Battery state evaluation method and system and storage medium

Abstract

The invention relates to the technical field of batteries, and discloses a battery state evaluation method and system and a storage medium, and the method comprises the steps: obtaining a manufacturing parameter set and an operation parameter set associated with a target battery; feature fusion is carried out on the manufacturing parameter set and the operation parameter set to obtain cross fusion features, and the cross fusion features are used for representing the interaction influence relation between the manufacturing parameter set and the operation parameter set; and performing state evaluation on the target battery based on the cross fusion feature to obtain an evaluation result. The battery health assessment method solves the problems of inaccurate assessment and large prediction deviation caused by the fact that an existing battery health assessment method cannot model a coupling effect of manufacturing and using.

Description

Technical Field

[0001] This invention relates to the field of battery technology, and specifically to a battery state assessment method, system, and storage medium. Background Technology

[0002] With the rapid development of new energy vehicles and the energy storage industry, accurate assessment of the health status of power batteries and management of fault risks have become crucial. Accurate determination of battery state of health (SOH) directly impacts system operational safety and asset value management. Currently, conventional battery health management methods primarily rely on data collected during use, such as capacity degradation statistics or optimized charging strategies. While these methods can reflect some aging phenomena, they struggle to comprehensively and deeply assess the complex performance degradation and potential fault risks caused by the combined effects of inherent manufacturing differences and post-use stresses.

[0003] Some existing technical solutions have attempted to introduce multi-dimensional data for more comprehensive battery health assessments. For example, health reports are generated by integrating multi-source information such as user behavior and environmental conditions through an edge-cloud collaborative architecture, or a health index is formed by combining multiple indicators such as voltage and internal resistance. However, these solutions have significant limitations at the data fusion level: they typically treat manufacturing process parameters (such as electrode characteristics and moisture content) and usage process data (such as fast charging frequency and temperature distribution) as parallel input features for comprehensive analysis, lacking feature engineering methods specifically for modeling the interaction between manufacturing and operating parameters. Due to the failure to effectively extract and utilize the cross-fusion features between the two types of parameters, existing methods struggle to accurately quantify the accelerated aging or failure modes triggered by specific manufacturing defects under specific usage conditions (e.g., the risk of gas generation in high-moisture batteries at high temperatures). This results in inaccurate assessments of the battery's health status throughout its entire lifecycle, biased predictions of remaining lifespan, and an inability to provide early warnings for failures caused by the coupling effect of manufacturing and usage. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide a battery state assessment method, system, and storage medium to solve the problems of inaccurate assessment and large prediction bias caused by the failure of existing battery health assessment methods to model the coupling effect between manufacturing and use.

[0005] In a first aspect, embodiments of the present invention provide a battery state assessment method, the method comprising:

[0006] Obtain the set of manufacturing parameters and the set of operating parameters associated with the target battery; The manufacturing parameter set and the operating parameter set are fused to obtain cross-fusion features, wherein the cross-fusion features are used to characterize the interaction and influence relationship between the manufacturing parameter set and the operating parameter set; The target battery is evaluated based on the cross-fusion features to obtain the evaluation results.

[0007] Furthermore, obtaining the set of manufacturing parameters and the set of operating parameters associated with the target battery includes: Obtain a pre-built battery base library, wherein the battery base library stores multiple preset manufacturing parameters of preset batteries, and the preset manufacturing parameters include at least electrode thickness, compaction density, formation efficiency, moisture content and material formulation batch. Obtain a pre-built battery runtime library, wherein the battery runtime library stores historical operating parameters of multiple preset batteries, and the historical operating parameters include at least the number of fast charge cycles, deep discharge frequency, single cell voltage difference, battery temperature difference, cumulative charge and discharge cycle count, and ambient temperature range. Obtain the set of manufacturing parameters for the target battery from the battery base library, and obtain the set of operating parameters for the target battery from the battery runtime library.

[0008] Furthermore, the feature fusion of the manufacturing parameter set and the operating parameter set to obtain cross-fused features includes: Static manufacturing features are extracted from the set of manufacturing parameters, wherein the static manufacturing features include at least one of the following: initial capacity, initial internal resistance, electrode thickness and deviation, compaction density, residual water content of material, negative electrode pre-lithiation rate, first-week efficiency of formation test, and electrode defect marking. Dynamic operating features are extracted from the set of operating parameters. The dynamic operating features include real-time state features and historical statistical features. The real-time state features include at least one of real-time state of charge value, real-time temperature value, and real-time individual cell pressure difference. The historical statistical features include at least one of cumulative cycle count, equivalent full charge cycle count, historical average depth of discharge, preset state of charge dwell time ratio, maximum temperature and pressure difference within a historical time period, and capacity reduction percentage within a historical time period. The static manufacturing features and the dynamic operation features are combined and calculated to generate the cross-fusion features.

[0009] Furthermore, the step of combining the static manufacturing features and the dynamic operational features to generate the cross-fusion features includes: Obtain a pre-built battery fault mode library, wherein the battery fault mode library stores a mapping relationship between preset fault modes and preset feature combinations; Based on the mapping relationship, at least one set of target static manufacturing features and target dynamic operating features associated with the target failure mode are determined from the static manufacturing features and the dynamic operating features; Logical combination operations are performed on the target static manufacturing features and target dynamic operation features associated with the target failure mode to generate the cross-fusion features.

[0010] Furthermore, the state assessment of the target battery based on the cross-fusion features to obtain the assessment result includes: Construct at least one hierarchical state index of the target battery based on the cross-fusion features; The target state value is obtained by weighted fusion of each of the hierarchical state indicators. The future state of the target battery is predicted based on the hierarchical state index, and the state prediction result is obtained. The hierarchical state index, the target state value, and the state prediction result are used as the evaluation results of the target battery.

[0011] Furthermore, constructing at least one hierarchical state index of the target battery based on the cross-fusion features includes: The capacity index of the target battery is calculated based on the cross-fusion features, wherein the capacity index is used to characterize the capacity decay state of the target battery; The consistency index of the target battery is calculated based on the cross-fusion characteristics, wherein the consistency index is used to characterize the performance differences between individual cells of the target battery; The risk index of the target battery is calculated based on the cross-fusion features, wherein the risk index is used to characterize the degree of potential failure risk of the target battery.

[0012] Furthermore, the prediction of the future state of the target battery based on the hierarchical state index to obtain the state prediction result includes: Obtain historical sequence data for each of the hierarchical state indicators; Calculate the predicted state indicators of each of the hierarchical state indicators in future time periods based on the historical sequence data. The remaining effective time and replacement reminder time of the target battery are calculated based on the predicted state indicators, and the remaining effective time and replacement reminder time are used as the state prediction results.

[0013] Furthermore, after evaluating the state of the target battery based on the cross-fusion features and obtaining the evaluation result, the method further includes: Obtain the business object tag associated with the target battery; For each of the aforementioned business object tags, generate early warning information corresponding to the evaluation results; The evaluation results and the early warning information are output to the business object corresponding to the business object label.

[0014] Secondly, embodiments of the present invention provide a battery state assessment system, the system comprising: The data acquisition module is used to acquire the set of manufacturing parameters and the set of operating parameters associated with the target battery; The feature fusion module is used to perform feature fusion on the manufacturing parameter set and the operating parameter set to obtain cross-fusion features, wherein the cross-fusion features are used to characterize the interaction and influence relationship between the manufacturing parameter set and the operating parameter set; The state assessment module is used to assess the state of the target battery based on the cross-fusion features and obtain the assessment result.

[0015] Thirdly, embodiments of the present invention provide a computer device, including: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the method described in the first aspect or any corresponding embodiment thereof.

[0016] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing computer instructions that cause a computer to perform the method described in the first aspect or any of its corresponding embodiments.

[0017] The method provided in this application has the following beneficial effects: The method provided in this application acquires the set of manufacturing parameters and the set of operating parameters of the target battery, comprehensively obtaining the battery's inherent manufacturing characteristics and acquired usage load, providing a comprehensive and multi-dimensional data foundation for subsequent evaluation; by performing feature fusion on the manufacturing parameters and operating parameters, a cross-fusion feature that can characterize the interaction between the two is constructed, thereby clarifying the complex failure modes under the coupling effect of manufacturing defects and usage behavior, and improving the ability to identify potential risks; based on the cross-fusion feature, the battery's state is evaluated, realizing the comprehensive quantification and accurate prediction of the battery's health status, remaining life and failure risk, forming an interpretable and traceable evaluation result, providing a reliable basis for battery safety management and value assessment. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1This is a schematic flowchart of a battery state assessment method according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating another battery state assessment method according to an embodiment of the present invention; Figure 3 This is a feature interaction and correlation analysis diagram according to an embodiment of the present invention; Figure 4 This is a multi-dimensional risk space visualization diagram according to an embodiment of the present invention; Figure 5 This is a performance comparison chart of the early warning methods according to embodiments of the present invention; Figure 6 This is a diagram illustrating the accuracy of early warning according to an embodiment of the present invention; Figure 7 This is a feature importance analysis diagram according to an embodiment of the present invention; Figure 8 This is a heat map of the fast charging × humidity interaction effect according to an embodiment of the present invention; Figure 9 This is a time series plot of SOH trajectory and risk score according to an embodiment of the present invention; Figure 10 This is a threshold boundary visualization and risk area division diagram according to an embodiment of the present invention; Figure 11 This is a depth quantization curve of the interaction effect according to an embodiment of the present invention; Figure 12 These are multi-scenario comparison verification diagrams according to embodiments of the present invention; Figure 13 This is a statistical comparison chart of risk level and characteristic distribution according to an embodiment of the present invention; Figure 14 This is a Top Contribution Feature Map according to an embodiment of the present invention; Figure 15 This is a comparison diagram of key features according to embodiments of the present invention; Figure 16 This is a SHAP contribution waterfall diagram according to an embodiment of the present invention; Figure 17 This is a contribution decomposition diagram according to an embodiment of the present invention; Figure 18 This is a feature distribution comparison diagram according to an embodiment of the present invention; Figure 19 This is a rule triggering frequency analysis diagram according to an embodiment of the present invention; Figure 20 This is a model confidence distribution diagram according to an embodiment of the present invention; Figure 21 This is a diagnostic path distribution diagram according to an embodiment of the present invention; Figure 22This is a rule and model consistency analysis diagram according to an embodiment of the present invention; Figure 23 This is a fusion decision cause analysis diagram according to an embodiment of the present invention; Figure 24 This is a cross-analysis diagram of diagnostic paths and fault types according to an embodiment of the present invention; Figure 25 This is a schematic diagram of the normal mode interface according to an embodiment of the present invention; Figure 26 This is a schematic diagram of a health index dashboard according to an embodiment of the present invention; Figure 27 This is a schematic diagram of a trend chart according to an embodiment of the present invention; Figure 28 This is a schematic diagram of an early warning panel according to an embodiment of the present invention; Figure 29 This is a schematic diagram of the ordinary mode model explanation panel according to an embodiment of the present invention; Figure 30 This is a schematic diagram of the expert mode interface according to an embodiment of the present invention; Figure 31 This is a schematic diagram of the expert mode model explanation panel according to an embodiment of the present invention; Figure 32 This is a structural block diagram of a battery state assessment system according to an embodiment of the present invention; Figure 33 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] According to embodiments of the present invention, a battery state assessment method, system, and storage medium are provided. It should be noted that the steps shown in the flowcharts in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowcharts, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0022] This embodiment provides a battery state assessment method. Figure 1 This is a flowchart of a battery state assessment method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: Obtain the set of manufacturing parameters and the set of operating parameters associated with the target battery.

[0023] In this embodiment, a pre-built structured database is accessed to extract static manufacturing data characterizing the inherent characteristics of the battery and dynamic operational data characterizing its usage history. Specifically, firstly, the battery's manufacturing parameter set is retrieved from the battery base library based on the unique identifier of the target battery (such as serial number or VIN code). This set includes key process and quality control parameters in the battery production process, such as "electrode thickness" and "compact density" used to measure electrode physical properties, "formation efficiency" reflecting the quality of the electrochemical formation process, "moisture content" affecting long-term reliability, and "material formulation batch" used to trace material sources and process consistency. Secondly, the battery's operational parameter set is obtained from the battery operational library based on the same battery identifier. This set includes time-series data and statistical characteristics collected by the BMS or OBD terminal during battery service, such as "number of fast charges" and "frequency of deep discharge" reflecting charging behavior, "single cell voltage difference" and "battery temperature difference" characterizing the cell equilibrium state, "cumulative charge-discharge cycle count" describing the battery's cumulative operating intensity, and "ambient temperature range" recording the influence of the external environment.

[0024] As an example, for a new energy vehicle with the target battery serial number ABC001, the following steps are taken: First, the pre-built battery database is queried based on this serial number to obtain its manufacturing parameter set, including: electrode thickness of 110μm for the positive electrode and 120μm for the negative electrode; compaction density of 3.4 g / cm³ for the positive electrode and 1.6 g / cm³ for the negative electrode; conversion efficiency of 95.2%; moisture content of 230ppm; and material formulation batch number of "positive electrode material FGH123-05, negative electrode graphite-G-09". Then, the battery operation database is queried based on the same serial number to obtain its operating parameter set for the past year, including: cumulative fast charge cycles of 152 times; deep discharge (DOD>80%) frequency of 15%; historical maximum single-cell voltage difference of 0.08V; maximum temperature difference within the battery pack of 8℃; total cumulative cycle count of 420 times; and recorded ambient operating temperature range of -10℃ to 45℃. At this point, the multi-source manufacturing and operational data set associated with the target battery has been successfully obtained.

[0025] Step S102: Perform feature fusion on the manufacturing parameter set and the operating parameter set to obtain cross-fusion features, wherein the cross-fusion features are used to characterize the interaction relationship between the manufacturing parameter set and the operating parameter set.

[0026] In this embodiment, based on prior knowledge of fault mechanisms, static manufacturing features characterizing the inherent properties of the battery and dynamic operating features characterizing its usage history are combined and calculated in a targeted manner to generate new features that can quantify the manufacturing-use coupling effect. First, features are extracted from the set: static manufacturing features (such as initial capacity, initial internal resistance, electrode thickness and deviation, compaction density, residual water content in materials, negative electrode pre-lithiation rate, first-week efficiency of formation test, electrode defect marking, etc.) reflect the inherent properties of the battery at the time of manufacture; dynamic operating features include instantaneous state features reflecting the instantaneous state (such as real-time state of charge (SOC), real-time temperature value, real-time single-cell pressure difference) and historical statistical features reflecting long-term cumulative effects (such as cumulative cycle count, equivalent full charge cycle count, historical average depth of discharge (DOD), preset SOC residence time ratio, maximum temperature and pressure difference within the historical cycle, capacity reduction percentage, etc.). Subsequently, by querying the battery fault mode library (which stores the mapping relationship between known fault modes and specific feature combinations), the target static manufacturing characteristics and target dynamic operating characteristics associated with the target fault mode (such as gas bulging, lithium dendrite precipitation, and accelerated degradation of consistency) are determined. Finally, logical combination operations (such as multiplication, ratio, weighted sum, or threshold-based piecewise function combination) are performed on these target features to generate the final cross-fusion features.

[0027] For example, to assess the risk of gas bulging, the residual water content of the material can be multiplied by the cumulative storage time at high temperatures (>40°C) as a cross-feature; to assess the risk of lithium plating, the compaction density can be combined with the percentage of fast charging times in the past 30 days. Through mechanism-driven feature engineering, the originally independent static and dynamic data are transformed into risk indicators that can directly indicate specific failure links.

[0028] Step S103: The state of the target battery is evaluated based on the cross-fusion features to obtain the evaluation results.

[0029] In this embodiment, a multi-layered quantitative evaluation system is constructed to comprehensively score the current state of the battery and predict its future trends. First, multi-dimensional hierarchical state indicators are calculated using cross-fusion features: a capacity index (quantifying the battery's energy storage capacity degradation state by combining current capacity retention rate, historical degradation rate, etc.), a consistency index (quantifying the degree of internal consistency degradation of the battery module by combining the dispersion of individual cell voltage / internal resistance, temperature difference, etc.), and a risk index (quantifying the potential probability of a battery safety accident by combining the frequency of events such as over-temperature and over-voltage with specific fault risks indicated by cross-features). Then, based on preset weights or weights obtained through machine learning, these hierarchical state indicators are weighted and fused to calculate a global target state value (e.g., a comprehensive health score of 0-100). The prediction phase then begins: historical sequence data for each tiered state indicator (e.g., monthly capacity index values ​​for the past 12 months) is acquired. Time series analysis or machine learning models (e.g., LSTM) are used to calculate the predicted state indicators for a future preset period (e.g., the next quarter). Based on these predictions, the remaining effective time (e.g., the estimated time required for battery capacity to decay to the failure threshold) and replacement reminder time (the recommended time for maintenance or replacement) can be further calculated. Both together constitute the state prediction result. Finally, the tiered state indicators reflecting the current state and the target state value, along with the future state prediction results, are packaged into a complete and structured evaluation result for the target battery.

[0030] In this embodiment of the application, obtaining the set of manufacturing parameters and the set of operating parameters associated with the target battery includes: Step A1: Obtain a pre-built battery base library, which stores multiple preset manufacturing parameters for preset batteries. The preset manufacturing parameters include at least electrode thickness, compaction density, formation efficiency, moisture content, and material formulation batch.

[0031] Specifically, it accesses or calls upon a centralized database that has been built and continuously maintained during the battery production stage. This database binds the production data of each battery to the battery entity through a unique battery ID, enabling digital file management for each battery and supporting personalized modeling and full lifecycle traceability for each battery, achieving individual difference modeling at the digital twin level. This battery base library is a structured data storage system whose core function is to persistently store the key static data generated by each battery throughout the entire manufacturing process before it leaves the factory, i.e., the preset manufacturing parameters. These parameters are not generated during operation, but rather are inherent genetic information of the battery, including at least: Electrode thickness: refers to the coating thickness of the positive and negative electrode active materials on the current collector, a key structural parameter affecting battery capacity, internal resistance, and mechanical stress; Compacted density: refers to the density of the electrode after the rolling process, directly affecting the porosity of the electrode and the lithium-ion transport rate; Formation efficiency: refers to the ratio of the reversible capacity after the formation of a stable SEI film during the first charge-discharge cycle to the initial charge capacity, a core indicator for measuring the quality of initial electrochemical formation; Moisture content: refers to the residual moisture content (usually expressed in ppm) in the electrolyte and materials before battery packaging, a key risk factor for triggering side reactions, gas generation leading to expansion, and safety degradation; Material formulation batch: refers to the specific chemical composition of the positive and negative electrode active materials used in the battery (such as the NCM ternary material ratio) and its production batch number, used to trace material characteristics and potential batch differences. These parameters are collected through the quality control system when the battery rolls off the production line and stored in this database after being bound to the battery's unique serial number (ID). By initiating a query request to the database using the unique identifier of the target battery (such as ID or VIN code), a complete, static set of manufacturing genetic data for the battery can be obtained for subsequent steps.

[0032] Step A2: Obtain the pre-built battery runtime library, which stores historical operating parameters of multiple preset batteries. The historical operating parameters include at least the number of fast charge cycles, deep discharge frequency, single cell voltage difference, battery temperature difference, cumulative charge and discharge cycle count, and ambient temperature range.

[0033] Specifically, it accesses a cloud or centralized database that continuously receives and stores dynamic usage data during the battery's service life. This battery runtime is a time-series database or big data platform, whose core function is to persistently store the historical operating parameters of each battery in real-world usage scenarios, collected and reported in real time by terminal devices such as the Battery Management System (BMS) and On-Board Diagnostics (OBD) test kits. These parameters record the battery's post-natal trajectory, including at least: Fast charging cycles: the cumulative number of times the battery is charged using high-power charging mode, used to assess the cumulative impact of high-rate charging stress on battery aging; Deep discharge frequency: characterizing the frequency with which the battery discharges to a low state of charge (SOC), used to quantify the accelerating effect of deep cycling on battery life; Individual cell voltage difference: recording the real-time voltage difference between individual cells within the battery pack, a key indicator for assessing the consistency and balance of the battery pack; Battery temperature difference: recording the real-time temperature difference at different monitoring points within the battery pack (e.g., between cells, between modules), reflecting the effectiveness of the thermal management system and the risk of thermal runaway; Cumulative charge-discharge cycle count: the cumulative number of times the battery undergoes a complete charge-discharge process, a fundamental indicator for measuring battery life and aging; Ambient temperature range: recording the external ambient temperature range in which the battery operates, used to analyze the impact of environmental thermal stress on battery performance and safety. A query request is initiated to the runtime library using the target battery's unique identifier (e.g., ID or VIN code) to obtain a dynamic set of operating load data accumulated since the battery was put into use.

[0034] Step A3: Obtain the set of manufacturing parameters for the target battery from the battery base library, and obtain the set of operating parameters for the target battery from the battery runtime library.

[0035] Specifically, based on the unique identifier of the target battery (such as the battery serial number or associated vehicle VIN code), data extraction is performed after data source access has been completed. Using this unique identifier as the query key, a request is first sent to the battery base library to retrieve and return all preset manufacturing parameters bound to that identifier. These parameters together constitute the manufacturing parameter set of the target battery. Next, using the same identifier, a request is sent to the battery runtime library to retrieve and return all relevant data recorded during the battery's historical operation. This data together constitute the operating parameter set of the target battery. By associating a general data source with a specific evaluation object through key identifiers, structured input data is provided for feature fusion.

[0036] In this embodiment of the application, feature fusion is performed on the manufacturing parameter set and the operating parameter set to obtain cross-fused features, including: Step B1: Extract static manufacturing features from the set of manufacturing parameters. The static manufacturing features include at least one of the following: initial capacity, initial internal resistance, electrode thickness and deviation, compaction density, residual water content of material, negative electrode pre-lithiation rate, first-week efficiency of formation test, and electrode defect marking.

[0037] Specifically, the raw manufacturing parameter set obtained from the battery base library is screened and transformed to extract quantitative indicators that have direct physical or chemical significance for battery life and reliability assessment, namely static manufacturing characteristics. These characteristics are essentially properties fixed at the factory, including at least the initial capacity. With initial internal resistance The nominal capacity and DC internal resistance of the battery, measured under standard testing conditions, serve as the benchmark values ​​for subsequent evaluation of capacity decay and internal resistance growth; electrode thickness. And deviation: refers to the average thickness of the positive and negative electrode coatings and their uniformity on the electrode sheet (usually expressed as standard deviation or maximum and minimum difference), reflecting the consistency of the coating process; compaction density The density of the electrode sheet after rolling directly affects the porosity and ion transport performance of the electrode; the residual water content of the material... The percentage of residual moisture in the internal materials (such as electrolyte and electrodes) before battery encapsulation is a key factor in inducing side reactions; negative electrode pre-lithiation rate: the ratio of the amount of lithium artificially added to the negative electrode before the first charge to the theoretically required amount of lithium, used to compensate for irreversible lithium loss during the first cycle; formation test first-week efficiency. The percentage of discharge capacity to charge capacity during the first charge-discharge cycle is a core indicator for evaluating the initial SEI film formation quality. Electrode defect labeling: Boolean (yes / no) labels indicate whether microscopic defects such as burrs, foreign objects, or poor welding were detected during the battery manufacturing process. By extracting these features with clear physical meaning, the raw manufacturing data is transformed into high-value inputs that can be directly understood and calculated by the model.

[0038] Step B2: Extract dynamic operating features from the set of operating parameters. The dynamic operating features include instantaneous state features and historical statistical features. Instantaneous state features include at least one of real-time state of charge value, real-time temperature value, and real-time individual cell pressure difference. Historical statistical features include at least one of cumulative cycle count, equivalent full charge cycle count, historical average depth of discharge, preset state of charge dwell time ratio, maximum temperature and pressure difference within the historical time period, and capacity reduction percentage within the historical time period.

[0039] Specifically, the time-series data obtained from the battery runtime library is processed and aggregated in multiple dimensions to extract quantitative indicators that reflect the current operating condition of the battery and the long-term cumulative aging effect, i.e., dynamic operating characteristics. The feature set is divided into two categories: one is instantaneous state features, representing the instantaneous state at the evaluation moment, including: real-time state of charge (SOC): the percentage of the battery's current remaining charge relative to its maximum usable capacity; and real-time temperature value. Current temperature of the battery pack or key cells; real-time cell pressure differential. The current difference between the highest and lowest individual cell voltages within the battery pack. Another type is historical statistical characteristics, obtained by statistically analyzing data within a historical time window. These are used to quantify long-term usage patterns and aging trends, including: cumulative cycle count. Total number of complete charge-discharge cycles since commissioning; Equivalent full charge cycles: Equivalent cycle count considering depth of discharge. This allows for more precise measurement of cumulative workload; historical average depth of discharge (DOD): the average depth of discharge per cycle within a statistical period; preset state of charge (SOC) dwell time ratio: the percentage of time the battery's SOC is in a preset high (e.g., >90%) or low (e.g., <20%) range; maximum temperature and pressure difference within a historical time period: the maximum temperature and pressure difference monitored within a specified period (e.g., the past 7 days); and percentage capacity decay within a historical time period: the percentage of capacity decay within a specified period (e.g., the past 30 days). This transforms raw time-series operational data into dynamic features with clear statistical significance and health indicator value, providing a foundation for subsequent interactive analysis with static features.

[0040] Step B3 involves combining and calculating static manufacturing features and dynamic operational features to generate cross-fusion features.

[0041] Specifically, after feature extraction, the health assessment module will use genetic features... and running feature sequences As input, an algorithm integrating preset rules and data-driven methods is used to calculate multi-dimensional battery health indicators (HI). A hierarchical indicator system is adopted: first, several secondary health indices are calculated, and then they are summarized into a final comprehensive health index (total score 0~100 points, the lower the score, the better the health and the lower the failure risk; in the following text, low score is good and high score is poor). For example, in the scenario of thermal runaway risk identification, the system does not rely solely on a single temperature threshold, but combines dynamic operating characteristics such as the rate of temperature rise, the trend of insulation resistance decline, and historical overheating records, and correlates them with the initial sealing process quality (manufacturing characteristics) of the battery to form a composite risk indicator. When a sharp increase in temperature is detected along with a sudden drop in insulation resistance, even if the absolute temperature value does not reach the traditional BMS threshold, it can be identified as a precursor to thermal runaway and an early warning can be issued.

[0042] In this embodiment of the application, static manufacturing features and dynamic operational features are combined and calculated to generate cross-fusion features, including: Step B301: Obtain a pre-built battery fault mode library, wherein the battery fault mode library stores the mapping relationship between preset fault modes and preset feature combinations.

[0043] Specifically, it invokes or loads a pre-built knowledge base based on battery domain knowledge (pre-defined experience) and historical fault data mining results. This battery fault mode library is a structured rule or knowledge base, the core of which is the mapping relationship between pre-defined fault modes and pre-defined feature combinations. Pre-defined fault modes refer to typical battery failure categories that have been recognized and defined, such as thermal runaway, capacity drop, accelerated consistency degradation, and lithium dendrite precipitation. Pre-defined feature combinations indicate the combination of one or more static manufacturing features and dynamic operating features that are typically associated with each fault mode during its occurrence or evolution, and have strong indicative power. This library not only contains the mapping between faults and features, but also records the combination features of manufacturing parameters and operating parameters when various faults occurred historically, for model training and inference reference. In addition, the fault library also defines health / fault indicators and their criteria, including capacity indicators, voltage indicators, internal resistance indicators, consistency indicators, etc., and associates them with corresponding health thresholds and fault judgment rules, forming a continuously evolving knowledge base generated by pre-defined knowledge and data mining.

[0044] For example, the combination of "high moisture content and duration of high-temperature environment" maps to the "gas-producing bulge" fault. This mapping relationship stores the logical association from feature combination to fault mode. The constructed knowledge base can be fully acquired into the current evaluation process's runtime environment through database query interfaces, file reading, or direct loading of data structures in memory, providing knowledge basis for the next step of locating relevant features based on the target fault mode.

[0045] Step B302: Based on the mapping relationship, determine at least one set of target static manufacturing features and target dynamic operating features that are associated with the target failure mode from the static manufacturing features and dynamic operating features.

[0046] Specifically, feature matching and filtering are performed based on a knowledge base. Guided by the mapping relationships stored in the battery failure mode library, and given a specific target failure mode (e.g., assessing "thermal runaway risk" or "capacity drop risk"), the mapping relationships are traversed to find all preset feature combinations associated with the target failure mode. For each such combination, a search is performed in the extracted pool of all static manufacturing features and dynamic operating features to match the specific feature items specified by the combination. These successfully matched features are identified as the target static manufacturing features and target dynamic operating features directly related to the assessment of the target failure mode. For example, if the mapping relationship indicates that "thermal runaway risk" is associated with the feature combination "(residual water content, high-temperature operating time)", then the static manufacturing feature named "residual water content" and the dynamic operating feature named "high-temperature operating time" will be found in the feature pool and marked as target features for assessing thermal runaway risk. This achieves precise positioning from the entire set of general features to a specific subset of features for specific failures.

[0047] Step B303: Perform logical combination operations on the target static manufacturing features and target dynamic operation features associated with the target failure mode to generate cross-fusion features.

[0048] Specifically, based on predetermined mathematical or logical rules, the static manufacturing characteristics and dynamic operational characteristics of the target are integrated and calculated. Logical combination operations refer to mathematical operations or logical judgments that reflect the coupling relationship between the two, such as directly multiplying two characteristic values, calculating their ratio, performing weighted summation, or performing AND / OR logical judgments after setting a threshold. For example: Characteristic X1: Residual water content of the material. ×High-temperature storage time ( ), used to assess the risk of gas production and expansion failure; Feature X2: Compacted density × percentage of fast charging cycles ( ), used to assess the risk of lithium dendrite precipitation; Feature X3: Difference in internal resistance consistency at the time of manufacture × Maximum temperature difference during operation ( These interactive features are used to assess the risk of accelerated degradation due to inconsistencies. Through mechanism-driven feature engineering, these features transform previously independent static and dynamic data into risk indicators that directly indicate specific failure paths. The result of this calculation is the cross-fusion feature. This feature is a newly quantified indicator whose value directly characterizes the contribution or influence of the inherent defects or characteristics represented by the target's static manufacturing features and the acquired usage stress or environmental conditions represented by the target's dynamic operating features on the risk of the target's failure mode.

[0049] In this embodiment of the application, the state of the target battery is evaluated based on cross-fusion features to obtain the evaluation result, including: Step C1: Construct at least one hierarchical state index of the target battery based on the cross-fusion characteristics.

[0050] Specifically, when calculating the status indicators for each layer, a combination of preset rules and data-driven approaches is used: health thresholds and scoring curves are set for key indicators (defining what level is considered normal and what level is considered severely degraded based on experience), while historical big data is used to determine the weights and combined mapping relationships of each indicator through statistics / machine learning. This fusion mechanism ensures that each index has both clear physical meaning and objectively reflects its impact on battery life.

[0051] In this embodiment of the application, constructing at least one hierarchical state index of the target battery based on cross-fusion characteristics includes: Step C101: Calculate the capacity index of the target battery based on the cross-fusion features, where the capacity index is used to characterize the capacity decay state of the target battery.

[0052] Specifically, based on a predefined quantification model, the indicators directly related to capacity decay in the cross-fusion features are weighted and comprehensively calculated to generate a scalar value characterizing the battery's capacity health status, namely the capacity index. This index is used to quantify the capacity degradation state of a target battery; a higher value indicates more severe degradation. The calculation process involves extracting key capacity degradation factors from cross-fusion features, including capacity retention rate (…). (Ratio of current maximum capacity to initial capacity), capacity decay rate ( (e.g., percentage decrease in capacity per month or per hundred cycles) and equivalent number of cycles ( (Considering the cumulative working intensity index of discharge depth). Subsequently, weighting coefficients are preset through pre-defined experience or data-driven methods ( , , Each element is processed through a specific function. (Used to map the original index values ​​to standardized scores) The processed results are weighted and summed, and the final capacity index is calculated according to the following formula:

[0053] in, For capacity index, The weights corresponding to the capacity retention rate. For capacity retention, The weights corresponding to the capacity decay rate, For capacity decay rate, The weights are the equivalent number of iterations. This represents the equivalent number of iterations.

[0054] This calculation process will incorporate manufacturing factors (implicit in the initial capacity). The influence of material properties on decay rate and usage factors (reflected in the number of cycles, usage habits, etc.) and The impact of capacity degradation is organically combined through feature fusion and transformed into a comprehensive score that allows for intuitive assessment of the degree of capacity degradation. The higher the index value, the more severe the capacity degradation and the worse the battery's health status in this dimension.

[0055] For example, the system specifically analyzes the coupling effect between fast charging behavior and battery internal humidity. If a battery's fast charging ratio is consistently higher than 30% (operational characteristic), and its residual water content is high or the humidity sensor reading rises abnormally during operation (a combination of manufacturing and environmental characteristics), even if the current SOH (capacity retention rate) decreases only slightly, the model will calculate a high capacity decay risk index through cross-features (such as "fast charging ratio × humidity level") and issue an early warning.

[0056] Step C102: Calculate the consistency index of the target battery based on the cross-fusion characteristics, whereby the consistency index is used to characterize the performance differences between individual cells of the target battery.

[0057] Specifically, based on a predefined quantification model, specific indicators related to consistency in the cross-fusion features are weighted and comprehensively calculated to generate a scalar value characterizing the internal performance consistency of the battery pack, namely the consistency index. This index is used to quantify the performance differences between individual cells in a target battery; a higher value indicates poorer consistency. The calculation process involves extracting key consistency elements from cross-fusion features, primarily including cell voltage dispersion. (Reflects the degree of voltage distribution dispersion of each cell), and the dispersion of internal resistance of a single cell. (Reflecting the degree of difference in internal resistance of each cell) and temperature uniformity index ( (e.g., the maximum temperature difference between modules reflects the uniformity of heat distribution). Subsequently, weighting coefficients are preset through experience or data-driven methods. , , Each element is processed through a specific function. (Used to map raw indicator values ​​to standardized scores) The processed results are then weighted and summed, i.e., according to the formula:

[0058] in, As a consistency index, The weights corresponding to the voltage dispersion of individual units. For individual unit voltage dispersion, The weights corresponding to the dispersion of the internal resistance of a single unit. The dispersion of the internal resistance of a single unit. The weights corresponding to the temperature uniformity index. This is an indicator of temperature uniformity.

[0059] This calculation process will reflect the fusion of manufacturing discreteness (such as initial internal resistance distribution) and operational imbalance (such as voltage and temperature differences generated during use), and transform it into a comprehensive score that can intuitively determine the degree of consistency degradation.

[0060] Step C103: Calculate the risk index of the target battery based on cross-fusion features, whereby the risk index is used to characterize the degree of potential failure risk of the target battery.

[0061] Specifically, based on a predetermined risk assessment model, indicators related to various specific faults and safety hazards in the cross-integration characteristics are comprehensively calculated to generate a scalar value that quantifies the potential danger level of the battery, namely the risk index. This index characterizes the potential failure risk of a target battery; a higher value indicates a greater likelihood of a safety accident or serious failure. The calculation process involves extracting and integrating multiple failure-related indicators from cross-functional features. These primarily include features directly reflecting abnormal states, such as the count of overvoltage / undervoltage events, the count of overtemperature events, the frequency of Diagnostic Trouble Codes (DTCs), and the insulation resistance level. Crucially, the calculation process deeply integrates risk factors from the manufacturing process. For example, static manufacturing features such as electrode defect markings and known defect tendencies associated with specific material formulation batches (e.g., micro-short circuit risk) are combined with the aforementioned dynamic operational features (e.g., calculating cross-features like "defect markings × recent overtemperature event frequency"). This reflects the coupling effect between the battery's inherent defects and acquired triggering conditions in the risk index. Finally, these integrated indicators are processed using weighted summaries or specific risk models (e.g., logistic regression, scoring cards) to output the risk index. This index achieves a comprehensive quantification and predictive assessment of deep-seated hidden dangers caused by the combined effects of manufacturing defects and operational stress, even before triggering a hard alarm in the BMS.

[0062] In addition, the stratification status index may also include the voltage characteristic index. (e.g., measuring voltage hysteresis, plateau changes), internal resistance growth index (Measures the increase in internal resistance at room temperature / low temperature), the efficiency degradation index Heta (reduction in charge / discharge energy efficiency), etc., are selected according to application requirements. These indices together constitute a multi-dimensional, hierarchical, and traceable health assessment system, making the assessment results more interpretable and providing business guidance value.

[0063] Step C2 involves weighted fusion of the various hierarchical state indicators to obtain the target state value.

[0064] Specifically, the calculated hierarchical status indicators (such as capacity index, consistency index, risk index, etc.) are linearly weighted and summed according to their respective importance or influence using preset weight coefficients to calculate a single comprehensive value, i.e., the target status value (e.g., a comprehensive health score of 0-100). The process includes: first, obtaining the value of each calculated hierarchical status indicator; then, assigning corresponding weight coefficients to each indicator. These weights can be set based on empirical judgments of the importance of different indicators in overall health assessment using domain knowledge, or they can be obtained through machine learning training and optimization on a large amount of historical data with end-life or health labels; finally, multiplying the values ​​of all indicators by their corresponding weights and summing the results, i.e., executing the formula:

[0065] in, For the target state value, The weights corresponding to the capacity index, For capacity index, The weights corresponding to the consistency index. As a consistency index, The weights corresponding to the risk index, This is a risk index.

[0066] By using a weighted fusion process, the sub-indicators reflecting different aspects of battery health are aggregated into an intuitive overall score, making it easy for users to quickly grasp the overall health level of the battery.

[0067] It should be noted that the health assessment and fault diagnosis in this application adopts a two-stage fusion reasoning framework of initial rule judgment and detailed model judgment. In the initial rule judgment stage, empirically based hard threshold rules are preset (e.g., temperature > 120℃, insulation resistance < 100 kΩ). In the real-time data stream, once any hard rule is triggered, the corresponding fault is immediately marked and the highest priority safety response process is initiated. In the detailed model judgment stage, for the vast majority of cases where hard rules are not triggered, a pre-trained machine learning model (e.g., XGBoost) is activated, inputting all features for comprehensive calculation, and outputting the probability distribution of various faults. The model can capture complex nonlinear relationships and weak collaborative signals between multi-dimensional features. In the result fusion and decision-making stage, the final diagnostic result is obtained by fusing the rule output and the model output according to a preset strategy. For example, if the fault output by the model with high confidence does not conflict with the rules, it is confirmed; if the two are inconsistent, a conservative principle is followed, prioritizing the judgment with higher risk. Simultaneously, the model's judgment can be explained using tools such as SHAP for review and continuous optimization of the rule base. This framework improved the recognition rate of complex edge cases and the overall fault diagnosis accuracy in testing.

[0068] Step C3: Predict the future state of the target battery based on the hierarchical state index to obtain the state prediction result.

[0069] In this embodiment of the application, the future state of the target battery is predicted based on the hierarchical state index to obtain the state prediction result, including: Step C301: Obtain historical sequence data for each hierarchical state indicator.

[0070] Specifically, for the target battery, the system retrieves and extracts the historical records of each calculated hierarchical status indicator (such as capacity index, consistency index, and risk index) from continuous historical time points in the system storage, forming a set of values ​​arranged in chronological order, i.e., historical sequence data. This historical sequence data is the foundation for time trend analysis and future prediction. This is implemented by querying a time-series database or historical record table specifically used to store health assessment results, based on the target battery's unique identifier (such as battery ID). For each hierarchical status indicator to be predicted, the calculated value of that indicator at each historical time point is obtained according to a preset time granularity (e.g., in "weeks" or "months" over the past 12 months). For example, the values ​​of the capacity index at the last assessment point of each week over the past 52 weeks are obtained, forming a time series of length 52. This ensures that the subsequent prediction model receives a structured time-series input reflecting the evolution trends of each health dimension.

[0071] Step C302: Calculate the predicted state indicators for each stratified state indicator in the future time period based on historical sequence data.

[0072] Specifically, time series analysis or sequence prediction models are used to model and infer the historical sequence data of each hierarchical state indicator, thereby calculating the estimated values ​​of these indicators at one or more time points within a specified future time period (such as the next quarter or the next 6 months). These estimated values ​​are the predicted state indicators. For each hierarchical state indicator to be predicted (such as capacity index, consistency index, etc.), its time-sorted historical sequence data is used as input and fed into a trained prediction model (e.g., a model based on Long Short-Term Memory (LSTM), Autoregressive Integrated Moving Average (ARIMA), or other machine learning algorithms). The model learns the change patterns, trends, and periodicity inherent in the historical data of the indicator and extrapolates or iterates to generate a sequence of predicted values ​​for a series of consecutive time points in the future. For example, inputting the monthly capacity index sequence of the past 24 months, the model outputs the predicted capacity index for each month of the next 6 months. These output predicted values ​​are the predicted state indicators of each indicator in the future time period, which constitute a quantitative prediction of the future evolution trend of various health dimensions of the battery.

[0073] Step C303: Calculate the remaining effective time and replacement reminder time of the target battery based on the predicted state indicators, and use the remaining effective time and replacement reminder time as the state prediction results.

[0074] Specifically, based on predefined failure or safety thresholds, trend analysis and threshold judgment are performed on the predicted state indicator sequences of each layered state indicator over future time periods to deduce two key time points as prediction conclusions. First, the remaining effective time is calculated by analyzing the changing trends of key predicted state indicators (usually with capacity index as the core, as it directly relates to energy storage capacity) over time. The future time point corresponding to the first time that the indicator reaches or exceeds the preset failure threshold (e.g., capacity index corresponds to capacity decay to 70% of initial capacity) is identified. The time from the current moment to that time point is the remaining effective time. Second, the calculation of the replacement warning time is more forward-looking. Typically based on safety or economic considerations, the predicted trends of risk index or consistency index are analyzed to determine the time point when they reach the warning threshold requiring intervention, or a safety buffer period is reserved in conjunction with the remaining effective time (e.g., 3 months in advance), thereby calculating the recommended replacement warning time for maintenance or replacement. Ultimately, these two calculated time parameters (such as "remaining effective time: 18 months" and "replacement reminder time: within 15 months") together constitute a quantitative prediction of the battery's future availability and are encapsulated as part of the state prediction result.

[0075] Step C4: Use the hierarchical state index, target state value, and state prediction result as the evaluation result of the target battery.

[0076] Specifically, the data set is formed by combining three types of information: multiple hierarchical status indicators (such as capacity index, consistency index, and risk index, each with its own independent score), a single comprehensive target status value obtained through weighted fusion (i.e., comprehensive health score), and a predicted status result including remaining effective duration and replacement reminder time. This result may be organized as a JSON object, database record, or a report in a specific format, containing multi-dimensional sub-indicators reflecting the current status, a comprehensive score reflecting the overall health level, and a time prediction reflecting future evolution trends. The completion of the evaluation process generates a standardized and interpretable battery health assessment conclusion that can be directly displayed on the user interface, provided to the API interface, or directly invoked by subsequent business decision-making logic.

[0077] To enhance the interpretability and decision support value of assessment results, model interpretation tools such as Shapley Additive exPlanations (SHAP) can be integrated. When assessment results indicate battery consistency degradation, SHAP analysis is invoked to automatically identify and report the most contributing characteristic factors. For example, for consistency degradation warnings, SHAP displays "cell voltage difference (…)" "Voltage and internal resistance dispersion" are the main contributors, leading to the report: "Consistency degradation mainly stems from significant deviations in voltage and internal resistance of individual cells from the group." Furthermore, if "production batch" characteristics are found to contribute significantly, the report will further indicate: "This battery belongs to Batch 7, and there may be batch-specific process differences; close monitoring is recommended." This mechanism transforms complex model judgments into easily understandable causal descriptions, improving operational efficiency and accuracy.

[0078] In this embodiment of the application, after evaluating the state of the target battery based on cross-fusion features and obtaining the evaluation results, as follows: Figure 2 As shown, the method also includes: Step S201: Obtain the business object tag associated with the target battery.

[0079] In this embodiment, the associated business role or system type identifier is determined and extracted based on the specific application scenario served by the target battery. These tags are used to distinguish different business application areas, such as "network risk control," "intelligent assessment," and "fleet operation and maintenance," with each tag corresponding to a specific business requirement and early warning strategy. The business type is parsed from the registration information or usage contract of the vehicle to which the battery belongs; tags are automatically matched from the configuration file of the battery deployment scenario (such as intelligent monitoring vehicles, intelligent verification vehicles, and intelligent management vehicles); or obtained through user manual specification, cloud database queries, etc. In addition, tags can be dynamically assigned according to battery data usage permissions or access platform type to ensure that the subsequently generated early warning information can adapt to the decision-making needs of different business objects.

[0080] As an example, the battery-related business object tags for new energy vehicles used in ride-hailing operations might include: fleet operation and maintenance (for vehicle dispatching and maintenance decisions), intelligent assessment (for risk quantification and assessment strategy analysis), and intelligent risk control (for battery status monitoring and risk warning management). After obtaining these tags, targeted warning content can be generated for each tag, such as sending "Battery consistency has decreased, balanced maintenance is recommended" to the fleet operation and maintenance end, and sending "Thermal runaway risk score has increased, assessment conditions are recommended" to the intelligent assessment platform, etc.

[0081] Step S202: For each business object tag, generate early warning information corresponding to the evaluation result.

[0082] In this embodiment, based on business object tags (such as intelligent risk control, intelligent assessment, fleet operation and maintenance, etc.), the assessment results output by the battery health assessment model are transformed into customized warnings or prompts that meet the needs of the business scenario. First, warning templates and rule bases associated with each business object tag are predefined or dynamically loaded. These templates and rules are designed based on business logic. For example, intelligent risk control focuses on battery status risk and operational reliability, intelligent assessment focuses on the probability of risk occurrence and quantitative assessment results, and fleet operation and maintenance focuses on operation and maintenance costs and equipment utilization efficiency. Second, based on the specific indicator values ​​in the assessment results (such as health score exceeding the threshold, risk index rising, detection of specific fault symptoms), key data (such as score, trend, prediction time) are filled in by the template, and rules are applied to generate text descriptions, risk levels (such as high, medium, low), and recommended measures. Finally, the warning information is formatted to ensure that it is suitable for subsequent output (such as JSON messages, visual report fragments). In addition, it can support dynamic adjustment of warning content. For example, when the same assessment result involves multiple business tags, information with different focuses can be generated respectively: for intelligent risk control, the impact of remaining life and operational reliability is highlighted; for intelligent assessment, the risk of thermal runaway and the frequency of historical failures are emphasized; for fleet operation and maintenance, specific maintenance suggestions and abnormal battery location are provided.

[0083] As an example, if the assessment results detect a significant decrease in the target battery's capacity index and an increase in its risk index, a warning message is generated for the fleet maintenance tag: "Battery health score drops to 65 (warning level: orange), capacity retention rate has dropped to 78%, accompanied by early lithium dendrite risk. Recommendation: Reduce fast charging frequency, schedule equalization maintenance, and focus on checking voltage consistency in the next monthly inspection." For the intelligent assessment tag, the message is: "The battery's thermal runaway risk score rises to 0.7 (high risk range), and the number of recent over-temperature events has increased. Recommendation: Review the assessment threshold, optimize the risk quantification model parameters, or add specific testing requirements for the battery thermal management system." In this way, the warning information can be accurately adapted to the decision-making processes of various business entities.

[0084] Step S203: Output the evaluation results and early warning information to the business object corresponding to the business object label.

[0085] In this embodiment, the output of early warning information is implemented based on the application interface, supporting real-time collection of vehicle BMS data through a plug-and-play OBD battery detection box and uploading it to the cloud via 4G / 5G. The cloud platform provides RESTful API services, offering standardized query interfaces to external systems (i.e., management platforms of different dimensions), returning JSON data such as health scores, SOH%, and risk events, supporting embedded applications in multiple industry scenarios. The specific process includes: First, maintaining an output routing table to map each business object tag (such as intelligent risk control, intelligent assessment) to a specific output terminal or interface, such as the API of the intelligent risk control platform, the assessment terminal of the intelligent assessment system, the visualization interface of the fleet management system, or the mobile application of maintenance personnel. Second, calling the corresponding interface protocol (such as RESTful API, WebSocket, or message queue) and encapsulation format (such as a JSON data packet containing health scores, SOH%, and a list of risk events, or a directly displayed HTML / PDF report) to push or respond to queries for the generated early warning information and associated detailed assessment results. In addition, the output process integrates authorization verification (such as APIKey / OAuth) and data anonymization mechanisms to ensure that business objects can only receive information within their authorized scope. For example, for fleet maintenance tags, early warning cards can be pushed in real time through a dedicated module integrated into the fleet management platform; for intelligent risk control tags, JSON data can be synchronized to the status assessment module of the intelligent risk control system via an encrypted API. Furthermore, this system also features a closed-loop learning and continuous optimization mechanism: whenever an actual fault occurs or a battery is retired, the battery's entire lifecycle data (genes + operation + fault results) is fed back to the fault database for model retraining and rule updates, continuously improving fault prediction accuracy and adaptability to new types of batteries.

[0086] As an example, consider a new energy vehicle belonging to a smart device management entity, whose business object tag includes smart asset management. When the assessment detects a deterioration in the vehicle's battery consistency index, an alert is generated: "The voltage range of individual cells within the battery pack remains above 0.1V, and the consistency health index has dropped to 40. Equalization maintenance is recommended." This information, along with detailed historical voltage dispersion curves and the specific affected module numbers, is automatically pushed to the smart asset management platform monitoring screen of the smart device management entity via a pre-integrated data interface. Simultaneously, a text message containing the key findings is sent to the corresponding maintenance equipment. The management entity can then promptly arrange maintenance based on this information, avoiding vehicle downtime losses due to battery issues.

[0087] To effectively communicate assessment results, a visual user interface (UI) tailored to different roles can be provided. Key UI elements include: a comprehensive health dashboard: visually displaying the battery's overall health index from 0-100 points, with color coding (e.g., green / yellow / orange / red) to differentiate health levels; sub-indicator trend charts: displaying historical change curves for key sub-indicators such as State of Health (SOH), internal resistance growth, and cell differential pressure in chart form, providing their percentile ranking within the same battery group (e.g., "Your battery's State of Health is better than 80% of similar vehicles"); a warning information center: prominently listing currently active fault risk warnings (e.g., "Thermal runaway risk: High"), each accompanied by a brief explanation of the cause and suggested measures (e.g., "Please immediately inspect the cooling system"); and in-depth diagnostic reports: users can click to view detailed reports, which not only include data for various indicators but also explain the main influencing factors of the current health score or specific warnings in a visual manner (e.g., feature contribution bar charts). For professional maintenance personnel, an expert mode can be switched to view more raw sensor data streams and model interpretation details. This design ensures that everyone from ordinary car owners to professional maintenance personnel can quickly understand the battery status and take appropriate action.

[0088] In another embodiment, the battery state assessment method of this application can be specifically applied to the early identification and warning of thermal runaway risk in power batteries. This embodiment provides a thermal runaway risk identification method based on cross-feature fusion and rule-model fusion decision-making. It not only collects multi-source dynamic features during battery operation (such as maximum temperature, insulation resistance, voltage fluctuations, overheating history, etc.), but also constructs explicit cross-features to characterize the multi-factor coupling mechanism. Furthermore, it employs a fusion decision-making mechanism combining a rule engine and a machine learning model to achieve hierarchical identification and accurate warning of thermal runaway precursors. The specific process is as follows: First, operational data, including at least the maximum temperature (max_temperature_C), insulation resistance (insulation_resistance_kohm), individual cell voltage fluctuation (delta_voltage_V), and overheat count, are collected from the vehicle BMS or monitoring terminal. Based on this, statistical characteristics (such as temperature standard deviation, temperature rise rate proxy, etc.) are calculated to form a basic feature vector for risk identification.

[0089] Secondly, to express the multi-factor coupling effect of thermal runaway, various cross-features are constructed on top of the basic features. For example, the temperature × insulation cross-feature: temp_x_inv_ins = max_temperature_C × (1 / insulation_resistance_kohm), is used to amplify the risk amplification effect of "high temperature superimposed on insulation degradation". The temperature × voltage cross-feature: temp_x_deltaV = max_temperature_C × delta_voltage_V, is used to characterize the simultaneous occurrence of thermal and electrical anomalies. Other coupling features include temperature rise rate × voltage, temperature rise rate × insulation, and overheating history × voltage. These cross-features can explicitly characterize the interaction between "long-term stress and short-term anomalies," improving the model's ability to identify complex risk patterns.

[0090] like Figure 3The figure shown is a feature interaction and correlation analysis diagram. This diagram comprehensively displays four parts: the feature correlation matrix, the cross-feature scatter plot, the temperature × voltage risk matrix, and the rule weight allocation, used to verify the effectiveness of the cross-feature construction and rule design in this embodiment. The key feature correlation matrix shows that the correlation coefficient between max_temperature_C and rule_score is 0.59, the correlation coefficient between overheat_count and rule_score is 0.32, and the correlation coefficient between insulation_resistance_kohm and rule_score is -0.13; the correlation coefficient between rule_score and model_proba is 0.08, and the correlation coefficient between max_temperature_C and model_proba is 0.06. This result indicates that the rule score is sensitive to temperature and overheating history, and the model output is not linearly determined by a single temperature, but relies more on the comprehensive pattern recognition of multiple features and cross-features. The cross-feature scatter plot shows that, with temp_x_inv_ins on the horizontal axis and model_proba on the vertical axis, high-risk / extremely high-risk samples clearly cluster in the higher-risk probability region, verifying the effectiveness of the "temperature × insulation coupling feature" in distinguishing high-risk samples. The temperature × voltage risk matrix shows high-risk hotspots at the combination of high temperature and large voltage fluctuations, indicating that the superposition of thermal and electrical anomalies can significantly increase the risk level, supporting the multi-feature cross-validation strategy. The figure clearly shows the weight allocation of each component of the rule engine (temperature 0.35, insulation 0.30, voltage 0.20, overheating history 0.15), reflecting the engineering logic of "temperature as the primary factor, insulation as the key safety boundary, and voltage and historical anomalies to enhance robustness." This figure, from multiple perspectives including statistical correlation, feature interaction visualization, and the rationality of rule design, jointly verifies the effectiveness of cross-feature fusion and rule engine design.

[0091] Four sub-items are set: temperature, insulation, voltage, and overheating history. Each sub-item is mapped to a sub-score based on a threshold, and then weighted and fused (e.g., temperature 0.35, insulation 0.30, voltage 0.20, overheating history 0.15). Therefore, the rule score is calculated as: rule_score = 0.35×score_T + 0.30×score_I + 0.20×score_V + 0.15×score_H, where score_T, score_I, score_V, and score_H are the normalized sub-scores for temperature, insulation, voltage, and overheating history, respectively. The rule engine serves as a fallback for engineering implementation, directly triggering an alert when high-risk combinations of conditions (such as high temperature and extremely low insulation) occur.

[0092] like Figure 4The figure shows a multi-dimensional risk space visualization. Through the three-dimensional risk space distribution, temperature × insulation heatmap, and rule-model joint distribution, it demonstrates that relying solely on the temperature dimension is insufficient to adequately distinguish risks; insulation and voltage fluctuations must be introduced for collaborative assessment. The temperature × insulation heatmap shows that the risk distribution is non-linear and interactive, meaning that risk does not simply correspond monotonically to temperature or insulation, but exhibits significant combined effects. The joint distribution of rule scores and model probabilities provides a partitioning diagram of medium and high risk thresholds, showing a consistent clustering trend in high-risk areas and complementarity at the boundary regions. This provides an intuitive basis for subsequent fusion decision-making strategies.

[0093] Using basic and cross-features as input, interpretable classification models such as random forest and gradient boosting tree are employed to output the thermal runaway risk probability `model_proba`. The model is trained through supervised learning and employs a class imbalance handling strategy to adapt to the scarcity of thermal runaway samples. The following fusion strategy is adopted: direct warning when rule scores reach the high-risk threshold; risk assessment based on model probability when rules are not triggered; and increased warning level when both rules and the model indicate medium to high risk. Tiered thresholds can be set, for example: medium risk (rule_score≈0.50, model_proba≈0.35), high risk (rule_score≈0.70, model_proba≈0.60), and the actual thresholds can be calibrated according to vehicle model, battery system, etc.

[0094] like Figure 5 The figure shows a performance comparison of the early warning methods. It compares the performance of the machine learning model, rule engine, and fusion strategy. The ROC curves show that the model's AUC is 0.911, the rule engine's AUC is 0.639, and the fusion strategy's AUC is 0.875. The performance comparison indicates that the rule engine, while pursuing a high recall rate (0.85), suffers from extremely high false positives (precision rate of only 0.01), making it difficult to apply in engineering alone. The model, while achieving a high precision rate (1.00), requires rules as a fallback to cover highly dangerous situations. The fusion strategy, while maintaining a high precision rate (0.909) and a stable recall rate (0.769), keeps the false positive rate at an extremely low level (0.1%), demonstrating its value for engineering deployment. Detailed statistics for the fusion strategy are: TP=10, FP=1, FN=3, TN=986, false negative rate 23.1%, accuracy 99.6%.

[0095] like Figure 6The figure shows the accuracy analysis of the early warning system. This figure further validates the accuracy of the method through PR curves, threshold analysis, and the optimal threshold confusion matrix. The average precision (AP) of the PR curve is 0.781, indicating good early warning capability even under imbalanced sample conditions. Threshold performance analysis determined the optimal threshold to be 0.061. Under this optimal threshold, the model's confusion matrix is ​​TN=987, FP=0, FN=3, TP=10, corresponding to Precision=1.000, Recall=0.769, and F1=0.870. The fusion strategy, while maintaining the same recall (0.769), has a corresponding Precision=0.909, F1=0.833, and a confusion matrix of TN=986, FP=1, FN=3, TP=10. This result demonstrates that the fusion strategy, after introducing rule fallback, only slightly increases the false positive rate by one, while effectively covering real-risk samples.

[0096] like Figure 7 The figure shows the feature importance analysis (Top 20 of Random Forest). This figure displays the ranking of feature importance in the Random Forest model, with cross-features ranking highly, such as rate_x_deltaV (importance 0.0248), overheat_x_deltaV (0.0205), and rate_x_inv_ins (0.0129). Meanwhile, manufacturing / batch-related features such as binder_content_pct (0.0149) also appear in the list of important features. This directly proves that the model's decision-making does not rely on a single temperature threshold, but rather comprehensively utilizes multiple factors such as temperature rise rate, voltage fluctuations, insulation level, historical anomalies, and manufacturing parameters, and explicitly characterizes the coupling relationships between them through cross-features, thereby improving the ability to identify thermal runaway risks and the interpretability of the model.

[0097] Finally, in the actual test scenario, when the battery temperature was detected to be about 130°C and the insulation resistance was below 50kΩ: the cross feature temp_x_inv_ins increased significantly; the rule score increased rapidly and triggered a high-risk fallback; the model output probability simultaneously entered the high-risk range; the fusion strategy output a high-risk warning and automatically executed safety strategies such as emergency cooling, power limiting, and guided shutdown, thereby achieving timely intervention for the precursors of thermal runaway.

[0098] The method has been tested and verified, achieving the following performance on the test dataset: model AUC=0.911, fusion strategy AUC=0.875, and rule engine AUC=0.639. The fusion strategy achieves a precision of 0.909, a recall of 0.769, and a false positive rate of only 0.1%, demonstrating strong engineering usability and safety early warning value. This embodiment illustrates that by employing a technical approach of cross-feature fusion + rule engine fallback + machine learning model fine-tuning + fusion hierarchical decision-making, multi-feature cross-validation and interpretable early warning of thermal runaway risk can be achieved, demonstrating the effectiveness and flexibility of the proposed method in specific safety management scenarios.

[0099] In another embodiment, the battery state assessment method of this application can also be specifically applied to the early identification and warning of the risk of accelerated capacity degradation in power batteries. This embodiment addresses the risks of "capacity plunge" or "accelerated capacity degradation" that may occur during battery use by providing a cross-coupling modeling method that combines fast charging behavior with moisture residue (humidity) indicators to achieve proactive identification of capacity degradation risks. The core of this method lies in: not only collecting battery health state indicators (such as SOH), but also focusing on extracting two types of features: "fast charging percentage" (reflecting user behavioral stress) and "humidity / residual moisture level" (reflecting sealing or residual moisture risk), and constructing the interaction features between the two. Thus, even when the battery SOH is still within the normal range (e.g., ≥85%), the potential risk of a precipitous capacity drop can be identified, providing users and maintenance personnel with a sufficient intervention window. The specific process is as follows: First, the following indicators are collected and constructed as inputs for capacity degradation risk identification: Fast Charge Ratio (F): Represents the proportion of fast charging behavior to total charging behavior within a given statistical window (e.g., the last 30 days, the last 90 days), which can be calculated by number of times, energy, or time. An example threshold is set at 30%. Humidity / Residual Moisture Index (Humidity, H, unit ppm): Sourced from humidity sensors inside the battery pack, coolant water content detection, and residual moisture records from the manufacturing process, used to characterize the risk level related to moisture. An example threshold is set at 200 ppm. State of Health (SOH) (Capacity Retention Rate, S): Used as background health information and result verification indicator, not the sole criterion. Cross-coupling feature (Fast Charge × Humidity Interaction Term, X_int): To characterize the nonlinear amplification effect under the combined action of fast charging and humidity, a formula such as "X_int=F×H" or "X_int=max(0,F)" is constructed. F_th)×max(0,H Interactive features such as "H_th") explicitly express the interaction pattern that the risk of capacity degradation increases significantly when "fast charging is high and humidity is high".

[0100] like Figure 8 The figure shows a heatmap (risk score distribution) of the interaction effect of fast charging and humidity. The graph uses the proportion of fast charging (F) (0-80%) as the horizontal axis and humidity (H) (0-400ppm) as the vertical axis, visually illustrating the distribution of the risk score R through a color change from green to red. The fast charging threshold (30%) and humidity threshold (200ppm) are marked with blue dashed lines, and the high-risk area (F>30% and H>200ppm) is indicated in the upper right corner with a semi-transparent fill. The results show that when only one factor is abnormal (F<30% or H<200ppm), the risk score increases only slightly; however, when both fast charging and humidity exceed the thresholds and enter the high-risk area, the risk score increases significantly and non-linearly. Figure 1 The system also provides a contour line (red dashed line) with a risk score R=0.7 to delineate higher-level alarm areas, and marks the case study point (F=40%, H=220ppm) with a black star. This point is located in the high-risk area, which verifies that even when SOH does not decrease significantly, the system can effectively identify the risk situation through cross features.

[0101] Secondly, a risk score R (range 0~1) is output by fusing a rule engine and a data-driven model, with higher values ​​indicating a higher risk of capacity degradation. A basic risk score is formed based on whether the fast charging percentage F and humidity H exceed thresholds (e.g., F_th=30%, H_th=200ppm). F, H, SOH, and cross-feature X_int are input into a trained scoring model (e.g., logistic regression, random forest), and the model risk score is output. R=max(R_rule, R_model) or a weighted fusion method can be used to ensure a "rule-based safety net" for clearly high-risk combinations, while simultaneously leveraging the model to improve the ability to identify boundary samples.

[0102] like Figure 9 The figure shown is a case study: a time series graph of SOH trajectory and risk score. This graph visually verifies the early warning capability of this embodiment through time series analysis. The upper graph shows the change of SOH over time, with its value only experiencing a sharp drop to 93% in month 8. The lower graph simultaneously shows the changes in risk score R and humidity H. The risk score had already risen to approximately 0.60 in month 0 (when SOH was still 100%) due to the excessive proportion of fast charging (40%) and humidity (220ppm), and has remained at a high level ever since. This proves that the system identified the risk and provided continuous warnings at least 6 months before a significant decline in SOH occurred, achieving proactive diagnosis.

[0103] like Figure 10The diagram shows the threshold boundary visualization and risk area division. The diagram clearly marks the two threshold boundaries, F=30% and H=200ppm, with blue dashed lines. Semi-transparent red and green areas visually represent the high-risk area (F>30% and H>200ppm) and the low-risk reference area (e.g., F<15% and H<50ppm), respectively. This visualization makes the physically-based area division rules of this embodiment readily apparent, greatly enhancing the interpretability of the solution and facilitating direct application and configuration in different business scenarios.

[0104] The following example thresholds are set: Medium-high risk threshold: R_medium=0.4; High-risk alarm threshold: R_alarm=0.7. The risk score R is used to grade the output: R<0.4: Low risk, capacity degradation is within the normal aging range; 0.4≤R<0.7: Medium-high risk, indicating accelerated capacity degradation risk, requiring attention and intervention; R≥0.7: High risk, triggering a strong alarm and recommending maintenance, usage restriction, or other measures. To enhance interpretability, physically meaningful regional rules can be defined, for example: High-risk region: F>30% and H>200ppm; Low-risk region: F<15% and H<50ppm. Through the dual output of "region identification + score grading," both the causes of risk and the intensity of risk can be explained.

[0105] like Figure 11 The figure shows a quantitative curve of the interaction effect. This figure, through two sets of curves, quantitatively demonstrates the significant interaction effect between fast charging and humidity. The left figure shows that the impact of humidity on the risk score varies at different fast charging percentages; the higher the fast charging percentage, the more significant the risk increase due to increased humidity. The right figure shows that the impact of fast charging percentage also differs at different humidity levels; the higher the humidity, the stronger the amplification effect of fast charging on risk. Both sets of curves together indicate that fast charging and humidity are not simply linearly superimposed, but rather exhibit a coupling pattern of mutually amplifying risks, providing direct evidence for the necessity of constructing cross-features.

[0106] like Figure 12 The figure shows a multi-scenario comparison and verification diagram. It sets up four representative scenarios (e.g., low risk: fast charging 10% / humidity 50ppm; case scenario: fast charging 40% / humidity 220ppm; extremely high risk: fast charging 60% / humidity 300ppm) and displays their positions in the fast charging-humidity space and their corresponding risk scores. The results show that, under the condition that all SOH is 98%, risk cannot be distinguished solely by SOH. However, the risk scores calculated by this method using cross-features exhibit a significant gradient difference (from 0.014 to 0.946). The risk score of the case scenario (0.598) has entered the medium-to-high risk range, verifying the ability of this method to distinguish different risk situations when SOH is normal.

[0107] like Figure 13 The figure shows a statistical comparison chart of risk levels and characteristic distributions. This chart uses a box plot to display the statistical distribution of fast charging percentage, humidity, SOH, and risk score R under different risk levels, and indicates the thresholds for each. Statistically, it verifies that: the baseline distributions of fast charging and humidity are mostly below their respective thresholds (30%, 200ppm), indicating that the threshold settings are reasonable; the SOH distribution shows that most samples are within the normal range, but the combination of fast charging and humidity can still identify medium-to-high risk; the distribution of the risk score R, combined with the tiered thresholds (0.4, 0.7), effectively achieves multi-level hierarchical output of "normal - attention - alarm".

[0108] Finally, the effectiveness of this embodiment is verified through six types of charts, including the aforementioned interaction effect heatmap, case time series, threshold boundary visualization, interaction effect quantification curve, multi-scenario comparison, and statistical distribution chart. For example, in the case study, when a battery had a SOH of 100%, a humidity of 220ppm, and a fast charging rate of 40%, the system calculated a risk score of approximately 0.60, identifying the risk of accelerated capacity degradation more than 6 months in advance, and continuously issuing warnings before the subsequent sharp drop in SOH, demonstrating significant forward-looking diagnostic and early warning capabilities. Differentiated prompts and handling suggestions are output based on the risk level: Medium-high risk (0.4≤R<0.7): Users are advised to reduce the frequency of fast charging and check the battery's sealing and waterproof structure; High risk (R≥0.7): A high-level alarm is triggered, fast charging and high-load conditions are restricted, and on-site inspection or repair is recommended. This embodiment achieves the identification of potential risks when "a single indicator is still normal, but the combined risk is significant" through cross-modeling of fast charging and humidity; it provides interpretable thresholds and regional rules, making it easy to implement in scenarios such as fleet management and financial risk control; it can provide early warnings before SOH has decreased significantly, giving users an intervention window of at least several months; it achieves refined management from "attention" to "alarm" through risk score stratification, avoiding delayed or excessive alarms; and it reduces reliance on the single indicator of SOH, improving the early perception and proactive management capabilities of capacity degradation risks.

[0109] In another embodiment, the battery state assessment method of this application can be further applied to the diagnosis and causal analysis of potential performance inconsistencies (consistency degradation) among individual power battery cells. This embodiment addresses the issues of cell parameter deviations and decreased consistency that occur during long-term use of battery packs, providing a technical solution that integrates an early warning model with SHAP (Shapley Additive Explanations) interpretability analysis. The core of this method lies in not only constructing an early warning model based on multi-dimensional consistency characteristics (such as voltage difference, internal resistance growth rate, temperature distribution, etc.) to identify risks, but more importantly, introducing the SHAP model interpretation method to analyze the feature-by-feature contribution of the early warning results for individual batteries. This clearly reveals the main factors leading to consistency degradation and their degree of influence, achieving a leap from "black-box early warning" to "white-box diagnosis," providing maintenance personnel with quantifiable and traceable decision-making basis. The specific process is as follows: First, the following multi-dimensional features are collected and constructed as inputs to the consistency degradation early warning model: Electrical performance features: including cell voltage difference (Delta_V) (the difference between the maximum and minimum cell voltages), voltage standard deviation (cell_voltage_std_dev) (the dispersion of voltage across all cells), and internal resistance growth rate (ir_increase_pct) (the percentage increase in cell internal resistance relative to its initial value). Thermal management and control features: including temperature standard deviation (temperature_std_dev) (the dispersion of temperature distribution between cells) and balancing count (the number of times the BMS performs cell balancing operations within a certain period). These features collectively characterize the battery pack's state from multiple perspectives, including electrical consistency, thermal consistency, and BMS intervention frequency. Simultaneously, a reference threshold based on historical data statistics (e.g., the 90th percentile) is preset for each feature for preliminary anomaly judgment at the rule level.

[0110] Secondly, using the aforementioned features as input, machine learning algorithms such as gradient boosting decision trees and random forests are employed to train a consistency degradation early warning model. The model output is a risk probability value P (ranging from 0 to 1), with higher values ​​indicating a greater likelihood of consistency degradation in the battery pack. This model can keenly detect subtle consistency anomalies and provide early warnings even when overall battery health indicators (such as state of equilibrium (SOH)) are still within the normal range.

[0111] After the model provides a warning probability for a single battery pack, the SHAP interpretation method is invoked to perform feature attribution analysis on this prediction. SHAP quantifies the direction and magnitude of each feature's influence by calculating its marginal contribution to the model output: a positive SHAP value indicates that the feature increases the warning probability and is a risk driver; a negative SHAP value indicates that the feature decreases the warning probability and is a risk mitigation factor. Through analysis, key features that lead to the battery pack being classified as high-risk (such as "individual cell voltage difference" contributing the most) can be clearly identified, and their contribution values ​​are compared with preset thresholds to achieve dual verification of rule anomalies and model intelligence.

[0112] like Figure 14 The chart shown is a Top Contribution Feature Chart. Presented as a horizontal bar chart, it lists the top 12 features with the greatest impact on consistency degradation warnings, arranged in descending order of SHAP contribution absolute value, along with their specific contribution values ​​and actual values. The "Individual Cell Voltage Difference (Delta_V)" has the highest SHAP contribution value (+1.8499), indicating it is the primary factor leading to the risk. "Internal Resistance Growth Rate" (+1.6573) and "Voltage Standard Deviation" (+0.8234) also have high positive contributions, indicating the combined effect of multiple anomalies. Meanwhile, some features, such as temperature standard deviation and equilibration frequency, show negative contributions (blue bars), meaning these factors mitigate the risk to some extent. This chart allows maintenance personnel to quickly focus on the most critical risk-driving features, providing a clear direction for subsequent targeted investigations.

[0113] like Figure 15 The chart shown is a comparison of key features. This chart, combining left and right sub-charts, comprehensively presents the rule anomalies and model impact of key features. The left bar chart visually compares the actual feature values ​​of the current vehicle (blue bars) with preset thresholds (orange bars, such as the 90th percentile Q90 of a normal fleet). For example, the actual value of "individual voltage difference" (1.3350 V) is significantly higher than the threshold (0.8500 V), indicating that it is an anomaly at the rule level. The right bar chart shows the SHAP contribution value and direction of the corresponding feature (red for positive, blue for negative). It can be seen that "individual voltage difference" also shows a prominent contribution value (+1.8499). This chart achieves dual verification and combined diagnosis of rule criteria (whether the value exceeds the threshold) and model intelligence (contribution magnitude and direction), enhancing the credibility of the conclusions.

[0114] To present the diagnostic results intuitively, the system generates a series of visual charts, forming a complete chain of evidence: like Figure 16The image shows a waterfall plot of SHAP contributions. This waterfall plot visually and transparently illustrates the complete decision-making accumulation process of the model from the baseline value to the final predicted output. The starting point (gray dashed line) represents the model's baseline value (approximately -2.5, corresponding to a low-risk tendency), and the ending point (red solid line) represents the model's final predicted value (0.769, corresponding to a high-risk probability). The red and blue bars in the middle represent the cumulative SHAP contribution effect of each feature. The illustrated example clearly shows that although the model initially tends towards low risk, the final output is significantly pushed up to a high-risk level due to the strong positive cumulative contributions of multiple features such as "single-cell voltage difference" (+1.8499) and "internal resistance growth rate" (+1.6573). This visualization fully unfolds the decision-making logic chain within the model, greatly improving the interpretability and transparency of the prediction process.

[0115] like Figure 17 The diagram shown is a contribution decomposition chart. This chart, through four sub-charts, decomposes and grasps the overall distribution of risk-driving and inhibiting factors. The pie charts in the upper left and upper right show the comparison of the number of positive and negative contribution features and their absolute contribution percentages, respectively. The case study shows that the number of positive contribution features (8) and their total contribution strength (approximately 75%) are both dominant, indicating that the risk is mainly driven by factors that increase the probability of early warnings. The bar charts in the lower left and lower right list the specific positive and negative contribution features and their ranking. Through this chart, maintenance personnel can comprehensively understand the "driving forces" (such as voltage differences and increased internal resistance) leading to consistency degradation and the "resistance" sources (such as good temperature uniformity) that mitigate it, helping to formulate a comprehensive maintenance strategy that balances both short-term and long-term solutions.

[0116] like Figure 18 The figure shown is a feature distribution comparison chart. This chart uses multiple parallel sub-charts to quantitatively compare the key feature values ​​of the current vehicle with the median values ​​(representing typical normal levels) of a normal vehicle fleet. Each sub-chart clearly shows the normal reference value (gray bars) and the actual value of the current vehicle (colored bars, red representing positive contributions), and labels the actual value, SHAP contribution value, and deviation factor relative to the normal value. For example, the actual value of "individual voltage difference" (1.33 V) is approximately 2.05 times the normal value (0.65 V), and the SHAP contribution is +1.8499, quantitatively confirming its abnormality and risk impact. This chart effectively connects the model diagnostic results with actual engineering experience (normal levels), providing accurate and objective quantitative evidence for risk assessment.

[0117] Finally, based on the above analysis, the system outputs a structured diagnostic report, which not only includes the risk level of consistency degradation but also clearly identifies: the main causal characteristics (e.g., "significantly excessive single-cell voltage difference, contributing the most"); the quantification of the degree of abnormality (e.g., "voltage difference is 2.05 times the normal value"); and the analysis of related factors (e.g., "synchronous abnormal internal resistance growth rate, exacerbating the risk"). Based on this, targeted suggestions are provided to maintenance personnel, such as: "Focus on inspecting and considering replacing single cells with abnormally low voltage," "Check the thermal management system to maintain temperature uniformity," and "Optimize the BMS balancing strategy." In fleet maintenance, this method can issue early warnings at least one maintenance cycle in advance before the overall performance of the battery pack significantly deteriorates, and indicates specific troubleshooting directions, thereby preventing the fault from worsening and reducing safety risks.

[0118] This embodiment combines model-based early warning with SHAP interpretation to clearly reveal "which features cause consistency problems" and "the extent of their respective impacts," solving the problem of opaque fault causes in traditional methods. Through threshold comparison and distribution comparison, the degree of feature anomalies is precisely quantified, making diagnostic conclusions objective and evidence-based. Visualization tools such as waterfall charts fully display the logical chain of model decision-making, making "black box" predictions understandable and verifiable. Even when macroscopic indicators such as SOH do not significantly deteriorate, early warnings can be issued based on subtle anomalies in multidimensional features, providing sufficient time for targeted maintenance such as replacing abnormal cells and adjusting balancing strategies, effectively preventing serious accidents that may be caused by consistency failures.

[0119] In another embodiment, the battery state assessment method of this application can also be specifically applied to construct a fault reasoning path that combines rules and models to achieve high-precision, interpretable diagnosis and early warning of various fault modes of power batteries. This embodiment addresses the shortcomings of traditional methods that rely solely on threshold rules (prone to false alarms and missed alarms) or purely data-driven models (lacking interpretability and reliability under extreme conditions) by proposing a two-stage reasoning fusion framework. The core of this framework is: first, using preset expert rules to quickly pre-judge obvious anomalies and provide a safety margin; then, for complex situations not covered by the rules or with uncertainties, calling a machine learning model for detailed judgment; and finally, combining the results of both based on a conservative fusion strategy to form a final diagnostic conclusion. This method can effectively identify high-risk faults such as thermal runaway, leakage, and overcharging, as well as hidden fault modes such as F02 and F03, significantly improving the accuracy, timeliness, and system robustness of fault diagnosis. The specific process is as follows: First, the system pre-defines a series of expert rules covering key monitoring indicators (such as temperature, voltage, current, insulation resistance, cell consistency, and mechanical shock), categorizing them into hard thresholds (corresponding to high-severity faults) and soft thresholds (corresponding to medium-severity anomalies). When real-time data triggers a hard threshold rule (e.g., temperature exceeding safety limits), the rule engine immediately identifies the corresponding fault and triggers the highest-priority safety response, achieving rapid interception and fallback. When data only triggers a soft threshold rule (e.g., cell consistency deviations but not exceeding hard limits), the system marks it as an early warning and proceeds to the subsequent model-based detailed analysis process. This stage can quickly process more than one-third of the samples, demonstrating high sensitivity to clear fault signals and real-time response capabilities.

[0120] like Figure 19 The chart shown is a rule trigger frequency analysis graph. This graph displays the number of times the seven preset expert rules were triggered in the test sample. Among them, the "poor cell consistency" rule was triggered most frequently (131 times), followed by the "overcharge risk" rule (83 times), verifying that the rule configuration can effectively cover common fault types. The triggering of high-severity rules (such as thermal runaway and leakage) ensures the immediate capture of high-risk events, providing an empirical basis for rapid mitigation in the initial rule judgment stage.

[0121] Secondly, for samples where the initial rule-based judgment did not directly identify a fault (including cases where no rule was triggered or only a soft threshold warning was triggered), a pre-trained machine learning model (such as XGBoost) was used for in-depth analysis. The model takes all features (including multi-dimensional "soft signals" such as voltage, current, temperature, and consistency) as input and outputs the confidence score or probability score for each potential fault type. The model showed high confidence in the test (average of approximately 97.2%), indicating that it has a good ability to distinguish between normal and abnormal patterns, and can effectively identify complex or ambiguous fault modes that are difficult for rules to capture, thus making up for the shortcomings of rule-based methods under boundary conditions.

[0122] like Figure 20 The figure shows the model confidence distribution. It illustrates the confidence distribution of 632 samples predicted by the model. The average confidence level of the model predictions is as high as 97.2%, with the vast majority of samples showing confidence levels concentrated in the high range of 95% to 100%, a median of 98.1%, and a standard deviation of only 1.9%. The 80% confidence threshold line clearly shows that most predictions far exceed this standard. This strongly demonstrates the model's highly reliable ability to distinguish between faults and normal states, providing crucial quantitative evidence for the effectiveness of the "model fine-tuning" stage.

[0123] The system integrates the initial rule-based judgment with the detailed model-based judgment, ensuring the reliability of the final diagnostic conclusion based on a conservative safety principle: Strong rule triggering priority: For high-severity faults detected by the rules (such as thermal runaway or leakage), the rule conclusion is adopted first, regardless of whether the model reports it, ensuring zero missed detections. Detailed model-based confirmation: When the model outputs a fault with high confidence and it does not conflict with the rule conclusion, the model diagnosis is directly confirmed. Conservative decision-making in case of inconsistency: When the rule and model conclusions are inconsistent, the more severe judgment direction is adopted (e.g., if the rule reports medium risk but the model reports high risk, the high risk is adopted; if the rule reports a warning but the model does not report a fault, the warning is still retained) to avoid missing significant risks. Soft threshold and model collaboration: For samples that trigger the soft threshold but the model does not confirm the fault, the system still outputs a warning status, achieving early risk indication. This fusion mechanism only requires joint judgment on a small number of samples (approximately 4%), effectively reducing the false alarm rate while ensuring full coverage of high-risk faults and improving the interpretability and acceptability of diagnostic conclusions.

[0124] like Figure 21 The pie chart shown is a distribution diagram of the diagnostic paths. This chart visually illustrates the path distribution of 1000 test samples within the "two-stage reasoning" framework. Of these, 36.8% of the samples were directly determined by rules, 26.2% were fine-tuned by the model, and 3.9% were decided through a fusion mechanism of rules and the model. This distribution directly verifies the effective division of labor and collaboration among the three core stages: initial rule judgment, fine-tuning by the model, and result fusion, demonstrating the operational efficiency of this framework.

[0125] like Figure 22 The diagram shows the consistency analysis between rules and the model. This diagram, through a pie chart on the left and a heatmap on the right, comprehensively analyzes the consistency between the rule and model diagnostic results. The left side shows that approximately 21.2% of the samples are completely consistent in their rule and model judgments, confirming that they can reach a consensus in a considerable number of cases, enhancing the rationality of the fusion mechanism. The heatmap on the right reveals the correspondence between rule judgments and model-predicted fault types. The high-value areas on the diagonal represent consistent situations, while the off-diagonal areas indicate inconsistencies that require the fusion strategy to focus on, providing specific scenario-based basis for fusion decisions.

[0126] like Figure 23The diagram shown illustrates the reasons behind the fusion decision-making process. It statistically displays the top 10 reasons and their frequencies used in the result fusion phase, clearly revealing the system's conservative decision-making logic. The most frequent reason is "exceeding the soft threshold but not the hard threshold, and the model prediction shows no fault" (331 times), reflecting the soft threshold warning mechanism. Meanwhile, situations such as "high / low severity level rules are triggered but the model does not report a fault, adopting a conservative principle" (358 times in total) directly demonstrate the safety-first fallback principle. This diagram transforms the fusion strategy from abstract principles into concrete, quantifiable operational rules, greatly enhancing the transparency and interpretability of the system's decision-making process.

[0127] like Figure 24 The diagram shows a cross-analysis of diagnostic paths and fault types. This heatmap illustrates the correspondence between different diagnostic paths (rule-based judgment, model prediction, fusion path, and soft threshold warning) and various fault types. It is clear that the rule-based path primarily captures hard threshold faults such as thermal runaway, leakage, and overcharging; the model-based path effectively identifies complex and hidden faults such as F02, F03, and M07; the fusion path handles a small number of cases requiring collaborative confirmation; and the soft threshold warning path covers a large number of early warning samples. This diagram comprehensively verifies the superior capabilities of the two-stage inference framework in achieving "rapid identification of obvious faults, in-depth judgment of complex faults, and early warning of potential risks" from the perspective of fault coverage.

[0128] Finally, through multi-dimensional charts and graphs verifying diagnostic path distribution, rule-model consistency analysis, fusion decision reasons, model confidence distribution, path-fault type cross-analysis, and rule triggering frequency, the method shown in this embodiment achieves: comprehensive fault coverage: rule paths mainly capture hard threshold faults (such as thermal runaway and overcharging), model paths effectively identify complex and hidden faults (such as F02 and F03), and soft threshold mechanisms provide early warnings. Efficient division of labor and collaboration: approximately 36.8% of samples are directly determined by rules, 26.2% by model-based fine-tuning, and 3.9% by the fusion mechanism, demonstrating the effectiveness of the "rapid screening-deep discrimination-conservative fusion" three-layer architecture. Clear decision-making logic: fusion reason analysis shows that the system strictly adheres to conservative principles (such as "high severity rule triggers immediate alarm" and "low severity rule triggers warning even if the model does not report"), ensuring the safety orientation and interpretability of diagnostic results.

[0129] This embodiment uses rules to quickly identify high-risk scenarios and combines them with models to fine-tune complex patterns, significantly reducing false alarms and missed alarms while achieving comprehensive diagnosis of various faults. Rules are based on explicit thresholds, model outputs include confidence levels, and the fusion strategy follows a publicly accepted conservative principle, making the final diagnostic conclusions easy to understand, verify, and trust. Rules and models complement and validate each other; even in cases of data anomalies, model uncertainty, or rule limitations, the fusion mechanism ensures robust diagnostic conclusions, improving the overall system's adaptability and reliability. It provides tiered outputs from rapid alarms to early warnings, clearly indicating the fault type and severity, facilitating targeted actions by maintenance personnel and improving battery safety management efficiency.

[0130] In another embodiment, the battery status assessment method of the present invention can be further extended to the visualization and user interaction of the assessment results, specifically providing a battery health index and early warning score interface display scheme for end users. This embodiment aims to solve the problems of unintuitive battery health assessment information presentation and insufficient interactivity. By designing a dual view of normal mode and expert mode, the complex multi-dimensional assessment results are transformed into intuitive information that users of different knowledge levels can understand, trust, and take action upon, thereby fully leveraging the value of the aforementioned assessment model and improving the user experience and operational efficiency of battery safety management. The specific interface display scheme is as follows: like Figure 25 The image shown is a schematic of the normal mode interface. This interface is designed for general users, with a simple layout and a clear focus. The core area integrates the battery health comprehensive index, key indicator trend previews, and a list of proactive warning information, allowing users to quickly and intuitively grasp the overall battery status and core risks, making it suitable for daily viewing.

[0131] On the one hand, the interface displays the battery's overall health score (range 0~100 points) in real time in the form of an eye-catching dashboard or scorecard.

[0132] like Figure 26The image shows a schematic diagram of the health index dashboard. This dashboard visually displays the battery's overall health score, clearly distinguishing health levels through color and scales. For example: 81-100 points (green) represent good health; 51-80 points (yellow) represent moderate health requiring attention; 21-50 points (orange) represent warnings requiring inspection; and 0-20 points (red) represent severe health requiring immediate action. This comprehensive index is derived from a weighted fusion of multiple tiered status indicators such as capacity decay, consistency degradation, and failure risk, allowing users to quickly grasp the overall battery condition without needing to understand the technical details. Alternatively, the interface can display the current values ​​and historical trends of key health sub-indicators in chart form, such as: State of Health (SOH) curve: reflecting the battery's capacity decay process; Internal resistance growth curve: monitoring internal aging and changes in conductivity; and Cell voltage / temperature difference trend: revealing the battery pack's internal consistency and thermal equilibrium status.

[0133] like Figure 27 The chart shown is a schematic diagram of the trend. This chart displays the historical trends and current values ​​of key battery health indicators (such as SOH and internal resistance) in the form of a curve, and provides users with percentile ranking information of this indicator among the same vehicle model (such as "Your battery capacity retention rate is better than 80% of the same vehicles"), helping users quantify the relative level of their own battery.

[0134] Fault Risk Warning Panel: The interface prominently displays potential fault risk information predicted by the assessment model. Each warning message clearly includes: fault type, such as thermal runaway risk, insulation failure, accelerated capacity decay, etc.; risk level / probability, distinguished by color (red / orange / yellow) and icon for high, medium, and low risk; and handling suggestions, providing concise and actionable advice (such as "Please immediately inspect the cooling system" or "It is recommended to reduce the frequency of fast charging").

[0135] like Figure 28 The image shows a schematic diagram of the warning panel. This panel clearly displays current battery fault risk warning information in a list format. Each piece of information includes the risk category, severity level, and specific operational suggestions. When the risk exceeds a preset threshold, the system proactively alerts the user through visual highlighting and other methods.

[0136] To enhance system transparency and user trust, the interface provides a hierarchical explanation function: Normal mode: When a user clicks on a warning, the main influencing factors are listed in concise text or bar charts (such as "Influencing factors: excessively high maximum temperature, decreased insulation resistance"), allowing users to intuitively understand the root cause of the problem.

[0137] like Figure 29The image shown is a schematic diagram of the model explanation panel (normal mode). This interface is designed for ordinary users. When a specific warning item is clicked, the main influencing factors that caused the warning are listed in a concise text or graphic format, allowing users to intuitively understand the root cause of the problem without having to deal with complex technical details.

[0138] Expert Mode: Operations and maintenance personnel can switch to this mode to view detailed model analysis results, including: feature contribution analysis charts (such as SHAP value visualization), clearly showing the specific contribution of each feature to the current risk assessment; raw data curves of key sensors for in-depth diagnosis and verification; and detailed logs of the assessment process and fusion decision-making to support technical review and auditing.

[0139] like Figure 30 The image shown is a schematic diagram of the expert mode interface. Building upon the standard mode, this interface provides professionals with more detailed data panels, access to advanced analysis charts (such as feature contribution analysis and raw data curves), and complete model explanations, meeting the needs of fleet maintenance and repair technicians for in-depth diagnosis and decision-making.

[0140] like Figure 31 The image shows a schematic diagram of the model interpretation panel (expert mode). This interface is designed for expert users and provides a detailed view of fault cause analysis, including model analysis graphs (such as SHAP contribution graphs) and in-depth technical information such as relevant raw sensor data curves, supporting professionals in accurate diagnosis and decision-making.

[0141] On the other hand, the standard mode interface integrates the aforementioned comprehensive indices, trend charts, and warning lists, featuring a clean and concise layout that highlights key information, making it suitable for daily viewing by vehicle owners to quickly understand battery status and risks. The expert mode, building upon the standard mode, adds a professional data panel, advanced analysis charts, and detailed model explanations to meet the in-depth diagnostic and decision-making needs of fleet maintenance and repair technicians. Users can switch between the two modes with a single click based on their role and needs. All displayed content supports real-time or near real-time updates, ensuring users receive the latest assessment results.

[0142] This embodiment, through the aforementioned interface design, transforms the complex results of multi-source data fusion and model evaluation into intuitive elements such as scores, charts, and colors, significantly lowering the cognitive threshold for users and improving the efficiency and accuracy of information acquisition. Real-time alerts and explicit suggestions significantly advance the fault identification process, enabling users to take preventative measures before battery performance noticeably declines or malfunctions occur, thus improving safety and battery lifespan. The design of both standard and expert modes precisely caters to the information needs and professional levels of different user groups, avoiding the confusion of ordinary users with technical details while providing ample diagnostic support for professionals, thereby enhancing the system's universality and acceptability. By providing model explanations and descriptions of influencing factors, the "black box" evaluation process is transformed into understandable and traceable "white box" information, effectively alleviating users' distrust of algorithmic decision-making and enhancing product credibility and market competitiveness.

[0143] In summary, this embodiment effectively delivers advanced battery health assessment and fault prediction capabilities from the backend to the frontend user through innovative user interface and interaction design, forming a complete closed loop from "data collection → intelligent assessment → result presentation → user action", thereby enhancing the practical value and user experience of the battery health management system.

[0144] This embodiment also provides a battery state assessment system for implementing the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0145] This embodiment provides a battery state assessment system, such as... Figure 32 As shown, it includes: a data acquisition module 100, a feature fusion module 200, and a state evaluation module 300, wherein the data acquisition module 100 is connected to the feature fusion module 200, and the feature fusion module 200 is connected to the state evaluation module 300. The data acquisition module 100 is used to acquire the set of manufacturing parameters and the set of operating parameters associated with the target battery; The feature fusion module 200 is used to fuse the manufacturing parameter set and the operating parameter set to obtain cross-fused features, wherein the cross-fused features are used to characterize the interaction relationship between the manufacturing parameter set and the operating parameter set; The state assessment module 300 is used to assess the state of the target battery based on cross-fusion features and obtain the assessment results.

[0146] In this embodiment, the data acquisition module 100 is specifically used to acquire a pre-built battery basic library, wherein the battery basic library stores multiple preset manufacturing parameters of preset batteries, the preset manufacturing parameters including at least electrode thickness, compaction density, formation efficiency, moisture content, and material formulation batch; acquire a pre-built battery operation library, wherein the battery operation library stores multiple preset battery historical operation parameters, the historical operation parameters including at least fast charge cycles, deep discharge frequency, single cell voltage difference, battery temperature difference, cumulative charge-discharge cycle count, and ambient temperature range; acquire the manufacturing parameter set of the target battery from the battery basic library, and acquire the operation parameter set of the target battery from the battery operation library.

[0147] In this embodiment, the feature fusion module 200 is specifically used to extract static manufacturing features from the set of manufacturing parameters, wherein the static manufacturing features include at least one of the following: initial capacity, initial internal resistance, electrode thickness and deviation, compaction density, residual water content of material, negative electrode pre-lithiation rate, first-week efficiency of formation test, and electrode defect marking; and to extract dynamic operating features from the set of operating parameters, wherein the dynamic operating features include real-time state features and historical statistical features, wherein the real-time state features include at least one of the following: real-time state of charge value, real-time temperature value, and real-time individual cell pressure difference, and the historical statistical features include at least one of the following: cumulative cycle count, equivalent full charge cycle count, historical average depth of discharge, preset state of charge dwell time ratio, maximum temperature and pressure difference within a historical time period, and capacity reduction percentage within a historical time period; and to combine and calculate the static manufacturing features and dynamic operating features to generate cross-fused features.

[0148] In this embodiment, the feature fusion module 200 is specifically used to obtain a pre-built battery fault mode library, wherein the battery fault mode library stores a mapping relationship between preset fault modes and preset feature combinations; based on the mapping relationship, at least one set of target static manufacturing features and target dynamic operating features associated with the target fault mode are determined from static manufacturing features and dynamic operating features; logical combination operations are performed on the target static manufacturing features and target dynamic operating features associated with the target fault mode to generate cross-fusion features.

[0149] In this embodiment of the application, the state evaluation module 300 is specifically used to construct at least one hierarchical state index of the target battery based on cross-fusion features; perform weighted fusion on each hierarchical state index to obtain the target state value; predict the future state of the target battery based on the hierarchical state index to obtain the state prediction result; and use the hierarchical state index, the target state value and the state prediction result as the evaluation result of the target battery.

[0150] In this embodiment of the application, the state assessment module 300 is specifically used to calculate the capacity index of the target battery based on cross-fusion characteristics, wherein the capacity index is used to characterize the capacity decay state of the target battery; calculate the consistency index of the target battery based on cross-fusion characteristics, wherein the consistency index is used to characterize the performance differences between individual cells of the target battery; and calculate the risk index of the target battery based on cross-fusion characteristics, wherein the risk index is used to characterize the potential failure risk level of the target battery.

[0151] In this embodiment of the application, the state assessment module 300 is specifically used to acquire historical sequence data of each layer state indicator; calculate the predicted state indicators of each layer state indicator in the future time period based on the historical sequence data; calculate the remaining effective time and replacement reminder time of the target battery based on the predicted state indicators, and use the remaining effective time and replacement reminder time as the state prediction result.

[0152] In this embodiment of the application, the system further includes: an output module, used to acquire business object tags associated with the target battery; generate early warning information corresponding to the evaluation result for each business object tag; and output the evaluation result and early warning information to the business object corresponding to the business object tag.

[0153] Please see Figure 33 , Figure 33 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 33 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system).

[0154] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0155] The memory 20 stores instructions executable by at least one processor 10 to cause at least one processor 10 to perform the method shown in the above embodiments.

[0156] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device as shown by a landing page for an app. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, which can be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0157] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0158] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0159] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0160] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for assessing battery state, characterized in that, The method includes: Obtain the set of manufacturing parameters and the set of operating parameters associated with the target battery; The manufacturing parameter set and the operating parameter set are fused to obtain cross-fusion features, wherein the cross-fusion features are used to characterize the interaction and influence relationship between the manufacturing parameter set and the operating parameter set; The target battery is evaluated based on the cross-fusion features to obtain the evaluation results.

2. The method according to claim 1, characterized in that, The acquisition of the set of manufacturing parameters and the set of operating parameters associated with the target battery includes: Obtain a pre-built battery base library, wherein the battery base library stores multiple preset manufacturing parameters of preset batteries, and the preset manufacturing parameters include at least electrode thickness, compaction density, formation efficiency, moisture content and material formulation batch. Obtain a pre-built battery runtime library, wherein the battery runtime library stores historical operating parameters of multiple preset batteries, and the historical operating parameters include at least the number of fast charge cycles, deep discharge frequency, single cell voltage difference, battery temperature difference, cumulative charge and discharge cycle count, and ambient temperature range. Obtain the set of manufacturing parameters for the target battery from the battery base library, and obtain the set of operating parameters for the target battery from the battery runtime library.

3. The method according to claim 1, characterized in that, The feature fusion of the manufacturing parameter set and the operating parameter set to obtain cross-fused features includes: Static manufacturing features are extracted from the set of manufacturing parameters, wherein the static manufacturing features include at least one of the following: initial capacity, initial internal resistance, electrode thickness and deviation, compaction density, residual water content of material, negative electrode pre-lithiation rate, first-week efficiency of formation test, and electrode defect marking. Dynamic operating features are extracted from the set of operating parameters. The dynamic operating features include real-time state features and historical statistical features. The real-time state features include at least one of real-time state of charge value, real-time temperature value, and real-time individual cell pressure difference. The historical statistical features include at least one of cumulative cycle count, equivalent full charge cycle count, historical average depth of discharge, preset state of charge dwell time ratio, maximum temperature and pressure difference within a historical time period, and capacity reduction percentage within a historical time period. The static manufacturing features and the dynamic operation features are combined and calculated to generate the cross-fusion features.

4. The method according to claim 3, characterized in that, The step of combining the static manufacturing features and the dynamic operation features to generate the cross-fusion features includes: Obtain a pre-built battery fault mode library, wherein the battery fault mode library stores a mapping relationship between preset fault modes and preset feature combinations; Based on the mapping relationship, at least one set of target static manufacturing features and target dynamic operating features associated with the target failure mode are determined from the static manufacturing features and the dynamic operating features; Logical combination operations are performed on the target static manufacturing features and target dynamic operation features associated with the target failure mode to generate the cross-fusion features.

5. The method according to claim 1, characterized in that, The state assessment of the target battery based on the cross-fusion features, to obtain the assessment result, includes: Construct at least one hierarchical state index of the target battery based on the cross-fusion features; The target state value is obtained by weighted fusion of each of the hierarchical state indicators. The future state of the target battery is predicted based on the hierarchical state index, and the state prediction result is obtained. The hierarchical state index, the target state value, and the state prediction result are used as the evaluation results of the target battery.

6. The method according to claim 5, characterized in that, The step of constructing at least one hierarchical state index of the target battery based on the cross-fusion features includes: The capacity index of the target battery is calculated based on the cross-fusion features, wherein the capacity index is used to characterize the capacity decay state of the target battery; The consistency index of the target battery is calculated based on the cross-fusion characteristics, wherein the consistency index is used to characterize the performance differences between individual cells of the target battery; The risk index of the target battery is calculated based on the cross-fusion features, wherein the risk index is used to characterize the degree of potential failure risk of the target battery.

7. The method according to claim 5, characterized in that, The prediction of the future state of the target battery based on the hierarchical state index, to obtain the state prediction result, includes: Obtain historical sequence data for each of the hierarchical state indicators; Calculate the predicted state indicators of each of the hierarchical state indicators in future time periods based on the historical sequence data. The remaining effective time and replacement reminder time of the target battery are calculated based on the predicted state indicators, and the remaining effective time and replacement reminder time are used as the state prediction results.

8. The method according to claim 1, characterized in that, After evaluating the state of the target battery based on the cross-fusion features and obtaining the evaluation result, the method further includes: Obtain the business object tag associated with the target battery; For each of the aforementioned business object tags, generate early warning information corresponding to the evaluation results; The evaluation results and the early warning information are output to the business object corresponding to the business object label.

9. A battery state assessment system, characterized in that, The system includes: a data acquisition module, a feature fusion module, and a state evaluation module, wherein the data acquisition module is connected to the feature fusion module, and the feature fusion module is connected to the state evaluation module; The data acquisition module is used to acquire the set of manufacturing parameters and the set of operating parameters associated with the target battery; The feature fusion module is used to perform feature fusion on the manufacturing parameter set and the operating parameter set to obtain cross-fusion features, wherein the cross-fusion features are used to characterize the interaction and influence relationship between the manufacturing parameter set and the operating parameter set; The state assessment module is used to assess the state of the target battery based on the cross-fusion features and obtain the assessment result.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to perform the method of any one of claims 1 to 8.

Images