An intelligent lithium iron phosphate battery capacity evaluation method, system, medium and product

Abstract

The application discloses a kind of intelligent lithium iron phosphate battery capacity evaluation method, system, medium and product, it is related to data processing technical field.Method includes: the theoretical battery capacity of target battery is calculated;When theoretical battery capacity is in the preset dangerous capacity range, calculate historical feature vector;Vector difference operation is carried out to current feature vector and historical feature vector, and aging vector is obtained;When detecting that target battery exists the fault belonging to preset fault type, and still in maintaining charge-discharge function, aging vector is decomposed into actual aging vector and fault vector;Update preset calibration function, and obtain new calibration function;The latest aging data in battery aging track is substituted into new calibration function, and the health capacity value is obtained;Based on fault vector, correct health capacity value by preset fault correction algorithm, and obtain actual capacity value.Implement the above technical scheme, improve the accuracy of capacity evaluation reflecting the real health status of battery.

Description

Technical Field

[0001] This application relates to the field of data processing, and in particular to a method, system, medium, and product for evaluating the capacity of intelligent lithium iron phosphate batteries. Background Technology

[0002] Lithium iron phosphate (LFP) batteries, as core components for energy storage and power output, have been widely used in key areas such as electric vehicles and distributed energy storage power stations. Battery capacity is the most critical indicator of battery state of health (SOH). Accurately assessing the capacity of LFP batteries directly affects the operational reliability, range, and decision-making regarding retirement and replacement of equipment, thus becoming a core function of the battery management system (BMS).

[0003] Currently, mainstream battery capacity assessments are typically based on calculations using real-time collected data such as voltage, current, and temperature. For example, the ampere-hour integration method is used to statistically determine the charge and discharge capacity, or an equivalent circuit model (ECM) is combined with algorithms such as Kalman filtering to estimate the battery's current state of capacity in real time, thus obtaining a theoretical battery capacity value. This calculation method based on real-time operating data usually provides a capacity assessment result with some reference value when the battery is in a healthy state or under stable operating conditions.

[0004] However, when a battery has been used for a long time and its calculated theoretical capacity enters a lower warning range (i.e., a dangerous capacity range), the accuracy of the traditional assessment methods mentioned above will decrease significantly. The main reason is that at this stage, the electrochemical characteristics inside the battery become extremely sensitive and complex: on the one hand, instantaneous fluctuations caused by real-time operating conditions (such as polarization voltage caused by current surges) are more pronounced in aged batteries, easily masking the battery's true static health characteristics; on the other hand, battery aging and degradation often occur simultaneously with potential minor faults (such as micro-short circuits), and the two exhibit a high degree of coupling in external data characteristics. Existing technologies struggle to effectively isolate operating condition interference and distinguish between normal aging and abnormal faults within this specific range, resulting in calculated capacity that often fails to reflect the battery's true healthy energy storage capacity, easily leading to misjudgments of the battery's remaining lifespan or missed detections of potential faults. Summary of the Invention

[0005] This application provides a method, system, medium, and product for evaluating the capacity of intelligent lithium iron phosphate batteries, which can improve the accuracy of capacity evaluation that reflects the true health status of the battery.

[0006] Firstly, this application provides a method for evaluating the capacity of an intelligent lithium iron phosphate battery. The method includes: calculating the theoretical battery capacity of the target battery based on real-time operational data collected in real time; when the theoretical battery capacity is within a preset dangerous capacity range, calculating a historical feature vector based on the target feature vector corresponding to each target's historical operational data and the local curvature of the end region in the battery aging trajectory of the target battery, wherein the target's historical operational data is operational data in a historical operational database that matches the current operating condition of the target battery; performing a vector difference operation between the current feature vector and the historical feature vector to obtain an aging vector, wherein the current feature vector is a feature vector characterizing the current state of the target battery; and when a preset fault is detected in the target battery... When a fault of a certain type occurs while the battery is still maintaining its charging and discharging function, the aging vector is decomposed into an actual aging vector and a fault vector. The actual aging vector is then added to the end of the battery's aging trajectory. Based on the anchor point operation data selected from the historical operation database and the battery's aging trajectory, the preset calibration function is updated to obtain a new calibration function. The anchor point operation data refers to the operation data in the historical operation data where the confidence level of the capacity calculation result is higher than a preset threshold. The latest aging data in the battery's aging trajectory is substituted into the new calibration function to calculate the healthy capacity value of the target battery under fault-free conditions. Based on the fault vector, the healthy capacity value is corrected using a preset fault correction algorithm corresponding to the preset fault type to obtain the actual capacity value of the target battery.

[0007] The above technical solution separates the "state features" generated by instantaneous operating condition fluctuations from the current feature vector exhibited by the battery, obtaining a feature vector reflecting the battery's current health state. Then, based on the separated feature vector, the battery capacity reflecting the battery's true health state is accurately calculated. This method addresses the core problem in battery capacity assessment where the long-term aging degradation signal, sudden fault degradation signal, and real-time operating condition fluctuation signal are mutually coupled and masked, making it difficult for the assessment model to accurately identify the battery's true physical health state. Through differential operations and operating condition matching, the interference of instantaneous operating condition fluctuations on health state assessment is eliminated. Furthermore, through vector decomposition, the degradation effects from different physical sources (long-term aging and sudden faults) are quantitatively separated, ensuring the depth of diagnosis and the accuracy of attribution, and avoiding the risk of multiple assessment inaccuracies caused by operating condition fluctuations, nonlinear aging, and concurrent faults. Meanwhile, the dynamic self-calibration mechanism based on high-confidence anchor data ensures that the core evaluation model can continuously adapt to the state evolution of the battery throughout its entire life cycle, avoiding the problem of model failure during long aging cycles, improving the ability of battery capacity evaluation results to reflect the true physical state of the battery, and thus improving the accuracy of capacity evaluation that reflects the true health state of the battery.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, the aging vector is decomposed into an actual aging vector and a fault vector, specifically including: comparing the aging vector with the health baseline of the target battery at the current stage of its life to calculate an abnormal deviation vector, the health baseline including one or more benchmark aging vectors; logically matching the abnormal deviation vector with a preset fault feature rule base to obtain the target fault mode of the target battery; projecting the abnormal deviation vector onto a preset feature fault signature corresponding to the target fault mode to obtain a fault vector, the preset feature fault signature being a standardized benchmark vector, the direction of which defines the degradation path that causes the target fault mode to cause changes in the battery's health state; and separating the fault vector from the aging vector to obtain the actual aging vector.

[0009] By employing the above technical solution, the abnormal deviation vector, which separates the effects of aging and reflects the characteristics of faults, is matched with a preset fault feature rule base to locate specific target fault modes from complex abnormal signals, thereby clarifying the root cause attributes of the anomaly. Then, the standardized feature signature corresponding to the fault mode is used for vector projection to extract the impact components belonging to the fault in the mixed abnormal signal and quantify them into fault vectors, so that the fault impact can be given a measurable physical expression. The actual aging vector obtained through the separation operation realizes the accurate separation of aging and fault impacts, providing a reliable input for subsequent capacity assessment based on the actual aging state, thereby improving the accuracy of capacity assessment that reflects the true health state of the battery.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, separating the fault vector from the aging vector to obtain the actual aging vector specifically includes: retrieving the corresponding coupling effect quantification equation from a preset fault and aging coupling effect knowledge base based on the target fault mode and the magnitude of the fault vector; substituting the magnitude of the fault vector into the coupling effect quantification equation to calculate the acceleration factor for each affected aging feature dimension and constructing an acceleration transformation matrix; obtaining the theoretical aging vector by solving a preset inverse optimization problem based on the aging vector and the acceleration transformation matrix, wherein the predicted vector after transformation of the theoretical aging vector by the acceleration transformation matrix has the smallest error with the aging vector; multiplying the theoretical aging vector by the acceleration transformation matrix to obtain the coupled aging increment vector; and summing the theoretical aging vector and the coupled aging increment vector to obtain the actual aging vector.

[0011] By employing the aforementioned technical solution, the physical coupling effect of faults accelerating the aging process is quantified into an "acceleration transformation matrix," and an inverse optimization problem is constructed, enabling in-depth tracing and reconstruction of the causes of hybrid degradation. Since the observed "aging vector" is itself the final manifestation after coupling, rather than a simple superposition of components, solving the inverse problem allows for the reverse derivation of the "theoretical aging vector" under fault-free acceleration from this hybrid result, thus locking in the battery's most fundamental and inherent degradation rate. Based on this, the theoretical aging is summed with the "coupled aging increment vector" explicitly calculated from the coupling effect, resulting in the "actual aging vector," which physically encompasses both the battery's basic aging and the permanent incremental aging caused by fault acceleration. This avoids underestimating the battery's true aging degree due to ignoring coupling effects, improves the accuracy of battery health status assessment, and consequently enhances the accuracy of capacity assessment reflecting the battery's true health status.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, after performing vector difference operation between the current feature vector and the historical feature vector to obtain the aging vector, the method further includes: when a fault not belonging to a preset fault type is detected in the target battery, and the battery is still maintaining charging and discharging functions, performing subspace projection decomposition on the aging vector based on the aging subspace composed of historical aging vectors under one or more historical healthy states, to obtain coplanar aging components that can be explained by historical experience and orthogonal anomaly components representing unknown faults; calculating the feature sensitivity vector of each feature dimension relative to the change in unit charge based on the historical feature vector and the corresponding historical state of charge data; performing a dimensional division operation between the absolute value of the orthogonal anomaly component and the absolute value of the feature sensitivity vector to obtain a dimension-constrained capacity vector; selecting the minimum value among all dimension components from the dimension-constrained capacity vector as the risk-constrained capacity; and selecting the smaller capacity between the risk-constrained capacity and the baseline healthy capacity corresponding to the coplanar aging component as the actual capacity value of the target battery.

[0013] By employing the aforementioned technical solution, a subspace based on historical healthy aging experience is constructed, and the current aging vector is orthogonally decomposed. This process isolates "orthogonal anomalous components" that cannot be explained by historical experience. A "feature sensitivity vector," directly related to physical charge, is introduced as a benchmark. The maximum capacity loss that this anomalous component may cause in each feature dimension is constrained and calculated. The minimum value among all dimensional constraints is selected as the "risk-constrained capacity." This transforms an abstract, unknown anomaly into a concrete capacity boundary value with safety as the highest criterion. By taking the smaller value between this risk-constrained capacity and the baseline healthy capacity calculated based on normal aging, the actual output capacity value ensures a highly reliable assessment result, whether in scenarios dominated by normal aging or where unknown faults are the primary limiting factor. This avoids the risk of overestimating battery usability due to unknown faults.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, the aging vector is decomposed by subspace projection based on an aging subspace composed of historical aging vectors under one or more historical healthy states to obtain coplanar aging components that can be explained by historical experience and orthogonal abnormal components representing unknown faults. Specifically, this includes: querying an initial reference subspace of the theoretical healthy aging behavior of the target battery at the current life stage from a preset historical aging subspace library, which stores aging subspaces of standard aging modes for each stage of the target battery's entire life cycle; sequentially projecting the aging vectors corresponding to each trajectory in the battery's aging trajectory onto the initial reference subspace to obtain a set of trajectory orthogonal components; identifying abnormal aging direction vectors in the set of trajectory orthogonal components; using the abnormal aging direction vectors as correction operators to correct the initial reference subspace to obtain a corrected aging subspace; and performing projection decomposition on the aging vector based on the corrected aging subspace to obtain coplanar aging components that can be explained by historical experience and orthogonal abnormal components representing unknown faults.

[0015] By employing the aforementioned technical solution, an aging subspace capable of dynamic learning and self-correction is constructed, enabling accurate modeling and tracking of individualized aging characteristics of batteries. Since the pre-defined initial baseline subspace only represents a theoretical standard aging pattern, and the actual aging trajectory of each battery deviates due to its unique operational history, this method utilizes the battery's own historical aging vector to continuously identify and learn "abnormal aging directions" that deviate from the standard pattern but have become inherent characteristics of the battery. These learned new directions are then integrated into and correct the original baseline subspace. This process transforms the "corrected aging subspace" from a universal model into a personalized model capable of accurately depicting the evolution of the battery's unique health state. Projective decomposition based on this battery-tailored subspace allows for a more accurate distinction between known, inherent aging behaviors (coplanar aging components) and truly unprecedented abnormal signals representing unknown faults (orthogonal abnormal components). This avoids misjudging individualized aging characteristics as unknown faults, improves the accuracy and reliability of unknown fault detection, and ultimately enhances the accuracy of capacity assessment reflecting the battery's true health state.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, after the step of selecting the minimum value among all dimensional components from the dimensional constraint capacity vector as the risk constraint capacity, the method further includes: dividing each dimensional component in the dimensional constraint capacity vector by the risk constraint capacity to obtain a collaborative risk feature vector; calculating the vector similarity between the collaborative risk feature vector and each historical risk signature stored in the historical risk feature database to obtain a collaborative similarity value set; mapping the maximum similarity value in the collaborative similarity value set through a preset nonlinear function to obtain a collaborative risk coefficient; multiplying the risk constraint capacity by the collaborative risk coefficient to obtain a collaborative risk capacity; and updating the risk constraint capacity based on the collaborative risk capacity.

[0017] By adopting the above technical solution, the focus is shifted from a single worst dimension to the overall "risk pattern" composed of the degradation degree of each dimension by normalizing the multi-dimensional constraint capacity vector into a collaborative risk feature vector. Since a single minimum capacity constraint value may not fully reveal the potential coupling risks between different feature dimensions, and specific risk "forms" or "signatures" are often associated with a more severe collaborative failure mode in the past, this paper proposes a method to improve capacity. By matching the current risk form with a historical risk feature database and calculating the collaborative risk coefficient based on the maximum similarity, a deeper inference can be made about the potential danger level of the current unknown fault based on historical experience. Finally, this coefficient is used to modify the original risk constraint capacity. This is equivalent to adding a risk superposition based on multi-dimensional collaborative failure modes on top of the single-dimensional "weakest link" assessment. This makes the final risk capacity not only reflect the limitation of the weakest link, but also incorporate the consideration of the specific risk mode formed by the co-evolution of the weakest link with other dimensions. This allows for more prudent and accurate capacity constraints on unknown faults with complex internal relationships that may lead to accelerated deterioration or systemic collapse, improving the safety and reliability of capacity assessment when encountering unknown risks with historical similarities.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, after the step of correcting the healthy capacity value by means of a preset fault correction algorithm corresponding to the preset fault type to obtain the actual capacity value of the target battery, the method further includes: when the actual capacity value is greater than or equal to the theoretical battery capacity, using the theoretical battery capacity as a physical boundary constraint, calculating the confidence level of the theoretical battery capacity; if the confidence level is higher than a preset confidence threshold, determining that the actual capacity value calculation is abnormal; forcibly clamping the actual capacity value to the theoretical battery capacity, or re-determining the actual capacity value based on a weighted fusion of the theoretical battery capacity and the actual capacity value, wherein the allocation of weights is positively correlated with the confidence level.

[0019] By adopting the above technical solution, a logic-based circuit breaker and calibration mechanism is established by introducing a high-confidence theoretical battery capacity as a physical boundary constraint. When the data-driven algorithm result shows a physically impossible "artificially high" value, this mechanism forces the evaluation result to be clamped to the physical upper limit, avoiding evaluation distortion caused by algorithm overfitting or abnormal noise. This not only ensures the absolute physical rationality of the final output capacity value, but also significantly improves the robustness and safety of the system under extreme operating conditions or data anomalies, eliminating safety hazards caused by capacity misjudgment.

[0020] In a second aspect, embodiments of this application provide a battery management system, including: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the battery management system to perform the method described in the first aspect and any possible implementation thereof.

[0021] Thirdly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a battery management system, cause the battery management system to perform the method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, this application provides a computer program product that, when run on a battery management system, causes the battery management system to perform the method described in the first aspect and any possible implementation thereof.

[0023] It is understood that the battery management system provided in the second aspect, the storage medium provided in the third aspect, and the computer program product provided in the fourth aspect are all used to execute the method provided in this application. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating a method for evaluating the capacity of a smart lithium iron phosphate battery in an embodiment of this application. Figure 2 This is another flowchart illustrating the intelligent lithium iron phosphate battery capacity evaluation method in the embodiments of this application; Figure 3 This is a schematic diagram of an exemplary hardware structure of the battery management system in an embodiment of this application. Detailed Implementation

[0025] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0026] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0027] In related technologies, battery management systems (BMS) often employ a combination of multi-parameter fusion and dynamic algorithms to achieve capacity assessment. This involves real-time acquisition of key parameters such as battery voltage, current, temperature, and charge-discharge cycle count, followed by preliminary calculations using basic algorithms like the ampere-hour integral method and open-circuit voltage method. The results are then corrected using equivalent circuit models or simplified electrochemical models. Some systems also incorporate machine learning algorithms to analyze capacity decay patterns in historical charge-discharge data and establish predictive models to improve the timeliness of the assessment. However, the real-time acquired parameter data often simultaneously includes both the "health characteristics" of long-term battery aging and the "state characteristics" of instantaneous operating condition fluctuations. This results in the capacity value calculated by existing assessment methods being a coupled result of these two characteristics, making it difficult to accurately reflect the battery's true health state and prone to misjudgment. Such misjudgment may prompt premature battery replacement, when in reality, the battery's "health characteristics" due to long-term aging have not reached the point where replacement is necessary, and the battery itself still possesses the capability for normal use.

[0028] The intelligent lithium iron phosphate battery capacity assessment method in this application calculates a historical feature vector that eliminates instantaneous operating condition fluctuations by matching operating conditions and combining the local curvature at the end of the aging trajectory. The difference between this vector and the current feature vector is used to obtain a pure aging vector, eliminating the influence of reversible interference caused by instantaneous operating conditions on capacity assessment. Based on the aging vector, the healthy capacity of the battery is calculated, so that the final actual capacity value can accurately reflect the irreversible health degradation caused by aging and remove the reversible interference caused by instantaneous operating conditions, thereby improving the accuracy of capacity assessment that reflects the true health state of the battery.

[0029] The following is combined with Figure 1 The method of the embodiments of this application will be described below.

[0030] Please see Figure 1 This is a flowchart illustrating a method for evaluating the capacity of a smart lithium iron phosphate battery in an embodiment of this application.

[0031] S101. Calculate the theoretical battery capacity of the target battery based on the real-time operating data of the target battery collected in real time.

[0032] Specifically, the system first collects the target battery's operating data in real time through a sensor network deployed inside and outside the battery pack. This data covers key parameters such as the battery's individual cell voltage, total voltage, charging and discharging current, temperature in each area, and number of cycles. The collection frequency is dynamically adjusted according to the operating conditions to ensure the timeliness and completeness of the data.

[0033] Subsequently, the theoretical battery capacity is calculated based on real-time operating data. Calculations are performed using the ampere-hour integral method, which integrates the charging and discharging current over time and incorporates the change in the current state of charge (SOC) to preliminarily estimate the theoretical battery capacity. Simultaneously, calculations are performed based on an equivalent circuit model. A pre-defined second-order RC equivalent circuit model is invoked, and voltage, current, and temperature data are input into the model. An extended Kalman filter algorithm is used to identify model parameters in real time, thereby deriving the theoretical battery capacity. Finally, the results from the two calculation methods are weighted and fused (the weights are dynamically adjusted according to their respective calculation confidence levels) to output a comprehensive theoretical battery capacity value, reflecting the battery's apparent capacity under current operating conditions, which can serve as a basis for the battery's current usable capacity.

[0034] Optionally, in some embodiments, the system may also employ a direct prediction method based on a deep learning model. Real-time operating data continuously collected within a preset time period (such as time series data of voltage, current, temperature, internal resistance, and expansion deformation over the past 5 minutes) is input into a trained convolutional neural network-long short-term memory network (CNN-LSTM) hybrid neural network model. This model extracts local features through convolutional layers and captures temporal dependencies through long short-term memory network (LSTM) layers. After one forward propagation calculation, it directly outputs the theoretical battery capacity prediction value. The model has been trained on battery experimental datasets covering various operating conditions and aging stages and can accurately learn the nonlinear mapping relationship between data and capacity.

[0035] S102. When the theoretical battery capacity is within the preset dangerous capacity range, calculate the historical feature vector based on the target feature vector corresponding to the historical operating data of each target and the local curvature of the end region in the battery aging trajectory of the target battery.

[0036] Among them, the target historical operating data is the operating data in the historical operating database that matches the current operating condition of the target battery.

[0037] Specifically, when the theoretical battery capacity is within the preset dangerous capacity range, the current operating parameters (including real-time temperature range, charge and discharge current fluctuation range, average SOC value, etc.) are constructed into a multi-dimensional retrieval vector. The cosine similarity algorithm is used to search the historical operation database, and historical data segments with similarity exceeding the preset threshold are selected as target historical operation data. Target feature vectors containing dimensions such as voltage offset, internal resistance change rate, and capacity decay factor are extracted from these data.

[0038] Meanwhile, curve fitting is performed on the terminal region of the battery aging trajectory (a time-series curve recording the aging state of the battery at different times) (such as the aging data points of the most recent 100 cycles). After fitting it into a quadratic curve using the least squares method, the second derivative of the terminal point of the curve is calculated to obtain the local curvature, thereby reflecting the recent aging rate change trend of the battery.

[0039] Subsequently, the system performs weighted adjustments on the target feature vector based on the local curvature: if the curvature is positive (accelerated aging), the aging-related dimension parameters in the target feature vector are adjusted incrementally according to the curvature; if the curvature is negative (decelerated aging), the corresponding decremental adjustment is applied according to the curvature. The adjusted target feature vector (if there are multiple, a weighted average is performed) is the historical feature vector.

[0040] If there is insufficient similar operating condition data in the historical database, a pre-trained Gaussian Process Regression (GPR) model is invoked. Based on a predetermined number of closest historical operating condition data, a virtual target feature vector adapted to the current operating condition is generated through interpolation. This is then corrected using local curvature to ensure the validity of the historical feature vector. This Gaussian Process Regression (GPR) model is a nonparametric probabilistic model based on Bayesian theory. It uses a kernel function to describe the nonlinear mapping relationship between input variables (operating condition parameters) and output variables (target feature vectors), outputting a prediction result with a probability distribution, thus quantifying prediction uncertainty.

[0041] During model training, samples covering various typical working conditions in a historical database are used as the training set. The input is a working condition parameter vector, and the output is the target feature vector under the corresponding working condition. During training, a squared exponential kernel function is used to measure the similarity between input samples. The kernel function hyperparameters (such as length scale and signal variance) are optimized by maximizing the marginal likelihood function, enabling the model to learn the potential correlation between working conditions and feature vectors. At the same time, cross-validation is used to adjust the hyperparameters to ensure the model's generalization ability.

[0042] Because the GPR model captures the similarity of input variables through kernel functions, it can construct the joint probability distribution of the input space using limited similar historical data. When new operating condition parameters are input, the probability distribution of the feature vector corresponding to the new input can be derived from the joint distribution of the training data based on the conditional probability formula. This enables interpolation to generate virtual target feature vectors that are adapted to the current operating condition. It is especially suitable for accurate mapping in scenarios with small samples or sparse operating condition data. Therefore, it can effectively generate reliable feature vectors for subsequent correction when historical data is insufficient.

[0043] S103. Perform vector difference operation between the current feature vector and the historical feature vector to obtain the aging vector.

[0044] Here, the current feature vector is the feature vector that represents the current state of the target battery.

[0045] Specifically, based on the current real-time operating data of the target battery, a current feature vector is constructed that is completely consistent with the dimensions and physical meaning of the historical feature vector. Its feature dimensions cover indicators that are sensitive to the battery health status, such as the average slope of the charging voltage curve within a specific SOC range, the time to reach a certain voltage plateau during constant current charging, ohmic internal resistance and polarization internal resistance, etc. The peak position, height and other features can be extracted from the current voltage-capacity curve through incremental capacity analysis (ICA), or constructed through the electrochemical impedance spectroscopy (EIS) features at key frequencies (such as the real and imaginary parts of impedance at 1Hz and 10Hz).

[0046] Subsequently, the current feature vector and the historical feature vector are subjected to Z-score standardization (based on the mean and standard deviation of each feature dimension in the historical database) to eliminate the influence of different physical units and numerical ranges. Then, element-wise subtraction is performed to obtain the difference vector, which is the aging vector. The direction of this vector represents the degradation path of the battery health status in the multi-dimensional feature space, and the magnitude represents the overall magnitude of degradation, providing a core basis for distinguishing normal aging from faults.

[0047] S104. When a target battery is detected to have a fault of a preset fault type, and is still maintaining the charging and discharging function, the aging vector is decomposed into the actual aging vector and the fault vector, and the actual aging vector is added to the end of the battery aging trajectory.

[0048] Specifically, the system continuously collects key operating parameters of the target battery through a real-time status monitoring module and compares them with preset fault alarm thresholds (such as voltage upper and lower limits, temperature critical values, current fluctuation thresholds, etc.). When a parameter is detected to exceed the threshold and the duration exceeds the filtering window, a preliminary fault alarm is triggered. Subsequently, a preset fault feature library (which stores typical feature maps of preset fault types such as internal micro-short circuits and electrode corrosion, including abnormal parameter combination patterns and time series change patterns) is called. The real-time abnormal parameter sequence is matched with the features in the library using dynamic time warping (DTW). If the matching degree exceeds the preset matching threshold, the fault type is confirmed to belong to the preset category. At the same time, the system checks whether the battery is still maintaining charging and discharging functions through functional validity verification: it checks whether the current charging and discharging efficiency is maintained at the preset standard of the rated value, whether the output voltage is stable within the working range, and whether the core safety parameters (such as the estimated value of internal short circuit resistance) have reached the emergency shutdown threshold. If the above conditions are met, it is determined that the battery is "still maintaining charging and discharging functions".

[0049] When a target battery is detected to have a fault of a preset fault type, but is still maintaining charge / discharge functionality, the following steps are taken: First, the aging vector is compared with the target battery's health baseline at its current life stage to calculate the abnormal deviation vector. The health baseline includes one or more baseline aging vectors. Next, the cumulative equivalent charge / discharge cycles (EFC) of the target battery since its inception are obtained from the Battery Management System (BMS). This number is obtained by recording the amount of charge / discharge during each charge / discharge cycle in real time and converting it to full-capacity cycle accumulation. Then, the cumulative EFC is compared with a preset life stage threshold to determine the target battery's current life stage.

[0050] For this lifecycle stage, two pre-constructed health baselines are invoked: First, a "group health baseline" built based on extensive aging experiment data from fault-free batteries of the same model. This baseline is presented using a multidimensional Gaussian distribution model, including a mean vector (average aging characteristics of the group) and a covariance matrix (range of health fluctuations), defining the objective boundaries of healthy aging at this stage. Second, an "individual healthy aging baseline" for the target battery, determined through a sliding window: a buffer is maintained for the target battery to store only the most recently confirmed fault-free cyclic aging vector (data must meet the requirements of no abnormal deviations and no fault alarms in the early stages). When the data volume exceeds a preset amount, a third-order polynomial is fitted to each feature dimension (such as internal resistance increment and voltage offset) with the cycle number as the horizontal axis. The current cycle number is substituted into the fitted curve in real time to obtain the predicted values ​​of each dimension, which are then combined into the individual expected aging vector. Each newly added valid data point removes the oldest data and refits the data; this process is not activated when there are new batteries or insufficient data.

[0051] Next, a baseline validity check is performed, and a healthy baseline is selected. If the "Individual Healthy Aging Baseline" is enabled (i.e., an individual's expected aging vector has been generated), the Mahalanobis distance of this individual's expected aging vector relative to the multidimensional Gaussian distribution of the "Group Healthy Baseline" is calculated. If the distance is less than a preset threshold, it indicates that the individual's expectation is within the range of group health fluctuations, and the individual baseline is valid; it is then selected as the final healthy baseline. If the distance exceeds the threshold, it indicates that the individual's aging trend deviates from the group's healthy range, and the individual baseline is unreliable; in this case, the mean vector of the "Group Healthy Baseline" is selected as the final healthy baseline. If the "Individual Healthy Aging Baseline" is not enabled (due to new batteries or insufficient data), the mean vector of the "Group Healthy Baseline" is directly used as the healthy baseline.

[0052] Next, the aging vector is subtracted from the selected health baseline dimension by dimension to obtain the original difference vector. Then, the Mahalanobis distance of the original difference vector in the multidimensional Gaussian distribution model of the "group health baseline" is calculated. If the distance exceeds a preset threshold, it indicates that the deviation exceeds the normal aging fluctuation range, and the original difference vector is identified as an abnormal deviation vector. If it does not exceed the threshold, it is determined to be a normal aging fluctuation, and the abnormal deviation vector is a zero vector, indicating that the current aging conforms to the health pattern.

[0053] Secondly, the abnormal deviation vector is logically matched with a preset fault feature rule base to obtain the target fault mode of the target battery. First, the preset fault feature rule base is invoked. This rule base is constructed based on fault injection experiments, electrochemical simulation data, and expert knowledge, and contains feature patterns (which can be multi-dimensional threshold rule sets or standardized prototype abnormal vectors) corresponding to each preset fault type (such as internal micro-short circuits, active material deactivation, etc.). Then, the abnormal deviation vector is matched with the rule base through a combination of logical judgment and pattern recognition: first, the features of each dimension of the abnormal deviation vector are analyzed to check whether it conforms to the core rules of a certain type of fault (such as the combination condition of specific dimension values ​​exceeding the threshold), and at the same time, the similarity between the vector and various fault prototype vectors in the library (such as cosine similarity or RBF kernel function value) is calculated. If the rule matching degree of a certain fault mode reaches the preset threshold and the similarity is the highest, the system identifies it as the target fault mode; if the matching degree of a single fault does not reach the threshold but is close to the "suspected composite" threshold, the system starts the sparse linear decomposition process. Through the least squares algorithm with L1 norm regularization, the abnormal deviation vector is decomposed into a linear combination of multiple prototype vectors. If the coefficients of some prototype vectors after decomposition are significantly non-zero, it is determined to be the composite mode of the corresponding fault, and the coefficients reflect the contribution weight of each fault.

[0054] Third, the abnormal deviation vector is projected onto the preset feature fault signature corresponding to the target fault mode to obtain the fault vector. The preset feature fault signature is a standardized reference vector, and the direction of the vector defines the degradation path of the battery health state change caused by the target fault mode. According to the determined target fault mode, the corresponding preset feature fault signature is retrieved from the database. This signature is a standardized unit vector (the direction corresponds to the fault feature degradation path) calibrated through multiple fault injection experiments. Let the abnormal deviation vector be A, and the fault signature be a unit vector S. The system calculates the projection component of A in the direction of S through the vector dot product, that is, the fault vector F=(A・S)・S, where the direction of F is consistent with S, the scalar (A・S) determines the magnitude (fault severity), and the unit vector S determines the direction (degradation path). If the fault signature is temperature-dependent, the system first obtains the current real-time temperature of the battery, calls the temperature compensation function obtained by fitting fault experimental data at multiple temperature points, dynamically corrects S to obtain a signature vector S_T adapted to the current temperature, and then calculates F according to the above projection formula.

[0055] Fourth, based on the target fault mode and the magnitude of the fault vector, the corresponding coupling effect quantification equation is retrieved from the pre-set fault and aging coupling effect knowledge base. First, the pre-set fault and aging coupling effect knowledge base is invoked. This knowledge base is constructed based on multi-factor accelerated aging tests (covering different combinations of fault types, severity, temperature, and cycling stages) and electrochemical simulation data. Coupling effect quantification equations are stored according to fault modes. Each equation uses fault severity (fault vector magnitude) as the core independent variable, while incorporating the current battery temperature and the stage of cumulative EFC as adjustment parameters. It outputs the acceleration factor function relationship for each aging characteristic dimension (such as capacity decay rate and internal resistance growth rate) (the form can be a piecewise exponential function, a polynomial, or a nonlinear model constrained by physical mechanisms). Then, using the target fault mode as a unique identifier, the corresponding dedicated sub-base in the knowledge base is located (the sub-base pre-stores the mapping relationship between different operating parameters under this fault mode and the coupling effect quantification equation). Next, the fault vector magnitude, real-time battery temperature, and cumulative EFC value under the current scenario are extracted as core retrieval parameters. Multi-dimensional parameter matching is performed in the sub-database (the matching dimension corresponds to the above three parameters, and rapid location is achieved by constructing a parameter space index). If the retrieved parameter combination is within the sub-database calibration range (i.e., all three parameter values ​​fall within the valid range of pre-stored data), a cubic spline interpolation algorithm is used to perform continuous fitting based on the discrete operating condition data within this range in the sub-database, generating a coupling effect quantification equation adapted to the current real-time operating condition. If the fault vector magnitude exceeds the sub-database calibration upper limit (other parameters are still within the range), a preset Sigmoid constraint function is called (its parameters are determined by fitting measured data within the sub-database calibration range to ensure consistency with known laws). This function constrains the acceleration factor. As the magnitude increases, the acceleration factor will gradually approach the physically reasonable upper limit (e.g., the reaction limit rate when the active material is completely deactivated), avoiding deviations from actual physical laws caused by unconstrained extrapolation. Finally, the coupling effect quantification equation obtained by the above retrieval interpolation or extrapolation calculation is output.

[0056] Fifth, substitute the magnitude of the fault vector into the coupling effect quantization equation to calculate the acceleration factor for each affected aging feature dimension, and construct the acceleration transformation matrix. First, substitute the magnitude of the fault vector, the current battery temperature, and the cumulative EFC value into the coupling effect quantization equation obtained in the previous step to calculate the acceleration factor corresponding to each aging feature dimension. For dimensions affected by the fault, the acceleration factor is a quantization value greater than 1 (reflecting the amplification factor of the aging rate); for unaffected dimensions, the acceleration factor is set to 1 by default. Then, using these acceleration factors as diagonal elements, construct a diagonal matrix with the same dimensions as the feature vector, i.e., the acceleration transformation matrix (all off-diagonal elements are 0 to ensure that the acceleration effects of each dimension are independent).

[0057] Sixth, based on the aging vector and the accelerated transformation matrix, the theoretical aging vector is obtained by solving a preset inverse optimization problem. The predicted vector after the theoretical aging vector is transformed by the accelerated transformation matrix has the smallest error with the aging vector.

[0058] The system first defines the objective function of the pre-defined inverse optimization problem as: minV theo (||T×V theo -V obs || 2 +λ×||V theo || 2 ), where the first term is the prediction error term, namely the acceleration transformation matrix T and the theoretical aging vector V. theo The product of (predicted vector) and the observed aging vector V obs The squared Euclidean distance between the predicted and actual values ​​is used to measure the deviation between the predicted and actual values. The second term is the Tikhonov regularization term, where λ is a preset regularization parameter (calibrated using offline experimental data, typically set to 0.01-0.1). By constraining the squared magnitude of the theoretical aging vector, it balances the fitting accuracy with the stability of the solution, avoiding drastic fluctuations in the solution due to matrix ill-conditionedness. The system finds the optimal V by solving this objective function. theo For low-dimensional spaces with feature dimensions less than a preset number, directly calculate the Moore-Ponros pseudo-inverse matrix T of T. + via V theo =T + ×Vobs yields analytical solutions; for high-dimensional spaces or when the number of T conditions exceeds the preset number of conditions (ill-conditioned conditions exist), the L-BFGS iterative algorithm is used, with V... obs The result of the inverse transformation of the diagonal elements of T is the initial guess value, and V is updated iteratively through gradient descent. theo (Calculate the objective function with respect to V in each iteration) theo The gradient is adjusted along the negative gradient direction until the objective function value drops below a preset percentage of the initial value or the maximum number of iterations is reached. Simultaneously, physical constraints are imposed on the solution: V is forced... theo Each feature dimension must be non-negative (because aging features represent cumulative degradation and cannot be negative). If a negative value appears during iteration, it is truncated to 0, resulting in the final V. theo This is the normal aging state vector when there are no faults.

[0059] Seventh, multiply the theoretical aging vector by the acceleration transformation matrix to obtain the coupled aging increment vector. First, multiply the theoretical aging vector by the acceleration transformation matrix to obtain the total aging prediction vector containing the fault acceleration effect. Then, remove the theoretical aging vector from the total aging prediction vector by vector subtraction. The difference between the two is the coupled aging increment vector. For high-dimensional feature spaces or scenarios with weak fault effects (matrix T is close to the identity matrix), to avoid loss of numerical accuracy, the system uses incremental matrix optimization calculation: first construct the increment matrix ΔT = T − I (where I is the identity matrix of the same dimension), and then directly use V... increment =ΔT×V theo Obtain the results. If the fault has a constant additional aging effect independent of the aging rate (such as the persistent effect of a specific chemical side reaction), the system retrieves the corresponding additional aging bias vector C from the preset fault and aging coupling effect knowledge base, and then uses V to... increment =ΔT×V theo The +C correction ensures that both the multiplicative acceleration effect and the additive superposition effect are covered, and the resulting coupled aging increment vector fully reflects the additional impact of the fault on the normal aging process.

[0060] Finally, the theoretical aging vector and the coupled aging increment vector are summed to obtain the actual aging vector. The theoretical aging vector and the coupled aging increment vector are summed dimension by dimension by vector addition to obtain the actual aging vector. This vector fully reflects the actual aging state of the battery under the influence of faults, and the actual aging vector is added to the end of the battery aging trajectory.

[0061] It should be noted that the high-dimensional vector operations, matrix transformations, iterative optimization problems (such as inverse optimization), and large-scale knowledge base retrieval and matching involved in this embodiment have certain computational resource requirements. Therefore, these computational tasks can be executed on a high-performance computing platform in the cloud. The cloud platform can provide elastic, on-demand computing power and storage resources to process real-time data streams from a large number of battery devices and execute complex algorithm models. When specific steps with extremely high real-time requirements (such as real-time data acquisition, preliminary data preprocessing, or preliminary fault triggering determination) need to be processed locally in the Battery Management System (BMS), the BMS will perform corresponding data preprocessing and instruction forwarding, send the processed data to the cloud for in-depth analysis, and apply the capacity assessment results or correction instructions returned from the cloud to the local system. This "end-cloud collaboration" deployment mode fully utilizes the real-time response capability of the BMS and the powerful computing capabilities of the cloud platform, ensuring the feasibility of the method and the accuracy of the assessment.

[0062] In some embodiments, these computational tasks can be offloaded to the vehicle for edge computing. The vehicle's high computing power and ample computing resources are utilized to execute these tasks, and the results are then fed back to the BMS.

[0063] S105. Based on the anchor point operation data and battery aging trajectory obtained from the historical operation database, update the preset calibration function to obtain a new calibration function.

[0064] Among them, the anchor point operation data are the operation data in the historical operation data where the confidence level of the capacity calculation result is higher than the preset threshold.

[0065] Specifically, anchor point operation data is first screened from the historical operation database. The historical capacity calculation results are scored using a preset confidence evaluation function (combining ampere-hour integral integrity, open-circuit voltage resting time compliance rate, and model residual root mean square). Data with scores higher than the threshold are selected as anchor points. Simultaneously, the actual aging vectors recorded in the battery aging trajectory for the timestamps corresponding to these anchor points are extracted to form a data pair sample set of (actual aging vector, high confidence capacity value).

[0066] Next, using this sample set as training data, incremental updates are performed on the pre-defined calibration function (such as a multilayer perceptron or Gaussian process regression model that maps multidimensional aging features to capacity values). For neural network models, mini-batch gradient descent is used to fine-tune network weights based on new samples, while historical samples are assigned dynamic weights according to a time decay factor to avoid excessive influence from old data. For Gaussian process models, new samples are directly added to the training set and the covariance matrix is ​​updated, optimizing the kernel function parameters by maximizing marginal likelihood estimation. A population health baseline constraint is introduced during the update process: when the number of new samples is insufficient, pseudo-anchor data matching the current battery aging stage is extracted from the aging database of the same battery model to assist training and prevent model overfitting.

[0067] The neural network model is a mathematical model constructed through multiple layers of processing units (neurons). It establishes a mapping by learning the nonlinear relationship between aging characteristics and capacity. The initial model is trained based on a large amount of historical aging data (features and corresponding capacities) of batteries of the same model. The prediction error is minimized by adjusting the connection weights between neurons. Subsequently, for each new anchor data point, the weights are fine-tuned using mini-batch gradient descent, while reducing the influence weight of old data. When data is insufficient, pseudo-anchor data of similar batteries is introduced to constrain the model. The final new model can directly input the latest aging data and output the fault-free healthy capacity. The Gaussian process model is a probabilistic mathematical model based on the similarity (covariance) between data points. It constructs a mapping by analyzing the distribution patterns of aging characteristics. The initial model uses the distribution characteristics of historical data to determine the kernel function parameters (rules describing feature similarity). When new anchor data is added, it is included in the dataset and the data distribution is recalculated. The model is updated by optimizing the kernel function parameters. When data is insufficient, pseudo-anchor data is also introduced to assist. The final new model can calculate the fault-free healthy capacity and prediction confidence based on the similarity between the latest aging data and historical data.

[0068] Finally, the prediction error of the updated function is checked using the reserved set of verification anchor points. If the mean absolute error is less than the preset error, it is confirmed to be effective, and a new calibration function is generated for subsequent capacity assessment. Otherwise, the update parameters are adjusted and the process is repeated.

[0069] S106. Substitute the latest aging data from the battery aging trajectory into the new calibration function to calculate the healthy capacity value of the target battery under fault-free conditions.

[0070] Specifically, the latest recorded actual aging vector is first extracted from the battery aging trajectory and used as input data to the new calibration function. If the new calibration function is a neural network model, the vector is input into the network's input layer. Through layer-by-layer weighted calculations and nonlinear activations (such as ReLU transformation) of neurons, the scalar value of the output layer is obtained after forward propagation, which is the health capacity. If it is a Gaussian process model, the posterior probability distribution is calculated based on the similarity between the vector and historical anchor data using a kernel function. The mean of the distribution is taken as the health capacity, and the variance can be output as a confidence index.

[0071] S107. Based on the fault vector, the healthy capacity value is corrected by a preset fault correction algorithm corresponding to the preset fault type to obtain the actual capacity value of the target battery.

[0072] Specifically, based on the target fault mode, the corresponding dedicated correction algorithm is first called from the preset fault correction algorithm library. Simultaneously, the fault vector magnitude and real-time operating parameters corresponding to the fault are extracted as input. The correction algorithm calculates the fault capacity correction amount used to correct the healthy capacity value. Subsequently, according to the preset combination rules of the correction algorithm (such as subtraction or multiplication), the fault capacity correction amount and the healthy capacity value are calculated to finally obtain the actual capacity value of the target battery. This value represents the actual output capacity of the battery under the current fault state after removing external operating condition interference. It includes both irreversible capacity loss caused by long-term battery aging and temporary or continuous capacity loss directly caused by the current fault. It is the most intuitive quantification of the battery's current actual energy storage capacity and can serve as the core basis for battery life assessment and fault early warning.

[0073] In this embodiment, by matching operating conditions and combining the local curvature at the end of the aging trajectory, a historical feature vector that can eliminate instantaneous operating condition fluctuations is calculated. The difference between this vector and the current feature vector is used to obtain a pure aging vector, thus eliminating the impact of reversible interference caused by instantaneous operating conditions on capacity assessment. Simultaneously, considering the different impact mechanisms of faults and aging, when a preset fault is detected, the aging vector is further decomposed into an actual aging vector and a fault vector. Based on the actual aging vector, the healthy capacity under fault-free conditions is calculated, and then a corresponding fault correction algorithm is applied for correction. This ensures that the final actual capacity value accurately reflects the irreversible health degradation caused by both aging and faults, while eliminating reversible interference from instantaneous operating conditions, thereby improving the accuracy of capacity assessment reflecting the true health state of the battery.

[0074] The following is combined with Figure 2 The methods of the embodiments of this application will be further explained below.

[0075] Please see Figure 2 This is another flowchart illustrating the intelligent lithium iron phosphate battery capacity evaluation method in this application embodiment.

[0076] S201. Calculate the theoretical battery capacity of the target battery based on the real-time operating data of the target battery collected in real time.

[0077] S202. When the theoretical battery capacity is within the preset dangerous capacity range, calculate the historical feature vector based on the target feature vector corresponding to the historical operating data of each target and the local curvature of the end region in the battery aging trajectory of the target battery.

[0078] S203. Perform vector difference operation between the current feature vector and the historical feature vector to obtain the aging vector.

[0079] S204. When a target battery is detected to have a fault of a preset fault type, and is still maintaining the charging and discharging function, the aging vector is decomposed into an actual aging vector and a fault vector, and the actual aging vector is added to the end of the battery aging trajectory.

[0080] S205. Based on the anchor point operation data and battery aging trajectory obtained from the historical operation database, update the preset calibration function to obtain a new calibration function.

[0081] S206. Substitute the latest aging data from the battery aging trajectory into the new calibration function to calculate the healthy capacity value of the target battery under fault-free conditions.

[0082] S207. Correct the healthy capacity value by using a preset fault correction algorithm corresponding to the preset fault type to obtain the actual capacity value of the target battery.

[0083] Steps S201-S207 and Figure 1 Steps S101-S107 in the illustrated embodiment are similar and can be found in the descriptions of steps S101-S107, which will not be repeated here.

[0084] S208. When a fault not belonging to the preset fault type is detected in the target battery, and the charging and discharging function is still maintained, the aging vector is decomposed by subspace projection based on the aging subspace composed of one or more historical aging vectors under historical health conditions, to obtain the coplanar aging component that can be explained by historical experience and the orthogonal abnormal component representing the unknown fault.

[0085] Specifically, firstly, the initial baseline subspace for the theoretical healthy aging behavior of the target battery at its current life stage is queried from a pre-defined historical aging subspace library. This library stores aging subspaces for standard aging modes at each stage of the target battery's entire life cycle. The current cumulative equivalent full-charge-discharge cycle count of the target battery is obtained to determine its life cycle stage. Then, the initial baseline subspace corresponding to the target battery's life cycle stage is queried from the pre-defined historical aging subspace library. This library stores standard subspace data for the healthy aging modes of batteries of the same model at each stage, categorized by life cycle stage.

[0086] Secondly, the aging vector corresponding to each trajectory in the battery aging trajectory is projected sequentially into the initial reference subspace to obtain the set of trajectory orthogonal components. First, the battery aging trajectory is retrieved, and all historical aging vectors marked as healthy (i.e., no fault alarms triggered) are extracted. These vectors record the multidimensional characteristics of normal aging of the target battery at different times. For each extracted historical aging vector, the system performs projection decomposition according to the definition of the initial reference subspace: if the initial reference subspace consists of a set of orthogonal basis vectors, the system calculates the projection coefficients by multiplying the vector with each basis vector, then weights and sums the coefficients with the corresponding basis vectors to obtain the projection component of the historical aging vector in the subspace. The projection component is then subtracted from the original vector to obtain the trajectory orthogonal component orthogonal to the subspace. If the initial reference subspace is implicitly defined by the autoencoder model, the system inputs the historical aging vector into the autoencoder to obtain the reconstructed vector (i.e., the projection component in the subspace) output by the model, and calculates the trajectory orthogonal component by the difference between the original vector and the reconstructed vector.

[0087] Third, identify abnormal aging direction vectors in the orthogonal component set of the trajectory. First, preprocess all orthogonal components in the set to remove outliers caused by noise or instantaneous fluctuations (e.g., using the Z-score method to remove vectors with excessive deviations from the mean). Then, extract the direction using a combination of statistical and pattern recognition methods: If principal component analysis is used, first calculate the covariance matrix of the processed vector set, obtain the direction of each principal component through eigenvalue decomposition, and select the first few principal component vectors whose variance contribution rate exceeds a preset threshold. These vectors represent the most concentrated directions in the orthogonal component set, i.e., abnormal aging directions. If density clustering is used, first normalize the vectors to unit vectors, then use the DBSCAN algorithm to cluster based on the cosine similarity between vectors, identify high-density clusters containing more than a threshold number of samples, calculate the weighted average direction of all vectors within the cluster (assigning weights according to time, with more recent vectors having higher weights), and use this as the abnormal aging direction vector corresponding to that cluster.

[0088] Fourth, the abnormal aging direction vectors are used as correction operators to correct the initial reference subspace, resulting in a corrected aging subspace. First, orthogonality purification is performed on each abnormal aging direction vector: the direction vector is projected onto the initial reference subspace, the projected components are calculated, and then vector subtraction is used to obtain residual components that are completely orthogonal to the initial subspace (these residual components only contain new information that cannot be interpreted by the initial subspace). The residual components are then normalized to ensure their magnitude is 1 and their direction remains unchanged. Next, the purified residual components are merged with the basis vectors of the initial reference subspace to form a new vector set. If the Gram-Schmidt orthogonalization method is used, the system processes the vectors in the set sequentially, subtracting the projection of each vector onto the orthogonally transformed vector, and then normalizing it, ultimately obtaining a set of mutually orthogonal new basis vectors. If based on an autoencoder model, the feature patterns corresponding to the residual components are incorporated into the model's latent space by fine-tuning the encoder parameters, enabling the latent space to more accurately map aging vectors containing individual features. The linear space spanned by these new basis vectors (or the expanded latent space) is the modified aging subspace.

[0089] Finally, based on the modified aging subspace, the aging vector is decomposed by projection to obtain coplanar aging components that can be explained by historical experience and orthogonal anomaly components representing unknown faults. If the modified aging subspace is defined by a new set of standard orthogonal basis vectors, the system calculates the dot product of the current aging vector with each basis vector to obtain projection coefficients, and then weights and sums the coefficients with the corresponding basis vectors to generate coplanar aging components. These components cover all aging information that can be explained by both the general laws of the group and the historical characteristics of the individual battery. Subsequently, the coplanar aging components are subtracted from the original aging vector to obtain orthogonal anomaly components that are completely orthogonal to the modified subspace, which centrally reflect unknown fault signals that cannot be explained by known experience. If a fine-tuned autoencoder model is used, the system inputs the current aging vector into the model, and the output reconstructed vector is the coplanar aging component. The difference between the input and the reconstructed vector is the orthogonal anomaly component, thus achieving the decomposition of nonlinear aging characteristics.

[0090] S209. Based on historical feature vectors and corresponding historical state of charge data, calculate the feature sensitivity vector of each feature dimension relative to the change in unit charge.

[0091] Specifically, the system first selects multiple complete charge and discharge process data of the target battery under healthy conditions where the range of state of charge variation exceeds a preset threshold from the historical operation database, ensuring that the data covers a sufficient range of charge variation and is free from fault interference. For each feature dimension (such as internal resistance, voltage plateau, temperature response, etc.), the system extracts the change in its feature value in each charge and discharge process, and records the cumulative charge and discharge capacity of the process. The slope of the linear relationship between the two (i.e., the feature change rate corresponding to a unit change in capacity) is calculated through piecewise linear regression, and the average slope of multiple processes is taken as the initial sensitivity of that dimension.

[0092] To adapt to the dynamic changes in sensitivity during aging, the initial sensitivity is adjusted by calling a preset sensitivity-SOH function model (such as a polynomial or lookup table) based on the current state of battery health (SOH). Finally, the adjusted sensitivities of each dimension are combined into a feature sensitivity vector, which quantifies the sensitivity of each feature dimension to changes in unit charge.

[0093] S210. Perform a dimensional division operation on the absolute value of the orthogonal outlier component and the absolute value of the feature sensitivity vector to obtain the dimensional constraint capacity vector.

[0094] Specifically, first take the absolute value of each dimension of the orthogonal anomaly component and the feature sensitivity vector to eliminate the direction influence. Then, perform per-dimension division: for the i-th dimension, calculate the capacity loss risk, and the calculation formula is \(C_i = |orthogonal\ anomaly\ component_i| / |feature\ sensitivity_i|\), where \(C_i\) represents the capacity loss risk equivalent to the anomaly in this dimension. For dimensions where the feature sensitivity is close to zero (indicating that the feature is not sensitive to capacity changes), the system marks its \(C_i\) as an invalid value (such as a very large number).

[0095] At the same time, introduce the noise threshold \(N_i\) for each dimension. If \(|orthogonal\ anomaly\ component_i| < N_i\) (the signal is submerged by noise), then set \(C_i\) to a special value (such as a very large number) and do not participate in subsequent risk assessments.

[0096] Finally, combine all valid \(C_i\) into a dimension-constrained capacity vector, and each component reflects the equivalent capacity risk of the unknown fault from different feature dimensions respectively.

[0097] S211. Select the minimum value among all dimension components in the dimension-constrained capacity vector as the risk-constrained capacity.

[0098] Specifically, traverse all valid components of the dimension-constrained capacity vector (excluding invalid values and special values), and find the minimum value through a single traversal or sorting operation. This value is the risk-constrained capacity.

[0099] S212. Divide each dimension component in the dimension-constrained capacity vector by the risk-constrained capacity to obtain a collaborative risk feature vector.

[0100] Among them, collaborative risk refers to a systematic risk caused by an unknown fault and presenting a specific association pattern (i.e., "collaborative pattern" or "fault fingerprint") in the multi-dimensional feature space.

[0101] Specifically, taking the risk-constrained capacity as the benchmark, perform a division operation on each valid component of the dimension-constrained capacity vector: \(R_i = dimension\ constrained\ capacity_i / risk\ constrained\ capacity\). After the operation, each component of the collaborative risk feature vector \(R\) is a dimensionless value, where the minimum value is 1.0 (corresponding to the dimension with the greatest risk), and other components ≥ 1.0. The numerical size reflects the multiple of the risk of this dimension relative to the "bottleneck dimension". For invalid or special value components, uniformly assign a preset very large value, which does not affect the overall pattern. To prevent numerical overflow (such as when the risk-constrained capacity is extremely small), the system sets an upper limit for \(R_i\), and truncates it forcibly if it exceeds, ensuring that the vector components are within a reasonable range.

[0102] S213. Calculate the vector similarity between the collaborative risk feature vector and each historical risk signature stored in the historical risk feature library to obtain a set of collaborative similarity numerical values.

[0103] Specifically, the historical risk feature database (which stores collaborative risk feature vectors of historical unknown faults of batteries of the same model) is retrieved. Using the current collaborative risk feature vector as the query signature, all historical risk signatures in the database are traversed, and the similarity is calculated one by one to obtain a set of collaborative similarity values. If cosine similarity is used, directional similarity is calculated by the ratio of the vector dot product to the product of the magnitudes; if the RBF kernel function is used, similarity is calculated based on Euclidean distance (the smaller the distance, the higher the similarity).

[0104] S214. Map the maximum similarity value in the set of collaborative similarity values ​​through a preset nonlinear function to obtain the collaborative risk coefficient.

[0105] Specifically, the maximum similarity S_max is extracted from the set of collaborative similarity values, and then substituted into a preset nonlinear function (such as the sigmoid function K_risk=A / (1+exp(-k*(S_max-x0)))+B) for mapping. The calculated value of K_risk is the collaborative risk coefficient, and the value of K_risk is greater than 1. Among them, the function parameters A, k, x0, and B are calibrated using historical data: when S_max is close to 0, K_risk is close to 1 (without additional adjustment); when S_max increases, K_risk increases smoothly; when S_max is close to 1, K_risk approaches its upper limit (associating with the severity of historical failures).

[0106] S215. Multiply the risk constraint capacity by the collaborative risk coefficient to obtain the collaborative risk capacity.

[0107] Specifically, the risk constraint capacity (basic risk) and the collaborative risk coefficient (historical correction factor) of S214 are multiplied by a scalar: Collaborative Risk Capacity = Risk Constraint Capacity × Collaborative Risk Coefficient, yielding the collaborative risk capacity. This capacity integrates direct evidence of current anomalous signals with analogical adjustments based on historical experience. If the current fault is highly similar to a high-risk historical fault (K_risk>1), the risk is amplified; if it is similar to a low-risk fault or has no match (K_risk≈1), the basic risk is maintained. To quantify uncertainty, the system can represent both as probability distributions (such as Gaussian distributions). The probability distribution of the collaborative risk capacity is obtained through the product of these distributions. The mean is used as a point estimate, and the variance reflects the assessment uncertainty, providing more comprehensive risk information for subsequent decision-making.

[0108] S216. Update the risk constraint capacity based on the collaborative risk capacity.

[0109] Change the value of the risk constraint capacity to the value of the collaborative risk capacity.

[0110] S217. Select the smaller of the risk constraint capacity and the benchmark healthy capacity corresponding to the coplanar aging component as the actual capacity value of the target battery.

[0111] Specifically, the baseline healthy capacity corresponding to the coplanar aging component is first calculated. The coplanar aging component is then substituted into a preset health characteristic-capacity calibration function (this function is built based on historical healthy battery data and maps healthy aging characteristics to corresponding capacity values) to obtain the capacity value caused solely by normal aging after excluding unknown faults; this is the baseline healthy capacity. Subsequently, the baseline healthy capacity is compared with the risk-constrained capacity, and the smaller of the two is selected as the actual capacity value.

[0112] To prevent early-stage unknown faults from causing abnormally large risk constraint capacity (such as exceeding the battery's rated capacity) due to weak risk signals, the system will pre-set a capacity upper limit (such as 1.1 times the battery's rated capacity), truncate the risk constraint capacity exceeding the upper limit, and then compare it with the baseline healthy capacity. This ensures that even in the early stages of a fault, risk factors can have a reasonable impact on the actual capacity assessment. The final output actual capacity value is the most conservative and reliable quantitative result of the battery's current safe energy storage capacity.

[0113] S218. When the actual capacity value is greater than or equal to the theoretical battery capacity, the theoretical battery capacity is used as a physical boundary constraint to calculate the confidence level of the theoretical battery capacity.

[0114] In this step, the "theoretical battery capacity" is considered the upper limit of the current physical capabilities of the battery. This is based on the first principles of battery electrochemical reactions, namely that the usable capacity of a battery cannot exceed its current physical storage capacity. Although the theoretical battery capacity may fluctuate due to operating noise, its calculation results have clear physical meaning after multiple verifications based on ampere-hour integration, open-circuit voltage calibration, and electrochemical models. Once the aforementioned "actual capacity value" exceeds this theoretical upper limit, logically speaking, this is highly likely due to accumulated errors or over-correction introduced during feature extraction or vector operations in previous steps.

[0115] To quantify the reliability of the theoretical battery capacity as a boundary, the system needs to calculate its confidence level. The confidence level calculation comprehensively considers multiple dimensions: first, the quality of the input data, checking the zero-drift noise level of the current sensor during the resting period and the continuity of the sampled data (frame drop rate); second, the model fit, evaluating the voltage prediction residual (RMSE) of the equivalent circuit model in the most recent charge-discharge cycle; and finally, the state stability, examining the variance range of the battery's current state of charge (SOC) estimate.

[0116] The system normalizes and sums the above indicators using a preset weighted scoring formula to obtain a confidence value between 0 and 1. The higher the value, the stronger the reliability of the theoretical battery capacity as a physical boundary, and the better it can be used as a benchmark to judge whether the actual capacity calculation is abnormal.

[0117] S219. If the confidence level is higher than the preset confidence threshold, the actual capacity value is determined to be abnormal.

[0118] The system pre-sets a confidence threshold (e.g., 0.85 or 0.9, which can be calibrated based on the stability of the battery chemistry and the accuracy of the BMS hardware). When the confidence level calculated by S218 exceeds this threshold, it means that the current "theoretical battery capacity" is very accurate, and its value is almost close to the actual physical limit of the battery. At this point, if the "actual capacity value" is still greater than or equal to this high-confidence theoretical value, then according to the law of conservation of physics, the only explanation is that there was a deviation in the calculation process of the "actual capacity value". This deviation may stem from excessive differencing due to improper matching of historical eigenvectors, or from incorrect gain introduced during the decoupling of faults and aging.

[0119] Therefore, the system determines that the "actual capacity value calculation is abnormal." This abnormality is not a physical anomaly of the battery itself, but rather a calculation distortion at the algorithm level. Conversely, if the confidence level is below the threshold, it indicates that the theoretical capacity itself may be lower than expected due to sensor noise or model parameter drift (i.e., the theoretical value is unreliable). In this case, even if the actual capacity value is higher than expected, it cannot be hastily identified as abnormal. The system will not trigger subsequent forced corrections, but will retain the calculation results or only perform slight smoothing, thereby avoiding "false positives" when the theoretical benchmark is inaccurate.

[0120] It should be noted that the theoretical battery capacity calculation model has been rigorously calibrated. When the confidence level is above a threshold, the calculated result constitutes the upper limit of the battery's physical capacity under current conditions. This is because the model is based on first principles of electrochemistry and has been verified through extensive experiments. Even with operating noise, the calculated result can only be lower due to measurement errors, not artificially high due to noise. Therefore, when the actual capacity value exceeds the high-confidence theoretical capacity, it indicates that there must be a calculation anomaly in the capacity assessment process.

[0121] S220, Force the actual capacity value to be clamped to the theoretical battery capacity, or, redetermine the actual capacity value based on the weighted fusion of the theoretical battery capacity and the actual capacity value, wherein the allocation of weights is positively correlated with the confidence level.

[0122] Specifically, for calculation results deemed abnormal, this step provides two specific correction strategies to ensure that the final output capacity value is physically reasonable and safe. The first strategy is "forced clamping," suitable for scenarios with extremely high confidence (e.g., greater than 0.95). In this case, the system directly modifies the "actual capacity value" to the value of the "theoretical battery capacity," essentially truncating the overflowing non-physical portion and ensuring that the evaluation result does not violate the physical upper limit. The second strategy is "weighted fusion," suitable for transitional scenarios with high confidence but not reaching the extreme value (e.g., between 0.85 and 0.95). The system recalculates using the formula C_final = w * C_theory + (1-w) * C_actual. Here, the weight coefficient w is positively correlated with the confidence calculated by S218 (e.g., w = (confidence - threshold) / (1 - threshold)). This means that the more reliable the theoretical capacity, the larger its proportion in the final result, and the more the system tends to trust the physical boundary; conversely, it retains more of the actual capacity value component. Through this dynamic weighting mechanism, the system can smoothly transition the results of different calculation modes while eliminating obvious computational overshoot, avoiding abrupt changes in the output curve, and thus obtaining a final actual capacity value that conforms to both physical constraints and algorithm characteristics.

[0123] The above describes the intelligent lithium iron phosphate battery capacity evaluation method in the embodiments of this application. The following describes the battery management system in the embodiments of this application in detail with reference to the above intelligent lithium iron phosphate battery capacity evaluation method.

[0124] Please see Figure 3 This is a schematic diagram of an exemplary hardware structure of the battery management system in an embodiment of this application.

[0125] In some embodiments, the battery management system 300 includes a computer device, which may be a terminal device. The computer device includes a processor 301, a memory 302, a sensor module 303, a communication module 304, an input device 305, and an output device 306 connected via a system bus. The processor 301 provides computing and control capabilities. The memory 302 includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores data. The sensor module 303 collects operating data and condition data of the battery during operation. The communication module 304 transmits the collected operating data and condition data to the processor and sends battery status information to the user. The input device 305 receives user-input commands. The output device 306 displays battery status and battery capacity. When the computer program is executed by the processor 301, it implements the intelligent lithium iron phosphate battery capacity evaluation method in the embodiments of this application.

[0126] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0127] In some embodiments of this application, a computer-readable storage medium is provided, including instructions that, when executed on the battery management system 300, cause the battery management system 300 to perform the intelligent lithium iron phosphate battery capacity evaluation method of the embodiments of this application.

[0128] In some embodiments of this application, a computer program product is also provided, which, when run on a battery management system 300, causes the battery management system 300 to execute the intelligent lithium iron phosphate battery capacity evaluation method of the embodiments of this application.

[0129] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0130] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as meaning "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0131] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0132] 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. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for evaluating the capacity of intelligent lithium iron phosphate batteries, characterized in that, include: The theoretical battery capacity of the target battery is calculated based on the real-time operating data of the target battery collected in real time. When the theoretical battery capacity is within the preset dangerous capacity range, the historical feature vector is calculated based on the target feature vector corresponding to each target historical operation data and the local curvature of the end region in the battery aging trajectory of the target battery. The target historical operation data is the operation data in the historical operation database that matches the current operating condition of the target battery. The current feature vector is compared with the historical feature vector to obtain the aging vector. The current feature vector is the feature vector that represents the current state of the target battery. When a fault of a preset fault type is detected in the target battery, and the battery is still maintaining its charging and discharging function, the aging vector is decomposed into an actual aging vector and a fault vector, and the actual aging vector is added to the end of the battery aging trajectory. Based on the anchor point operation data obtained from the historical operation database and the battery aging trajectory, the preset calibration function is updated to obtain a new calibration function. The anchor point operation data refers to the operation data in the historical operation data whose confidence level of the capacity calculation result is higher than a preset threshold. The latest aging data in the battery aging trajectory is substituted into the new calibration function to calculate the healthy capacity value of the target battery under fault-free conditions. Based on the fault vector, the healthy capacity value is corrected by a preset fault correction algorithm corresponding to the preset fault type to obtain the actual capacity value of the target battery.

2. The method according to claim 1, characterized in that, The step of decomposing the aging vector into an actual aging vector and a fault vector specifically includes: The aging vector is compared with the health baseline of the target battery at the current stage of its life to calculate the abnormal deviation vector, wherein the health baseline includes one or more benchmark aging vectors. Logically match the abnormal deviation vector with a preset fault feature rule base to obtain the target fault mode of the target battery. The abnormal deviation vector is projected onto the preset feature fault signature corresponding to the target fault mode to obtain the fault vector. The preset feature fault signature is a standardized reference vector, and the direction of the vector defines the degradation path that causes the target fault mode to cause changes in battery health status. The fault vector is separated from the aging vector to obtain the actual aging vector.

3. The method according to claim 2, characterized in that, The step of separating the fault vector from the aging vector to obtain the actual aging vector specifically includes: Based on the target fault mode and the magnitude of the fault vector, the corresponding coupling effect quantification equation is retrieved from the preset fault and aging coupling effect knowledge base. Substitute the magnitude of the fault vector into the coupling effect quantification equation to calculate the acceleration factor for each affected aging feature dimension, and construct the acceleration transformation matrix. Based on the aging vector and the acceleration transformation matrix, a theoretical aging vector is obtained by solving a preset inverse optimization problem. The predicted vector after the theoretical aging vector is transformed by the acceleration transformation matrix has the smallest error with the aging vector. Multiply the theoretical aging vector by the acceleration transformation matrix to obtain the coupled aging increment vector; The actual aging vector is obtained by summing the theoretical aging vector and the coupled aging increment vector.

4. The method according to claim 1, characterized in that, After the step of performing vector difference operation between the current feature vector and the historical feature vector to obtain the aging vector, the method further includes: When a fault not belonging to the preset fault type is detected in the target battery, and the battery is still maintaining the charging and discharging function, the aging vector is decomposed by subspace projection based on the aging subspace composed of historical aging vectors under one or more historical health states, to obtain coplanar aging components that can be explained by historical experience and orthogonal abnormal components representing unknown faults. Based on the historical feature vectors and the corresponding historical state of charge data, calculate the feature sensitivity vector of each feature dimension relative to the change in unit charge. The absolute value of the orthogonal anomaly component is divided dimension by dimension by the absolute value of the feature sensitivity vector to obtain the dimension constraint capacity vector. From the dimensional constraint capacity vector, select the minimum value among all dimensional components as the risk constraint capacity; The smaller of the risk-constrained capacity and the baseline healthy capacity corresponding to the coplanar aging component is selected as the actual capacity value of the target battery.

5. The method according to claim 4, characterized in that, The subspace projection decomposition of the aging vectors, based on an aging subspace composed of one or more historical aging vectors under historical health states, yields coplanar aging components that can be explained by historical experience and orthogonal anomaly components representing unknown faults. Specifically, this includes: From the preset historical aging subspace library, query the initial benchmark subspace of the theoretical healthy aging behavior of the target battery at the current life stage. The historical aging subspace library stores the aging subspace of the target battery in each stage of the standard aging mode throughout its entire life cycle. The aging vector corresponding to each trajectory in the battery aging trajectory is sequentially projected onto the initial reference subspace to obtain the set of trajectory orthogonal components; Identify the abnormal aging direction vectors in the set of orthogonal components of the trajectory; The abnormal aging direction vector is used as a correction operator to correct the initial reference subspace, resulting in a corrected aging subspace; Based on the modified aging subspace, the aging vector is decomposed by projection to obtain coplanar aging components that can be explained by historical experience and orthogonal anomaly components representing unknown faults.

6. The method according to claim 4, characterized in that, After the step of selecting the minimum value among all dimensional components from the dimensional constraint capacity vector as the risk constraint capacity, the method further includes: Divide each dimension component of the dimensional constraint capacity vector by the risk constraint capacity to obtain the collaborative risk feature vector. The collaborative risk feature vector is compared with each historical risk signature stored in the historical risk feature database to calculate the vector similarity, thus obtaining a set of collaborative similarity values. The maximum similarity value in the set of collaborative similarity values ​​is mapped through a preset nonlinear function to obtain the collaborative risk coefficient; The risk constraint capacity is obtained by multiplying the risk constraint capacity by the collaborative risk coefficient. Update the risk constraint capacity based on the collaborative risk capacity.

7. The method according to claim 1, characterized in that, After the step of correcting the healthy capacity value using a preset fault correction algorithm corresponding to the preset fault type to obtain the actual capacity value of the target battery, the method further includes: When the actual capacity value is greater than or equal to the theoretical battery capacity, the theoretical battery capacity is used as a physical boundary constraint to calculate the confidence level of the theoretical battery capacity. If the confidence level is higher than the preset confidence threshold, the actual capacity value is determined to be abnormal. The actual capacity value is forcibly clamped to the theoretical battery capacity, or the actual capacity value is re-determined based on a weighted fusion of the theoretical battery capacity and the actual capacity value, wherein the allocation of weights is positively correlated with the confidence level.

8. A battery management system, characterized in that, include: One or more processors and memory; The memory is coupled to the one or more processors, the memory being used to store computer program code, the computer program code including computer instructions, the one or more processors invoking the computer instructions to cause the battery management system to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium storing computer instructions, characterized in that, When the computer instructions are executed on the battery management system, the battery management system performs the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product is run on the battery management system, it causes the battery management system to perform the method as described in any one of claims 1-7.

Images