a high-current operating state to zero current and remains rested until a stable threshold is reached, if the SOC value is higher than a set percentage at this time, it is identified as a high SOC resting point; the stable threshold includes the duration of the state; if the SOC value is lower than a set percentage at this time, it is identified as a low SOC resting point, and the high SOC resting points and low SOC resting points are filtered and verified to obtain a set of resting points; the high-current operating state refers to a current absolute value greater than a set value.

[0009] In one embodiment, the set of stationary points is dynamically calibrated using a pre-established OCV-SOC relationship curve, and the stationary points are matched and filtered using a set cross-day state continuation mechanism to obtain paired stationary point combinations and SOC changes. This is achieved through the following steps: using the pre-established OCV-SOC relationship curve, each stationary point in the set is dynamically calibrated; the SOC value of each stationary point is obtained by querying the OCV-SOC relationship table using an interpolation algorithm; a cross-day state continuation mechanism is designed to store the information of the stationary points; a bidirectional pairing strategy is used to match high SOC stationary points and low SOC stationary points to obtain multiple pairs of stationary point combinations; the SOC difference between each pair of stationary point combinations is calculated; and the SOC change is calculated for combinations whose SOC differences meet set conditions.

[0010] In one embodiment, the capacity decay is calculated based on the SOC change using physical rules, and the estimated SOH value is obtained and its confidence level is assessed through the following steps: The capacity decay of the battery is calculated based on the SOC change using physical rules and the resting point combination; the current estimated SOH value is derived by combining the battery's state data; the state data includes initial capacity and charge / discharge efficiency; a confidence assessment system is set up according to confidence factors; the estimated SOH value is comprehensively assessed based on the confidence assessment system, and a scaling factor and mapping function are introduced to obtain the confidence level; the confidence factors include measurement error, model uncertainty index, and data quality.

[0011] In one embodiment, the prediction of SOH (State of Health) by training a random forest model and the quantification of the uncertainty index predicted by the model through out-of-bag estimation are achieved through the following steps: extracting statistical features closely related to battery aging from the historical operating data of the battery and constructing a feature vector; the statistical features include voltage features, current features, SOC (State of Charge) features, temperature features, and energy features; using the SOH values ​​that meet the set criteria in the historical operating data as training labels, a random forest regression model is trained using a machine learning algorithm, and the model hyperparameters are optimized through grid search and cross-validation; the feature vector is input into the trained random forest regression model to obtain the SOH prediction value, and the uncertainty index predicted by the model is quantified using the out-of-bag estimation or the prediction variance between trees built into the random forest.

[0012] In one embodiment, the SOH value is obtained by adaptively weighted fusion of the estimated and predicted SOH values ​​based on the confidence level and the uncertainty index, or by performing a moving average of historical data using decaying weighted filtering. This is achieved through the following steps: if only the estimated SOH value is valid, it is used as the SOH value; if only the predicted SOH value is valid, it is used as the SOH value; if both the estimated and predicted SOH values ​​are valid, they are dynamically weighted and fused based on the confidence level and the uncertainty index to obtain the SOH value; if both the estimated and predicted SOH values ​​are invalid, a decaying weighted filtering is used to perform an exponentially decaying moving average of the historical data weights in chronological order to obtain the SOH value.

[0013] According to one aspect of the present invention, an online SOH estimation device is provided, the device comprising: a data acquisition and preprocessing module, configured to acquire and preprocess battery operating data in real time, and intelligently detect SOC resting points based on the operating data using a multi-mode resting point identification mechanism to obtain a set of resting points; the operating data includes voltage, current, SOC, temperature, and timestamp; the preprocessing includes noise reduction, interpolation, alignment, and filtering; and a resting point calibration and matching module, configured to dynamically calibrate the set of resting points using a pre-established OCV-SOC relationship curve, and match and filter the resting points using a set cross-day state continuation mechanism to obtain paired resting point groups. The system includes the sum of SOC changes; the settling point information includes type, time, and SOC; a rule-based dual prediction module is used to calculate capacity decay based on the SOC change using physical rules, obtain an estimated SOH value and assess confidence, and simultaneously predict the predicted SOH value by training a random forest model and quantify the uncertainty index predicted by the model through out-of-bag estimation; the physical rules include ampere-hour integrals; and an online SOH estimation module is used to obtain the SOH value by weighting the estimated SOH value and the predicted SOH value based on the confidence level and the uncertainty index through adaptive weighted fusion, or by performing a moving average processing on historical data through decay weight filtering to obtain the SOH value.

[0014] According to one aspect of the present invention, an electronic device includes at least one processor and at least one memory, wherein computer-readable instructions are stored on the memory; the computer-readable instructions are executed by one or more of the processors to cause the electronic device to implement the online SOH estimation method as described above.

[0015] According to one aspect of the invention, a storage medium stores computer-readable instructions thereon, which are executed by one or more processors to implement the online SOH estimation method as described above.

[0016] The beneficial effects of the technical solution provided by this invention are: In the above technical solution, this invention collects real-time operating data such as battery voltage, current, SOC, temperature, and timestamps. After preprocessing including denoising, interpolation, alignment, and filtering, it intelligently detects high / low SOC resting points using a multi-mode resting point identification mechanism and dynamically calibrates the SOC value using the OCV-SOC relationship curve. A cross-day state continuation mechanism is used to match and filter resting point combinations, calculate the SOC change, and derive capacity decay based on physical rules (such as ampere-hour integration) to generate an estimated SOH value and confidence assessment. Simultaneously, statistical features such as voltage and current are extracted from historical data to train a random forest model to predict the SOH value, and the uncertainty index is quantified through out-of-bag estimation. Finally, based on the confidence and uncertainty index output by the physical rules and machine learning, an adaptive weighted fusion or attenuated weighted filtering is used to process the historical data using a moving average to obtain a highly robust SOH value. This method integrates the advantages of physical constraints and data-driven approaches, overcoming the limitations of traditional methods in terms of low data utilization and poor reliability under frequency modulation conditions, significantly improving estimation accuracy and engineering practicality. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating an online SOH estimation method according to an exemplary embodiment; Figure 2 This is a flowchart illustrating the online SOH estimation method in an application scenario. Figure 3 yes Figure 2 A schematic diagram of the cross-day status continuation mechanism in the corresponding application scenario; Figure 4 yes Figure 2 A diagram illustrating the pairing and filtering of stationary points in corresponding application scenarios; Figure 5 yes Figure 2 A schematic diagram of attenuation weighted filtering in the corresponding application scenario; Figure 6 This is a block diagram illustrating an online SOH estimation apparatus according to an exemplary embodiment; Figure 7 This is a hardware structure diagram of an electronic device according to an exemplary embodiment; Figure 8This is a block diagram illustrating an electronic device according to an exemplary embodiment. Detailed Implementation

[0019] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0020] Those skilled in the art will understand that, unless specifically 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 this disclosure means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0021] This invention provides an online SOH estimation method. By integrating physical rules and a random forest model, it achieves high-precision online estimation of the SOH of energy storage batteries under frequency regulation conditions. This solves the technical problems of low data utilization, poor reliability, and insufficient estimation accuracy of traditional methods in complex charging and discharging scenarios, significantly improving engineering practicality and adaptability. This online SOH estimation method is applicable to online SOH estimation devices, which can be electronic devices. The online SOH estimation method in this invention can be applied to various scenarios, such as online SOH estimation.

[0022] Please see Figure 1 This invention provides an online SOH estimation method, which is applicable to electronic devices.

[0023] In the following method embodiments, for ease of description, the execution subject of each step of the method is an electronic device, but this does not constitute a specific limitation.

[0024] like Figure 1 As shown, the method may include the following steps: Step 110: Collect battery operating data in real time and preprocess it. Use a multi-mode resting point identification mechanism to intelligently detect SOC resting points based on the operating data to obtain a set of resting points.

[0025] In one possible implementation, high-precision sensors are used to continuously monitor various key operating parameters of the battery in real time. Filtering techniques are used to denoise the operating parameters, interpolation methods are used to fill in missing values ​​in the operating parameters, data is aligned according to the timestamps of the operating parameters, and low-pass filtering is used to smooth the operating parameters.

[0026] The operating data includes voltage, current, SOC, temperature, timestamps, etc., and the preprocessing includes noise reduction, interpolation, alignment, filtering, etc. The filtering techniques include moving average filtering or Kalman filtering, etc., none of which are specified here.

[0027] In one possible implementation, the operating data is traversed. When the battery suddenly changes from a high-current operating state to zero current and remains stationary until a stable threshold is reached, if the SOC value is higher than a set percentage, it is identified as a high SOC stationary point. If the SOC value is lower than a set percentage, it is identified as a low SOC stationary point. The high SOC stationary points and low SOC stationary points are then filtered and verified to obtain a set of stationary points.

[0028] Among them, the stability threshold includes the duration of the state; the high current working state refers to the absolute value of the current being greater than the set value. The set percentage for judging the high SOC rest point and the low SOC rest point can be different. The stability threshold, set value, and set percentage can all be set according to the actual application scenario.

[0029] Specifically, high-precision sensors are used to monitor key operating parameters of the battery in real time, such as voltage, current, SOC, temperature, and timestamp. Moving average filtering or Kalman filtering techniques are employed to denoise the raw data, reducing measurement errors and random interference. Interpolation methods (such as linear interpolation and spline interpolation) are used to fill in missing values ​​in the data, ensuring data continuity. Multi-parameter data are aligned according to timestamps to ensure the accuracy of each parameter at the same time point. Low-pass filtering is used to smooth the data, further reducing the impact of high-frequency noise.

[0030] In the above process, this embodiment of the invention uses high-precision sensors to collect multi-dimensional battery operation data in real time, and combines preprocessing techniques such as denoising, interpolation, alignment, and smoothing to ensure the accuracy and continuity of the data, providing a reliable foundation for subsequent analysis. This step effectively improves data quality and reduces estimation errors caused by data problems.

[0031] Step 120: Dynamically calibrate the set of stationary points using the pre-established OCV-SOC relationship curve, and match and filter the stationary points using the set cross-day state continuation mechanism to obtain the paired stationary point combination and SOC change.

[0032] In one possible implementation, the OCV-SOC relationship curve, established experimentally beforehand, is used to dynamically calibrate each stationary point in the stationary point set. The SOC value of each stationary point is obtained by querying the OCV-SOC relationship table using an interpolation algorithm. A cross-day state continuation mechanism is designed to save the stationary point information. A two-way pairing strategy is used to match high-SOC stationary points with low-SOC stationary points to obtain multiple pairs of stationary point combinations. The SOC difference between each pair of stationary point combinations is calculated. For combinations whose SOC difference meets the set conditions, the SOC change is calculated.

[0033] The settling point information includes type, time, SOC, etc., which are not specified here.

[0034] Specifically, when a battery abruptly changes from a high-current operating state (absolute current > 10A) to zero current and remains stationary for a stable threshold (e.g., 1200 seconds) with a SOC value higher than 90%, it is identified as a high SOC resting point. Similarly, when the SOC value is below 20%, it is identified as a low SOC resting point.

[0035] Furthermore, using the pre-established OCV-SOC relationship curve, an interpolation algorithm is employed to perform precise lookups in the discrete OCV-SOC relationship table to obtain the true SOC value for each stationary point. A state memory system with timeout protection (e.g., 7 days) is designed to extend the valid stationary point information to subsequent calculation cycles. All possible combinations of high-low SOC stationary points are traversed, and the maximum ΔSOC (SOC difference) selection principle is used for pairing, and the SOC change is calculated.

[0036] In the above process, this embodiment of the invention effectively captures the battery's resting point information under complex operating conditions through a multi-mode resting point identification mechanism, combined with OCV dynamic calibration and cross-day state continuation technology, and achieves accurate calibration of the resting point SOC. This step significantly improves data utilization and provides accurate SOC change data for subsequent SOH estimation.

[0037] Step 130: Calculate the capacity decay based on the change in SOC using physical rules, obtain the estimated SOH value and assess the confidence level. At the same time, predict the SOH value by training a random forest model and quantify the uncertainty index predicted by the model through out-of-bag estimation.

[0038] In one possible implementation, the battery capacity decay is calculated based on the change in SOC (State of Charge) according to a combination of physical rules and resting point. The current SOH (State of Health) estimate is derived by combining the battery's state data. A confidence assessment system is set up based on confidence factors. The SOH estimate is then comprehensively assessed based on the confidence assessment system, and a scaling factor and mapping function are introduced to obtain the confidence level.

[0039] The physical rules include ampere-hour integrals, the state data includes initial capacity, charge and discharge efficiency, and the confidence factors include measurement error, model uncertainty index, data quality, etc., none of which are specified here.

[0040] In one possible implementation, statistical features closely related to battery aging are extracted from the battery's historical operating data to construct a feature vector. The SOH values ​​that meet the set criteria in the historical operating data are used as training labels. A random forest regression model is trained using a machine learning algorithm, and the model's hyperparameters are optimized through grid search and cross-validation. The feature vector is then input into the trained random forest regression model to obtain the predicted SOH value. The uncertainty index predicted by the model is quantified using the out-of-bag estimation built into the random forest or the prediction variance between trees.

[0041] The statistical characteristics include voltage characteristics, current characteristics, SOC characteristics, temperature characteristics, and energy characteristics, etc., which are not specified here.

[0042] Specifically, based on the change in SOC according to physical rules and the combination of resting points, the battery capacity decay is calculated, and the current SOH estimate is derived. A confidence assessment system is set up according to factors such as measurement error, model uncertainty, and data quality to conduct a comprehensive confidence assessment of the SOH estimate.

[0043] Furthermore, statistical features closely related to battery aging (such as voltage, current, SOC, temperature, and energy characteristics) are extracted from historical battery operating data to construct feature vectors. Using SOH values ​​that meet predetermined criteria from historical data as training labels, a random forest regression model is trained, and its hyperparameters are optimized through grid search and cross-validation. The feature vectors are then input into the trained random forest model to obtain predicted SOH values, and the uncertainty of the model's predictions is quantified using the random forest's built-in out-of-bag estimation or the prediction variance between trees.

[0044] In the above process, this embodiment of the invention combines the accuracy of the physical model and the flexibility of the machine learning model through dual-model prediction using physical rules and machine learning. Physical rule prediction provides high-confidence SOH estimates, while machine learning prediction achieves continuous SOH inference through learning from the full dataset. This dual-model prediction mechanism significantly improves the accuracy and reliability of SOH estimation.

[0045] Step 140: The SOH value is obtained by weighting the estimated SOH value and the predicted SOH value according to the confidence level and uncertainty index through adaptive weighted fusion, or by performing moving average processing on historical data through attenuated weighted filtering.

[0046] In one possible implementation, if only the SOH estimate is valid, then the SOH estimate is used as the SOH value; if only the SOH prediction is valid, then the SOH prediction is used as the SOH value. If both the SOH estimate and the SOH prediction are valid, then the SOH estimate and the SOH prediction are dynamically weighted and fused according to the confidence level and uncertainty index to obtain the SOH value. If both the SOH estimate and the SOH prediction are invalid, then a decaying weighted filter is used to perform an exponentially decaying moving average of the historical data weights in chronological order to obtain the SOH value.

[0047] Specifically, based on the confidence and uncertainty of the physical rules and machine learning predictions, a decision is made as to whether to perform weighted fusion. If both are effective, dynamic weighted fusion is performed based on the confidence and uncertainty to obtain the final SOH value. If neither the physical rules nor the machine learning predictions are effective, a decaying weight filter is used to perform a moving average on the historical data to obtain the SOH value. The decaying weight moving average algorithm ensures the smoothness and stability of the output SOH value.

[0048] In the above process, this embodiment of the invention achieves optimized processing of SOH estimation results through an adaptive weighted fusion and attenuated weighted filtering output mechanism. Adaptive weighted fusion fully utilizes the advantages of physical rules and machine learning predictions, improving the accuracy and reliability of the estimation results; while attenuated weighted filtering output ensures the smoothness and stability of the output results, effectively addressing situations where data is missing or abnormal.

[0049] Through the above process, this embodiment of the invention achieves high-precision online estimation of battery SOH under complex frequency modulation conditions by data acquisition and preprocessing, static point identification and dynamic calibration, dual-model prediction using physical rules and machine learning, adaptive weighted fusion and output. This method combines the accuracy of the physical model with the flexibility of the machine learning model, significantly improving the accuracy and reliability of SOH estimation and providing effective technical support for battery lifecycle management.

[0050] In one application scenario, the online SOH estimation method of the present invention is used to estimate SOH online.

[0051] like Figure 2 As shown, the following steps may be included: Step S1: Data acquisition and preprocessing.

[0052] Specifically, high-precision sensors are used to continuously monitor key operating parameters of the energy storage battery in real time, such as voltage, current, SOC, temperature, and timestamp. Moving average filtering or Kalman filtering techniques are used to denoise the collected operating parameters, interpolation methods are used to fill in missing values ​​in the data, multi-parameter data are aligned according to timestamps, and low-pass filtering is used to smooth the data.

[0053] In the above process, the embodiments of the present invention ensure the quality and reliability of the data on which subsequent analysis depends through multi-dimensional data acquisition and comprehensive preprocessing technology, laying a solid foundation for accurate estimation of SOH.

[0054] Step S2: Static point detection.

[0055] Specifically, the preprocessed operating data is analyzed traversally. When the battery abruptly changes from a high-current operating state (absolute current value greater than a set value, such as 10A) to zero current and remains stationary for a stable threshold (e.g., 1200 seconds), the SOC value at this point is used for judgment. If the SOC value is higher than 90%, it is identified as a high SOC stationary point; if the SOC value is lower than 20%, it is identified as a low SOC stationary point. The identified high and low SOC stationary points are then filtered and verified to form a set of stationary points.

[0056] like Figure 3 As shown, in practical applications, taking three days as an example, on the first day, there is a high SOC static point (SOC=95%) at time point A1 and the state is saved, and a low SOC static point (SOC=18%, ΔSOC=77%) at time point B1; on the second day, there is a high SOC static point (SOC=94%) at time point A2 and the state is saved, but there is no low SOC point, so point B1 from the first day is used; on the third day, there is a time point A3 but no high SOC point, so point A2 from the second day is used, and a low SOC static point (SOC=17%, ΔSOC=77%) at time point B3 is used, demonstrating the application of the cross-day state continuation mechanism in static point detection.

[0057] In the above process, the embodiments of the present invention effectively capture the static state information of the battery under specific operating conditions through a precise static point detection mechanism, providing key data points for subsequent SOC calibration and SOH estimation.

[0058] Step S3: If the static point detection is successful, proceed to the physical rule path.

[0059] Step S31: Multi-mode stationary point identification and OCV dynamic calibration of SOC.

[0060] Specifically, after successfully detecting a stationary point, multi-mode recognition is performed on the stationary point to accurately distinguish different types of stationary states. Next, using a pre-established OCV-SOC relationship curve, each stationary point in the stationary point set is dynamically calibrated. An interpolation algorithm is then used to precisely query the OCV-SOC relationship table to obtain a more accurate SOC value for each stationary point.

[0061] In the above process, the embodiments of the present invention effectively improve the accuracy of the SOC value at the resting point through multi-mode recognition and dynamic calibration technology, providing more accurate data support for subsequent capacity calculation and SOH estimation.

[0062] Step S32: Cross-day state continuation and bidirectional optimal pairing.

[0063] Specifically, a cross-day state continuation mechanism is designed to store relevant information about the static points, such as type, time, and SOC, so as to facilitate state continuation and matching between different days. During the pairing process, a bidirectional optimal pairing strategy is adopted to traverse all possible combinations of high-low SOC static points.

[0064] like Figure 4 As shown, the pairing process begins by acquiring all static points and generating candidate pairing combinations. Then, combination screening is performed. For valid combinations, the change in SOC ΔSOC and the ampere-hour integral value Q are calculated. The combination with the largest ΔSOC is selected after sorting by ΔSOC. The SOHp is calculated and the optimal result is output. If ΔSOC < 2%, the change is considered too small and the combination is discarded. If the SOH is abnormal, the combination with abnormal results is also discarded. When pairing fails, machine learning is used to supplement the results.

[0065] In the above process, the embodiments of the present invention fully utilize the static point data of different time periods through cross-day state continuation and bidirectional optimal pairing technology, thereby improving data utilization and ensuring the rationality and effectiveness of pairing combinations.

[0066] Step S33: Calculate the ampere-hour integral and assess the confidence level, Wp.

[0067] Specifically, based on the SOC change of the bidirectional optimal pairing at rest point, combined with the battery's initial capacity, charge / discharge efficiency, and other state data, the battery's capacity decay is calculated using the ampere-hour integration method, thereby deriving the current SOH estimate. Simultaneously, a confidence assessment system is established based on factors such as measurement error, model uncertainty, and data quality to comprehensively assess the confidence of the SOH estimate, and scaling factors and mapping functions are introduced to obtain specific confidence values.

[0068] In the above process, the embodiments of the present invention not only obtained the estimated SOH value of the battery through ampere-hour integral calculation and confidence assessment technology, but also quantitatively assessed its reliability, providing an important basis for subsequent weighted fusion.

[0069] Step S4: If the static point detection fails, proceed to the data-driven path.

[0070] Step S41: Feature engineering extraction.

[0071] Specifically, statistical features closely related to battery aging are extracted from the battery's historical operating data, including voltage features, current features, SOC features, temperature features, and energy features, to construct a multi-dimensional feature vector.

[0072] In the above process, the embodiments of the present invention fully explore the information related to battery aging in historical data through comprehensive feature engineering extraction, providing rich input features for machine learning models and helping to improve the prediction accuracy of the models.

[0073] Step S42: Random Forest Prediction and Uncertainty Assessment (Wrf).

[0074] Specifically, using historical SOH values ​​that meet set criteria (such as high data quality and accurate SOH values) as training labels, a random forest regression model is trained using machine learning algorithms, and the model's hyperparameters are optimized through grid search and cross-validation. The constructed feature vectors are then input into the trained random forest regression model to obtain the predicted SOH values. Simultaneously, the uncertainty of the model's predictions is quantified using the random forest's built-in out-of-bag estimation or by calculating the prediction variance between trees.

[0075] In the above process, the embodiments of the present invention realize continuous prediction of battery SOH through random forest model prediction and uncertainty assessment technology, and quantitatively evaluate the reliability of the prediction results, providing key information for subsequent weighted fusion.

[0076] Step S5: Adaptive weighted fusion.

[0077] Specifically, the validity of the SOH estimate (Wp) obtained from the physical rule path and the uncertainty index (Wrf) of the SOH prediction obtained from the data-driven path are assessed to determine their effectiveness. If only the SOH estimate is valid, it is directly used as the SOH value; if only the SOH prediction is valid, it is directly used as the SOH value; if both are valid, the SOH estimate and the SOH prediction are dynamically weighted and fused based on the confidence level and uncertainty to obtain the SOH value.

[0078] In the above process, the embodiments of the present invention fully utilize the advantages of physical rule models and machine learning models through adaptive weighted fusion technology, dynamically adjust the weights according to the reliability and effectiveness of both, and improve the accuracy and reliability of SOH estimation results.

[0079] like Figure 5As shown, in the weighted fusion process, the current SOH value has a weight of 1.0, the previous SOH value has a weight of 0.8, the previous two SOH values ​​have a weight of 0.64, the previous three SOH values ​​have a weight of 0.512, the previous four SOH values ​​have a weight of 0.410, and so on. As the number of iterations increases, the weights decrease. The weighted sum of each SOH value is then normalized, and the final output is the filtered SOH value.

[0080] Step S6: Attenuation weighted filtering and SOH result output.

[0081] Specifically, if both physical rules and machine learning predictions are ineffective, a decaying weighted filter is used to apply a moving average to the historical data. The weights of the historical data are exponentially decayed according to time sequence. Each historical SOH value is multiplied by its corresponding weight, summed, and then normalized to obtain the filtered SOH output value. Finally, the final SOH result is output.

[0082] In the above process, the embodiments of the present invention, through attenuation weighted filtering technology, can still output relatively stable and reliable SOH results even in the case of missing or abnormal data, thereby improving the robustness of the entire estimation system.

[0083] Through the above process, this embodiment uses energy storage batteries as the application object to elaborate on the specific implementation process of the online SOH estimation method. High-precision data acquisition and comprehensive preprocessing ensure data quality. Static point detection is used to distinguish battery states. When detection is successful, multi-mode recognition, dynamic calibration, cross-day state continuation, bidirectional optimal pairing, ampere-hour integral calculation, and confidence assessment are performed through physical rule paths. When detection fails, feature engineering extraction, random forest prediction, and uncertainty assessment are performed through data-driven paths. Then, adaptive weighted fusion combines the results from both methods, and attenuated weighted filtering is used to process historical data when necessary, ultimately outputting a reliable SOH result. This method effectively solves the problems of low data utilization and poor reliability of traditional methods under complex operating conditions, improves the accuracy and stability of SOH estimation, and provides strong support for the health management of energy storage batteries.

[0084] In another application scenario, in large-scale energy storage power stations, the online SOH estimation method proposed in this embodiment of the invention is used to construct an online SOH estimation system for batteries.

[0085] The system includes a data acquisition and preprocessing module, a multi-mode stationary point detection module, an OCV dynamic calibration module, a cross-day status management module, a two-way pairing optimization module, an ampere-hour integral calculation module, a feature engineering extraction module, a random forest prediction module, a confidence assessment module, an adaptive fusion module, and an attenuation filter output module.

[0086] Specifically, the data acquisition and preprocessing module utilizes high-precision sensors to collect real-time data on the battery pack's voltage, current, SOC, temperature, and timestamps. Data imputation involves interpolation to ensure data continuity for missing data points. Deduplication removes duplicate data records to avoid computational redundancy. Interpolation performs linear interpolation on data with discontinuous timestamps, standardizing the sampling interval to 1 second. Timestamp alignment ensures strict alignment of timestamps for voltage, current, and SOC data. Filtering uses a low-pass filter to smooth data fluctuations and reduce noise interference. This provides high-quality, standardized input data for subsequent modules.

[0087] Multi-mode resting point detection module: Traverses the pre-processed data stream to monitor changes in the battery's operating state in real time. High SOC resting point: A sudden change from high current operating state (absolute current > 10A) to zero current, with a resting time reaching a preset threshold (e.g., 1200 seconds), and the SOC value is above 90%. Low SOC resting point: Similarly, a sudden change from high current operating state to zero current, with a resting time reaching a preset threshold, and the SOC value is below 20%. Accurately identifying the battery's resting state under specific operating conditions provides key data points for subsequent SOC calibration and SOH estimation.

[0088] Specifically, a dual resting point detection standard is established to capture resting opportunities in high SOC and low SOC regions respectively. High or low SOC resting points may occur after battery charging and discharging. However, due to polarization, the battery voltage is not equal to the open circuit voltage when it first enters resting. Therefore, the resting period needs to meet a certain duration to be determined as a high or low SOC resting point. Due to current sampling errors, ampere-hour integration errors, etc., the SOC corresponding to high and low SOC resting points is not the actual SOC of the battery itself. It needs to be corrected according to the SOC-OCV relationship to obtain the true SOC.

[0089] The high and low SOC resting point identification includes the following: High SOC resting point detection conditions: a sudden change from high current operating state to zero current (the absolute value of the current at the previous moment is >10A, and the current at the next moment is 0), that is, regardless of whether the battery changes from charging state or discharging state to resting state, it can be continuously rested until a stable threshold (such as 1200 seconds) is reached and the SOC value is higher than 90%. At this time, the resting point of this frame 1200s after this point is defined as the high SOC resting point.

[0090] Low SOC settling point detection conditions: The current changes abruptly from a high current operating state to zero current (the absolute value of the current at the previous moment is >10A, and the current at the next moment is 0), and the settling point is maintained until a stable threshold is reached (e.g., 1200 seconds) and the SOC value is below 20%. Since there may be multiple settling points within the same day, the settling time may not meet the requirements; there may be long settling points, but the SOC point does not meet the requirements.

[0091] The OCV dynamic calibration module performs dynamic calibration at the detected resting point using a pre-established OCV-SOC relationship curve. It queries the OCV-SOC relationship table by interpolating the value in the discrete OCV-SOC relationship table based on the resting open-circuit voltage (OCV) at the resting point. Finally, it corrects the SOC value at the resting point by combining the query results with experimentally obtained lithium battery characteristics, resulting in a precise SOC baseline value. This improves the accuracy of the resting point SOC value, providing more accurate data support for subsequent capacity calculations and SOH estimations.

[0092] Among them, the pre-established OCV-SOC relationship curve is obtained through experiments to obtain the SOC of lithium batteries under multiple OCVs, i.e. Corresponding to ,Right now Corresponding to Then determine that the obtained open-circuit voltage V is between and In the middle, the battery's SOC at this point is between... and Between them, the following are the pre-established key relationships: .

[0093] Therefore, it is only necessary to set a reasonable stabilization time window after resting to ensure that the battery voltage stabilizes after sufficient resting, and then use an interpolation algorithm to perform accurate lookup in the discrete OCV-SOC relationship table.

[0094] Cross-Day Status Management Module: A status memory system with timeout protection (e.g., 7 days) is designed to store information such as the type, time, and SOC of the resting point. Cross-Day Status Continuation Mechanism: Information Storage: Valid resting point information for each day is saved to the database, including resting point type, timestamp, and SOC value. Status Continuation: If only one of a high SOC or low SOC resting point is detected on a given day, this information is saved to the next day and paired with the resting point for calculation. Timeout Protection: A 7-day timeout protection period is set; resting point information that is not paired after this period will be cleared. This maximizes the utilization of lithium battery charge and discharge information, captures more aging information, and ensures the continuity of SOH estimation.

[0095] Bidirectional pairing optimization module: It iterates through all possible combinations of high- and low SOC static points and pairs them using the maximum ΔSOC selection principle. Pairing process: Combination traversal: Pairwise pairings are performed for each day's high-SOC static points and low-SOC static points. ΔSOC calculation: The SOC difference ΔSOC between each combination is calculated. Optimal pairing: The effective combination with the largest ΔSOC is selected for capacity calculation, while a reasonable SOH range verification mechanism (0.8-1.1) is set to exclude abnormal calculation results. This ensures the rationality and effectiveness of the paired combinations and improves the accuracy of SOH estimation.

[0096] Ampere-hour integration calculation module: Based on the SOC change of the optimal pairing and resting point combination, combined with the battery's initial capacity, charge / discharge efficiency, and other state data, it calculates the battery's capacity decay using the ampere-hour integration method. Capacity calculation: Calculates the battery's actual capacity based on the SOC change and current integral value between paired resting points. SOH estimation: Calculates the current estimated SOH value (SOH = Actual Capacity / Nominal Capacity) based on the battery's initial nominal capacity. This reveals the battery's capacity decay and allows for the derivation of the current SOH value.

[0097] The calculation of the SOC change between paired static points is specifically implemented as follows: .

[0098] The ampere-hour integral value between two paired resting point combinations does not need to consider intermediate charge / discharge conditions or continuous resting conditions. The specific implementation is as follows: .

[0099] SOH calculated by the Coulomb count variant: .

[0100] Confidence It is determined by two factors: .

[0101] in, Scaling factor The function is used to... Map to the interval (0,1).

[0102] Feature Engineering Extraction Module: Extracts statistical features closely related to battery aging from historical battery operating data. Feature extraction includes: Voltage features: mean, variance, skewness, kurtosis, extreme values, quantiles. Current features: mean, standard deviation, charge / discharge statistics, switching frequency, time ratio. State of Charge (SOC) features: range, variation, volatility, initial value. Temperature features: mean, extreme values, fluctuation range. Energy features: throughput, charge / discharge energy, efficiency indicators. This module fully leverages historical data to extract battery aging-related information, providing rich input features for machine learning models.

[0103] Random Forest Prediction Module: This module uses SOH values ​​that meet predefined criteria from historical data as training labels to train a random forest regression model. Model Training: The high-confidence SOH values ​​output by the aforementioned physical rules are used as supervision signals, combined with extracted feature vectors, to train the random forest model. Model Update: The model is periodically updated online using new high-confidence samples to adapt to changes in battery aging status. SOH Prediction: The feature vector for the current day is input into the trained model to obtain the predicted SOH value. This enables continuous prediction of battery SOH, improving the timeliness and accuracy of the estimation.

[0104] Among them, the model outputs a high-confidence SOH value. It is used as a supervisory signal for training. Its mathematical form is: .

[0105] Among them, T i This represents a single decision tree. Through grid search and cross-validation, the key hyperparameters of the model were optimized as follows: the number of trees is 100, the maximum tree depth is 10, the minimum number of samples per leaf node is 5, and the minimum number of samples required for node splitting is 10. These settings aim to balance the model's fitting ability and generalization performance, effectively suppressing overfitting.

[0106] The confidence assessment module evaluates the confidence levels of the SOH estimates obtained from the physical rule-based path and the SOH predictions obtained from the data-driven path. Physical rule confidence (Wp): A confidence assessment system is established based on factors such as measurement error, model uncertainty, and data quality. Machine learning uncertainty index (Wrf): The uncertainty of the prediction is quantified using out-of-bag estimation built into random forests or by calculating the prediction variance between trees. This quantitative evaluation of the reliability of the SOH estimates and predictions provides a basis for subsequent weighted fusion.

[0107] The model incorporates a confidence assessment based on prediction consistency. This is achieved by calculating the standard deviation of all decision tree predictions. To quantify the uncertainty of this forecast : .

[0108] This is then converted into confidence weights: .

[0109] This mechanism enables the model to self-assess the reliability of its output, providing a crucial basis for subsequent weighted fusion with physical rules.

[0110] Furthermore, the out-of-bag estimation built into random forests can be used to calculate the prediction variance between trees: .

[0111] in, This is an uncertainty index.

[0112] The adaptive fusion module: It determines the validity of the SOH estimate (Wp) obtained from the physical rule path and the uncertainty (Wrf) of the SOH prediction obtained from the data-driven path. Validity judgment: If only the SOH estimate is valid, it is directly adopted; if only the SOH prediction is valid, it is also directly adopted; if both are valid, dynamic weighted fusion is performed based on confidence and uncertainty. Weighted fusion: The fusion ratio of the SOH estimate and prediction is allocated according to the confidence weight to obtain the final fused SOH value. This fully leverages the advantages of the physical rule model and the machine learning model to improve the accuracy and reliability of the SOH estimation results.

[0113] Specifically, determine whether the physical rules for that day have outputted valid values. If yes, proceed to weighted fusion; otherwise, adopt directly. As the final output. When and When both exist simultaneously, a weighted average is taken from the two values: .

[0114] Attenuation Filter Output Module: If both physical rules and machine learning predictions are ineffective, an attenuation weighted filter is used to perform a moving average processing on the historical data. Weight Allocation: The weights of historical data are exponentially decayed according to time order, with the most recent data assigned the highest weight. Each historical SOH value is multiplied by its corresponding weight, summed, and then normalized to obtain the filtered SOH output value. Even in cases of missing or anomaly-prone data, this module can still output relatively stable and reliable SOH results, improving the robustness of the entire estimation system.

[0115] Specifically, the latest calculation result is assigned the highest weight; then, the weights are reduced exponentially based on historical data. Furthermore, the length of historical data retention is limited to ensure the algorithm's response speed. The final output is the SOH value after smoothing and filtering, which is also the final output value of the entire system. The final SOH value after filtering is obtained through weighted summation and normalization.

[0116] Through the above process, this embodiment of the invention achieves real-time and accurate estimation of battery SOH by coordinating modules such as high-precision data acquisition and comprehensive preprocessing, multi-mode static point detection and calibration, cross-day state management, bidirectional pairing optimization, ampere-hour integral calculation, feature engineering extraction, random forest prediction, confidence assessment, adaptive weighted fusion, and attenuation weight filtering. This method effectively solves the problems of low data utilization and poor reliability of traditional methods under complex operating conditions, providing strong support for the health management of energy storage batteries.

[0117] The following are embodiments of the apparatus of the present invention, which can be used to execute the online SOH estimation method involved in the present invention. For details not disclosed in the apparatus embodiments of the present invention, please refer to the method embodiments of the online SOH estimation method involved in the present invention.

[0118] Please see Figure 6 This invention provides an online SOH estimation device 800.

[0119] The online SOH estimation device 800 includes, but is not limited to: a data acquisition and preprocessing module 810, a static point calibration and matching module 830, a rule model dual prediction module 850, and an online SOH estimation module 870.

[0120] The data acquisition and preprocessing module 810 is used to acquire and preprocess the battery's operating data in real time. It uses a multi-mode resting point identification mechanism to intelligently detect the SOC resting point based on the operating data to obtain a set of resting points. The operating data includes voltage, current, SOC, temperature, and timestamp. The preprocessing includes noise reduction, interpolation, alignment, and filtering.

[0121] The stationary point calibration matching module 830 is used to dynamically calibrate the stationary point set using a pre-established OCV-SOC relationship curve, and to match and filter stationary points in conjunction with the set cross-day state continuation mechanism to obtain paired stationary point combinations and SOC changes; the stationary point information includes type, time, and SOC.

[0122] The rule-based model dual prediction module 850 is used to calculate capacity decay based on the change in SOC using physical rules, obtain an estimated SOH value and assess confidence, and simultaneously predict the SOH value by training a random forest model and quantify the uncertainty index predicted by the model through out-of-bag estimation; the physical rules include ampere-hour integrals.

[0123] The online SOH estimation module 870 is used to obtain the SOH value by weighting the estimated SOH value and the predicted SOH value according to the confidence level and uncertainty index through adaptive weighted fusion, or by performing moving average processing on historical data through attenuated weight filtering.

[0124] It should be noted that the online SOH estimation provided in the above embodiments is only an example of the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed. That is, the internal structure of the online SOH estimation device will be divided into different functional modules to complete all or part of the functions described above.

[0125] Furthermore, the online SOH estimation device and the online SOH estimation method provided in the above embodiments belong to the same concept, and the specific way in which each module performs its operation has been described in detail in the method embodiments, and will not be repeated here.

[0126] Figure 7 A schematic diagram of the structure of an electronic device according to an exemplary embodiment is shown.

[0127] It should be noted that this electronic device is merely an example adapted to the present invention and should not be construed as providing any limitation on the scope of use of the present invention. Furthermore, this electronic device should not be interpreted as requiring or depending on having... Figure 7 One or more components of the exemplary electronic device 2000 shown.

[0128] The hardware structure of electronic devices 2000 can vary significantly due to differences in configuration or performance, such as... Figure 7 As shown, the electronic device 2000 includes: a power supply 210, an interface 230, at least one memory 250, and at least one central processing unit (CPU) 270.

[0129] Specifically, power supply 210 is used to provide operating voltage for various hardware devices on electronic device 2000.

[0130] Interface 230 includes at least one wired or wireless network interface 231 for interacting with external devices. Of course, in other examples adapted to this invention, interface 230 may further include at least one serial-to-parallel conversion interface 233, at least one input / output interface 235, and at least one USB interface 237, etc. Figure 7 As shown, this does not constitute a specific limitation.

[0131] The memory 250 serves as a carrier for resource storage and can be a read-only memory, random access memory, disk, or optical disk, etc. The resources stored on it include the operating system 251, application programs 253, and data 255, etc., and the storage method can be temporary storage or permanent storage.

[0132] The operating system 251 is used to manage and control the various hardware devices and application programs 253 on the electronic device 2000, so as to enable the central processing unit 270 to perform calculations and processing on the massive data 255 in the memory 250. It can be Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0133] Application 253 is a computer-readable instruction based on operating system 251 that performs at least one specific task, and may include at least one module ( Figure 7 (Not shown), each module may contain computer-readable instructions for electronic device 2000. For example, the SOH online estimation device can be considered as application 253 deployed on electronic device 2000.

[0134] Data 255 may be signal information, etc., and is stored in memory 250.

[0135] The central processing unit 270 may include one or more processors and is configured to communicate with the memory 250 via at least one communication bus to read computer-readable instructions stored in the memory 250, thereby performing operations and processing on the massive amounts of data 255 stored in the memory 250. For example, the online SOH estimation method can be completed by the central processing unit 270 reading a series of computer-readable instructions stored in the memory 250.

[0136] Furthermore, the present invention can also be implemented through hardware circuits or a combination of hardware circuits and software. Therefore, the implementation of the present invention is not limited to any specific hardware circuit, software, or combination thereof.

[0137] Please see Figure 8 This invention provides an electronic device 4000, which may include: a desktop computer, a laptop computer, a server, etc., with sensor recognition capabilities.

[0138] exist Figure 8 In this context, the electronic device 4000 includes at least one processor 4001 and at least one memory 4003.

[0139] The data interaction between the processor 4001 and the memory 4003 can be achieved through at least one communication bus 4002. This communication bus 4002 may include a path for transmitting data between the processor 4001 and the memory 4003. The communication bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The communication bus 4002 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 8 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0140] Optionally, the electronic device 4000 may further include a transceiver 4004, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 4004 is not limited to one type, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of the present invention.

[0141] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. Processor 4001 may also be a combination that implements computing functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0142] The memory 4003 may be a ROM (Read Only Memory) or other type of static storage device capable of storing static information and instructions, RAM (Random Access Memory) or other type of dynamic storage device capable of storing information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program instructions or code in the form of instructions or data structures and accessible by the electronic device 4000, but not limited thereto.

[0143] The memory 4003 stores computer-readable instructions, and the processor 4001 can read the computer-readable instructions stored in the memory 4003 through the communication bus 4002.

[0144] The computer-readable instructions are executed by one or more processors 4001 to implement the online SOH estimation method in the above embodiments.

[0145] Furthermore, this embodiment of the invention provides a storage medium storing computer-readable instructions, which are executed by one or more processors to implement the online SOH estimation method described above.

[0146] This invention provides a computer program product including computer-readable instructions stored in a storage medium. One or more processors of an electronic device read the computer-readable instructions from the storage medium, load and execute the computer-readable instructions, thereby enabling the electronic device to implement the online SOH estimation method as described above.

[0147] Compared with related technologies, the beneficial effects of the present invention are: 1. This invention can significantly improve the success rate of SOH estimation; through the adaptive weighted fusion mechanism of physical rule model and random forest model, the success rate of traditional methods is increased from <50% to >90%, effectively solving the estimation failure problem caused by fragmented data under frequency modulation conditions.

[0148] 2. This invention has a higher data utilization rate; through multi-mode static point identification and cross-day state continuation mechanism, it makes full use of incomplete charging and discharging segments in frequency modulation scenarios, improving data utilization rate by more than 40%, and breaking through the dependence of traditional methods on complete cycles.

[0149] 3. This invention can ensure physical interpretability; by using the high confidence SOH value provided by the physical rule model as the gold standard node, combined with the continuous prediction results of the machine learning model, the fused output always has a clear physical meaning, thus solving the reliability problem of the "black box" model of the data-driven method.

[0150] 4. This invention has strong robustness; through a dual-reset reliability assessment system (physical rule confidence Wp and machine learning uncertainty Wrf) and decaying weighted moving average filtering output, it can still provide a reasonable SOH value through historical data trend prediction when the physical rule fails, ensuring the continuous and stable operation of the system.

[0151] 5. This invention can adapt to the entire battery life cycle; through a regularly updated random forest model and a dynamic weighted fusion strategy, it automatically adapts to changes in different aging stages and operating states of the battery, avoiding the accumulation of estimation errors caused by model parameter drift in traditional methods.

[0152] 6. This invention has practical engineering value; through online monitoring and real-time filtering output mechanism, it meets the timeliness requirements of SOH estimation for energy storage power stations, while reducing reliance on manual calibration and lowering maintenance costs. It has been verified in actual frequency regulation energy storage power stations.

[0153] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0154] The above description is only a partial embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for online estimation of SO₂, characterized in that, The method includes: The system collects and preprocesses battery operating data in real time. A multi-mode resting point identification mechanism is used to intelligently detect SOC resting points based on the operating data, resulting in a set of resting points. The operating data includes voltage, current, SOC, temperature, and timestamp. The preprocessing includes noise reduction, interpolation, alignment, and filtering. The set of stationary points is dynamically calibrated using a pre-established OCV-SOC relationship curve. The stationary points are then matched and filtered using a set cross-day state continuation mechanism to obtain paired stationary point combinations and SOC changes. The stationary point information includes type, time, and SOC. The capacity decay is calculated based on the change in SOC using physical rules to obtain an estimated SOH value and assess the confidence level. Simultaneously, a random forest model is trained to predict the SOH value, and the uncertainty index predicted by the model is quantified by out-of-bag estimation. The physical rules include ampere-hour integrals. The SOH value can be obtained by weighting the estimated SOH value and the predicted SOH value according to the confidence level and the uncertainty index using adaptive weighted fusion, or by performing a moving average processing on historical data through attenuated weight filtering.

2. The online SOH estimation method as described in claim 1, characterized in that, The real-time acquisition and preprocessing of battery operating data includes: High-precision sensors are used to continuously monitor various key operating parameters of the battery in real time, and filtering techniques are employed to denoise these operating parameters; the filtering techniques include moving average filtering or Kalman filtering. The missing values ​​in the running parameters are filled using interpolation methods, the data is aligned according to the timestamps of the running parameters, and the running parameters are smoothed using low-pass filtering.

3. The online SOH estimation method as described in claim 1, characterized in that, The multi-mode stationary point identification mechanism intelligently detects SOC stationary points based on the operational data to obtain a set of stationary points, including: The operating data is traversed. When the battery suddenly changes from a high current operating state to zero current and remains stationary until a stable threshold is reached, if the SOC value is higher than a set percentage at this time, it is identified as a high SOC stationary point. The stable threshold includes the duration of the state. If the SOC value is lower than the set percentage at this time, it is identified as a low SOC resting point, and the high SOC resting points and low SOC resting points are screened and verified to obtain a set of resting points; the high current working state refers to the absolute value of the current being greater than the set value.

4. The online SOH estimation method as described in claim 1, characterized in that, The process involves dynamically calibrating the set of stationary points using a pre-established OCV-SOC relationship curve, and matching and filtering the stationary points using a set cross-day state continuation mechanism to obtain paired stationary point combinations and SOC changes, including: Using the OCV-SOC relationship curve established in advance through experiments, each stationary point in the set of stationary points is dynamically calibrated. The SOC value of each stationary point is obtained by querying the OCV-SOC relationship table through an interpolation algorithm. A cross-day state continuation mechanism is designed to store the information of the stationary points. A two-way pairing strategy is used to match the high SOC stationary points and the low SOC stationary points to obtain multiple pairs of stationary point combinations. The SOC difference between each of the stationary point combinations is calculated. For the combinations whose SOC difference meets the set conditions, the SOC change is calculated.

5. The online SOH estimation method as described in claim 1, characterized in that, The process of calculating capacity decay based on the change in SOC using physical rules, obtaining an estimated SOH value, and assessing confidence level includes: The capacity decay of the battery is calculated based on physical rules and the SOC change of the resting point combination, and the current SOH estimate is derived by combining the battery's state data; the state data includes initial capacity and charge / discharge efficiency. A confidence assessment system is set up based on confidence factors. The SOH estimate is then comprehensively assessed based on the confidence assessment system, and a scaling factor and mapping function are introduced to obtain the confidence level. The confidence factors include measurement error, model uncertainty index, and data quality.

6. The online SOH estimation method as described in claim 1, characterized in that, The process of predicting the SOH prediction value by training a random forest model and quantifying the uncertainty index predicted by the model through out-of-bag estimation includes: Statistical features closely related to battery aging are extracted from the historical operating data of the battery to construct a feature vector; the statistical features include voltage features, current features, SOC features, temperature features, and energy features. Using the SOH values ​​that meet the set criteria from the historical running data as training labels, a random forest regression model is trained using a machine learning algorithm, and the model hyperparameters are optimized through grid search and cross-validation. The feature vector is input into the trained random forest regression model to obtain the SOH prediction value, and the uncertainty index is predicted using the random forest's built-in out-of-bag estimation or the prediction variance quantification model between trees.

7. The online SOH estimation method as described in claim 1, characterized in that, The method of obtaining the SOH value by weighting and fusing the estimated and predicted SOH values ​​based on the confidence level and the uncertainty index using adaptive weighted fusion, or by performing a moving average processing on historical data using attenuated weighted filtering, includes: If only the SOH estimate is valid, then the SOH estimate is used as the SOH value; if only the SOH prediction is valid, then the SOH prediction is used as the SOH value. If both the estimated SOH value and the predicted SOH value are valid, then the estimated SOH value and the predicted SOH value are dynamically weighted and fused according to the confidence level and the uncertainty index to obtain the SOH value; If both the estimated and predicted SOH values ​​are invalid, then the SOH value is obtained by using a decaying weighted filter to perform an exponentially decaying moving average of the weights of the historical data in chronological order.

8. An online SOH estimation device, characterized in that, The device includes: The data acquisition and preprocessing module is used to acquire and preprocess the battery's operating data in real time. It uses a multi-mode resting point identification mechanism to intelligently detect SOC resting points based on the operating data, obtaining a set of resting points. The operating data includes voltage, current, SOC, temperature, and timestamp. The preprocessing includes noise reduction, interpolation, alignment, and filtering. The stationary point calibration and matching module is used to dynamically calibrate the set of stationary points using a pre-established OCV-SOC relationship curve, and to match and filter the stationary points in conjunction with a set cross-day state continuation mechanism to obtain paired stationary point combinations and SOC changes; the stationary point information includes type, time, and SOC; The rule-based dual-prediction module is used to calculate the capacity decay based on the change in SOC using physical rules, obtain the estimated SOH value and evaluate the confidence level, and simultaneously predict the SOH value by training a random forest model and quantify the uncertainty index predicted by the model through out-of-bag estimation; the physical rules include ampere-hour integrals; The online SOH estimation module is used to obtain the SOH value by weighting the estimated SOH value and the predicted SOH value according to the confidence level and the uncertainty index through adaptive weighted fusion, or by performing a moving average processing on historical data through attenuated weighted filtering to obtain the SOH value.

9. An electronic device, characterized in that, include: At least one processor and at least one memory, wherein, The memory stores computer-readable instructions; The computer-readable instructions are executed by one or more of the processors, causing the electronic device to implement the online SOH estimation method as described in any one of claims 1 to 7.

10. A storage medium having computer-readable instructions stored thereon, characterized in that, The computer-readable instructions are executed by one or more processors to implement the online SOH estimation method as described in any one of claims 1 to 7.

Images