Method and system for predicting residual life of battery of electrochemical energy storage system

Abstract

The invention relates to the technical field of electrochemical energy storage, and discloses a method and a system for predicting the residual life of a battery of an electrochemical energy storage system. According to the method, multi-dimensional operation data such as temperature distribution, current ripples and voltage attenuation curves in the charge-discharge cycle process of a target battery pack are collected, and feature weight parameters are generated in combination with real-time working conditions and historical database matching results; a degradation characteristic mode in a full life cycle is extracted as a reference template, and a dynamic degradation characteristic space is established; constructing a multi-scale coupled life prediction model, and constructing an adaptive evaluation function by iteratively updating model parameters and reference template physical field coupling parameters; triggering cross-cycle data verification, and carrying out working condition compatibility verification; and generating a feedback matrix based on the prediction error index and the verification result, and dynamically adjusting the mapping relation between the feature weight and the reference template. The method improves the accuracy and adaptability of battery residual life prediction, and is suitable for an electrochemical energy storage system under complex working conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electrochemical energy storage, in particular to a method and system for predicting the remaining life of a battery in an electrochemical energy storage system. BACKGROUND

[0002] In an electrochemical energy storage system, accurate prediction of the remaining life of the battery is an important prerequisite for stable operation of the system. With the rapid development of the new energy industry, electrochemical energy storage technology is widely used in renewable energy grid connection, smart grid peak shaving and other scenarios, and the life degradation characteristics of the battery as the core component of the energy storage system directly affect the economy and safety of the system.

[0003] Currently, battery remaining life prediction methods rely on single-dimensional operating data, such as life estimation based only on voltage or current changes, which cannot fully reflect the degradation state of the battery under complex operating conditions. In actual operation, the battery pack often faces problems such as dynamic charging and discharging rate, uneven temperature distribution, and single data dimension prediction model is prone to large errors. At the same time, existing models mostly use fixed degradation feature templates and do not consider the dynamic changes of battery degradation patterns in different operating cycles, resulting in a gradual decline in prediction accuracy in long-term cycling.

[0004] Traditional prediction methods lack effective cross-cycle data validation mechanisms, and when the battery operating conditions change suddenly, the model is difficult to quickly adapt to the new operating mode, and is prone to distorted prediction results. These problems make it difficult for existing prediction methods to meet the high-precision and high-adaptability requirements of electrochemical energy storage systems for battery life prediction, limiting the efficient management and optimal operation of the energy storage system. SUMMARY

[0005] The present application aims to provide a method and system for predicting the remaining life of a battery in an electrochemical energy storage system to solve the problems raised in the background.

[0006] To achieve the above-mentioned purpose, the present application provides a method for predicting the remaining life of a battery in an electrochemical energy storage system, which comprises:

[0007] Collecting multi-dimensional operating data of the target battery pack during the charging and discharging cycle, the multi-dimensional operating data including temperature distribution data, current ripple data and voltage attenuation curve, and generating feature weight parameters according to the matching results of real-time operating conditions and historical operating database;

[0008] Extracting the degradation feature mode of the target battery pack throughout its life cycle, the degradation feature mode being generated by analyzing the capacity attenuation trajectory under different charging and discharging rates, and using the degradation feature mode as a reference template for establishing a dynamic degradation feature space;

[0009] The multi-scale coupled life prediction model is constructed, model parameters are iteratively updated in the direction of minimizing prediction error, feature reconstruction instructions are generated each time iteration, and the physical field coupling parameters of the benchmark template are updated, a model adaptive evaluation function is constructed based on the correlation between feature weight parameters and real-time operation data;

[0010] Cross-cycle data verification is triggered according to the amplitude of the feature reconstruction instruction, the verification data package matching the current charge and discharge mode is extracted from the historical operation database, and the working condition compatibility of the verification data package is verified through the electrochemical conversion rule in the benchmark template;

[0011] The prediction error index output by the model adaptive evaluation function and the verification data package verification result generate a feedback matrix, which dynamically adjusts the mapping relationship between the feature weight parameters and the benchmark template.

[0012] Preferably, the multi-scale coupled life prediction model is constructed, model parameters are iteratively updated in the direction of minimizing prediction error, including:

[0013] The degradation feature mode is mapped to a high-dimensional tensor, each dimension corresponds to the adjustable range of the physical field coupling parameter, and the correlation coefficient of the electrochemical parameter is calculated based on the current charge and discharge mode;

[0014] An error neighborhood table is constructed according to the correlation coefficient, historical failure cases associated with the current degradation feature mode and their physical field coupling parameter differences are recorded, and a candidate reconstruction instruction set is generated;

[0015] A double-constraint screening mechanism is used to select a target reconstruction instruction from the candidate reconstruction instruction set, the constraint conditions include the error reduction rate of the candidate reconstruction instruction in historical correction and the parameter compatibility threshold of the electrochemical conversion rule;

[0016] The target reconstruction instruction is applied to the degradation feature mode, the parameter values of the corresponding dimensions in the high-dimensional tensor are dynamically updated, and the screening threshold of the error neighborhood table is adjusted to trigger the elimination of low-correlation data;

[0017] The working condition matching degree of the updated degradation feature mode and the verification data package is recalculated, and the matching degree change is fed back to the adjustment process of the feature weight parameter.

[0018] Preferably, the double-constraint screening mechanism is used to select a target reconstruction instruction from the candidate reconstruction instruction set, including:

[0019] The error reduction rates of each candidate reconstruction instruction in a preset time window are counted, and the ratio of the number of historical successful corrections to the priority change of the reconstruction instruction is normalized to the initial screening weight;

[0020] The parameter constraint field corresponding to the electrochemical conversion rule in the reference template is extracted, the coverage ratio of the parameter dimension corresponding to the candidate reconstruction instruction to the preset constraint field is calculated, and the coverage ratio is mapped as a compatibility constraint weight;

[0021] The initial screening weight and the compatibility constraint weight are dynamically superimposed, and a superposition coefficient is adjusted according to the distribution density of low-correlation data in the error neighborhood table, so that the compatibility constraint weight obtains a higher proportion in the high-risk area of prediction error;

[0022] According to the comprehensive screening probability, an instruction selection operation is performed, and the compatibility state of the candidate reconstruction instruction and the core parameter of the reference template is verified. If there is a conflict, a standby reconstruction instruction queue is enabled.

[0023] Preferably, the dynamic superposition of the initial screening weight and the compatibility constraint weight comprises:

[0024] Based on the comparison between the proportion of low-correlation data in the error neighborhood table and the historical maximum data capacity, the error risk level is calculated by a multi-segment function;

[0025] A dynamic superposition factor is generated according to the error risk level, a threshold segmented adjustment mechanism is adopted, and the historical sliding mean value is used as a stability factor to control the fluctuation amplitude of the superposition factor;

[0026] The initial screening weight and the dynamic superposition factor are associated and calculated, and the associated result is fused with the compatibility constraint weight;

[0027] The action strength of the stability factor is adjusted according to the deviation degree of the superposition factor and the historical sliding mean value.

[0028] Preferably, the model adaptive evaluation function is constructed based on the association relationship between the feature weight parameter and the real-time running data, comprising:

[0029] According to the distribution characteristics of the feature weight parameter, the spatio-temporal correlation degree of each dimension parameter in the real-time running data and the battery degradation rate is calculated;

[0030] A dynamic correction factor is generated based on the spatio-temporal correlation degree, the parameter change trend of the high attenuation rate area and the temperature abnormal area is associated, and the action range of the correction factor is restricted by the physical field coupling parameter of the reference template;

[0031] The feature weight parameter, the dynamic correction factor and the spatio-temporal correlation degree are fused to construct the model adaptive evaluation function, and a penalty term related to the deviation of the physical field coupling parameter is introduced.

[0032] Preferably, the dynamic correction factor is generated based on the spatio-temporal correlation degree, comprising:

[0033] Based on the spatiotemporal correlation and the inverse correlation between electrochemical parameter fluctuations and capacity decay gradients, the difference in parameter changes between the high decay rate region and the temperature anomaly region is calculated.

[0034] Generate an initial correction factor based on the difference in parameter changes, associate the parameter sensitivity in the high decay rate region with the electrochemical response delay in the temperature anomaly region, and constrain the rate of change of the correction factor through the physical field coupling parameters of the benchmark template;

[0035] Based on the historical correction record of the physics coupling parameters, a decay control factor is applied to the initial correction factor.

[0036] Preferably, the generation of a feedback matrix by combining the prediction error index output by the model adaptive evaluation function and the verification data packet verification result includes:

[0037] Calculate the error and compatibility coupling coefficient based on the prediction error index and the compatibility compliance rate in the verification data packet check result;

[0038] An initial feedback matrix is ​​generated based on the error and compatibility coupling coefficient, the compatibility loss in high-error areas is associated with feature reconstruction instructions, and the update strength of the feedback matrix is ​​constrained by the historical correction record of the benchmark template.

[0039] A multi-dimensional feedback model is constructed by integrating the error with the compatibility coupling coefficient, the prediction error index and the compatibility compliance rate.

[0040] Preferably, the method further comprises:

[0041] Establish a quantitative model for battery pack degradation status and analyze the primary and secondary characteristic quantities in multi-dimensional operating data;

[0042] Construct a topological network of degradation feature space and encode the historical degradation patterns of different charge and discharge stages into topological nodes;

[0043] Calculate the similarity path between the current running data and the topological nodes through the dynamic time warping algorithm;

[0044] When the similarity path exceeds the preset threshold, the prediction model parameter reset mechanism is triggered.

[0045] Preferably, the calculating of the similarity path between the current running data and the topological node by the dynamic time warping algorithm includes:

[0046] Perform sliding window segmentation on the current running data to generate multiple data segments and their timestamp indexes;

[0047] Extract the electrochemical eigenvector of each data segment and calculate the dynamic bending path corresponding to the topological node eigenvector;

[0048] The similarity matrix is generated based on a dynamic bending path, and the optimal matching path is located through a backtracking algorithm.

[0049] The model correction instruction is generated according to the deviation degree of the optimal matching path from a preset threshold.

[0050] Preferably, the application further comprises a prediction system for the remaining life of a battery of an electrochemical energy storage system, for implementing the prediction method for the remaining life of a battery of an electrochemical energy storage system as described above, the system comprising:

[0051] A data acquisition module acquires multi-dimensional operating data of the target battery pack during the charging and discharging cycle process, and generates a feature weight parameter according to the matching result of the real-time operating condition and the historical operating database;

[0052] A template construction module extracts the degradation feature mode of the target battery pack throughout its life cycle, and takes the degradation feature mode as a benchmark template for establishing a dynamic degradation feature space;

[0053] A model iteration module constructs a multi-scale coupled life prediction model, drives the model parameters to iteratively update in the direction of minimizing the prediction error, and constructs a model adaptive evaluation function based on the correlation between the feature weight parameter and the real-time operating data;

[0054] A verification triggering module triggers cross-cycle data verification according to the amplitude of the feature reconstruction instruction, and performs operating condition compatibility verification on the verification data packet through the electrochemical conversion rule in the benchmark template;

[0055] A feedback generation module generates a feedback matrix through the prediction error index output by the model adaptive evaluation function and the verification data packet verification result, and dynamically adjusts the mapping relationship between the feature weight parameter and the benchmark template.

[0056] Compared with the prior art, the application has the following beneficial effects:

[0057] By acquiring multi-dimensional operating data, including temperature distribution, current ripple, and voltage attenuation curve information, the degradation characteristics of the battery under different operating conditions can be fully captured, avoiding the information partiality caused by a single data dimension. The feature weight parameter is generated based on the matching result of the real-time operating condition and the historical database, so that the model can dynamically adjust the influence weight of each feature according to the actual operating state, improving the adaptability to complex operating conditions.

[0058] The degradation feature pattern in the whole life cycle is extracted as a benchmark template, and a dynamic degradation feature space is established, which breaks through the limitation of traditional fixed template. By analyzing the capacity attenuation trajectory under different charge and discharge rates, the benchmark template can reflect the degradation law of the battery under diversified operating conditions, providing a more practical feature reference for subsequent prediction. At the same time, the physical field coupling parameters of the benchmark template are updated during the model iteration process, further enhancing the tracking ability of the template to the dynamic degradation process of the battery.

[0059] The multi-scale coupled life prediction model is optimized in the direction of minimizing prediction error through parameter iteration, which improves the prediction accuracy of the model. The adaptive evaluation function constructed by combining feature weight parameters and real-time running data can monitor the prediction effect of the model in real time, and dynamically adjust the model parameters according to the error situation, so that the model can maintain stable prediction performance in long-term operation.

[0060] The introduction of cross-cycle data verification mechanism can effectively solve the problem of insufficient model adaptability when the working condition changes. This mechanism enables the model to quickly call similar historical data for verification and calibration when the working condition changes, ensuring the consistency and reliability of the prediction results in different operation cycles.

[0061] Through the feedback matrix, the mapping relationship between the feature weight parameters and the benchmark template is dynamically adjusted, realizing the collaborative optimization of each component of the model. The dynamic update of feature reconstruction instructions and physical field coupling parameters enables the model to continuously adapt to the nonlinear changes in the battery degradation process, maintaining high prediction accuracy throughout the battery life cycle, and providing a more reliable life assessment basis for the battery management of electrochemical energy storage systems. BRIEF DESCRIPTION OF DRAWINGS

[0062] Figure 1 The working principle diagram of the electrochemical energy storage system battery residual life prediction method described in the present application;

[0063] Figure 2 The flowchart of multi-scale model parameter iterative update;

[0064] Figure 3 The flowchart of dynamic weight superposition control;

[0065] Figure 4 The flowchart of model adaptive evaluation function construction;

[0066] Figure 5 The flowchart of feedback matrix generation. DETAILED DESCRIPTION

[0067] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the scope of protection of the present application.

[0068] With reference to Figure 1 The present application provides a method for predicting the remaining life of a battery of an electrochemical energy storage system, the method comprising:

[0069] High-precision life prediction is achieved through multi-dimensional data fusion and dynamic modeling. The system first collects multi-dimensional operating data such as temperature distribution data, current ripple data, and voltage attenuation curves during the charging and discharging cycle of the target battery pack, generates characteristic weight parameters through real-time working condition and historical database matching analysis. A full life cycle degradation characteristic mode analysis framework is established, and a reference template is generated based on the capacity attenuation trajectory under different charging and discharging rates. A multi-scale coupled life prediction model is constructed, and an iterative optimization mechanism is used to drive model parameter updating, generate characteristic reconstruction instructions, and adjust physical field coupling parameters. Through a cross-cycle data verification mechanism, the verification data package is checked for working condition compatibility, and a feedback matrix is constructed using the prediction error index generated by the model adaptive evaluation function and the verification results to dynamically optimize the mapping relationship between the characteristic weight parameters and the reference template.

[0070] Embodiment 1: With reference to Figure 2 The core of this implementation is to convert battery degradation characteristics into a computable high-dimensional data structure, and continuously optimize model parameters through an iterative mechanism to ultimately achieve accurate prediction of the remaining life of the battery.

[0071] The system first performs tensor mapping processing on the collected battery degradation characteristics. This process converts multi-dimensional operating parameters such as temperature distribution, current ripple, and voltage attenuation into mathematical expressions with clear physical meaning. Each dimension corresponds to a physical field coupling parameter, which defines the electrochemical behavior boundary of the battery under different working conditions. The establishment of the tensor structure takes into account the interaction between parameters, enabling the model to handle multiple interrelated degradation factors simultaneously. During the charging and discharging cycle, the system calculates the correlation coefficients of electrochemical parameters in real time, which reflect the relative influence of different parameters on the battery degradation rate.

[0072] The construction of the error neighborhood table is a key step in this implementation. This data structure stores historical records of failure cases similar to the current battery state, including the corresponding physical field parameter combinations and the final failure modes they caused. The system classifies these historical cases through clustering algorithms, identifying case groups with similar degradation trajectories. Each case group is associated with a set of candidate reconstruction instructions, which contain parameter adjustment suggestions that can improve model prediction accuracy. The formation of the candidate instruction set takes into account the historical effectiveness of parameter adjustments and the compatibility between different parameter combinations.

[0073] The double-constraint screening mechanism is responsible for selecting the optimal adjustment scheme from numerous candidate instructions. This mechanism operates in two complementary dimensions: on the one hand, it evaluates the performance of each candidate instruction in historical correction cases, quantifying its actual effect on reducing prediction errors; on the other hand, it detects the matching degree of candidate instructions with the current electrochemical conversion rules, ensuring that parameter adjustments do not violate basic physical and chemical laws. The screening process uses a dynamic weight allocation strategy to automatically adjust the relative importance of the two dimensions according to the current prediction state of the model. When the model prediction results fluctuate greatly, the system will prioritize parameter compatibility constraints; in the relatively stable prediction stage, it will focus more on error reduction effects.

[0074] The execution of the target reconstruction instruction involves multiple coordinated operations. The system first analyzes the instruction content to determine the physical field coupling parameters that need to be adjusted and their target value range. The parameter adjustment process uses a gradual strategy, gradually approaching the optimal value through multiple small adjustments to avoid model instability caused by a single large adjustment. After each parameter update, the system re-evaluates the data relevance in the error neighborhood table, eliminating historical cases with low matching degree to the new parameter combination. This dynamic updating mechanism ensures that the model is always based on the most relevant historical data for prediction.

[0075] The working condition matching degree calculation module is responsible for verifying the effectiveness of parameter adjustment. This module compares the updated degradation feature mode with the validation data set in multiple dimensions, calculating the matching degree in each feature dimension. The matching degree evaluation not only considers the closeness of static parameter values, but also analyzes the consistency of parameter change trends. The system establishes a special feedback channel to transmit matching degree change information to the feature weight adjustment module, forming a closed-loop optimization system. This feedback mechanism enables the model to continuously fine-tune its internal parameters based on actual effects, gradually improving prediction accuracy.

[0076] The adjustment of physical field coupling parameters follows certain constraints. The system maintains a parameter safety boundary database, which defines the allowable adjustment range and rate of change for each parameter. Before executing each reconstruction instruction, a boundary check is performed to prevent the parameters from entering a dangerous region. For critical parameters, the system also sets up a mutual restraint relationship, which automatically limits the adjustment amplitude of related parameters when a certain parameter approaches the upper limit. This protection mechanism effectively prevents model failure caused by uncontrolled parameters.

[0077] During the model iteration process, special attention is paid to the degradation characteristics of different time scales. Short-term fluctuation characteristics reflect the instantaneous state changes of the battery, medium-term trend characteristics embody the stage characteristics of the degradation process, and long-term decay characteristics show the overall aging trajectory of the battery. The system uses a multi-resolution analysis method to process data at these three time scales simultaneously, ensuring that the prediction results reflect the current state and conform to long-term change patterns. The fusion of characteristics at different time scales is achieved through a weighting algorithm, and the weight coefficients are dynamically adjusted according to the prediction time span.

[0078] The error correction strategy adopts a hierarchical response mechanism. For small-scale prediction deviations, the system performs routine parameter fine-tuning; when moderate deviations occur, the reconstruction of the error neighborhood table and the update of the candidate instruction set are triggered; when major prediction errors are encountered, a comprehensive model diagnosis and parameter reset process is initiated. This hierarchical response ensures daily operational efficiency while providing adequate measures for abnormal situations.

[0079] Embodiment 2: Refer to Figure 3 , the dynamic weight superposition algorithm of the double-constraint screening mechanism and its specific application in the battery life prediction model. The core of this implementation is to establish a multi-dimensional decision-making framework, and to achieve precise screening of reconstruction instructions through intelligent weight allocation strategies, thereby optimizing the model parameter adjustment process.

[0080] An evaluation system for candidate reconstruction instructions is constructed at the initial stage of system operation. This system includes two independent and complementary evaluation dimensions. The historical correction effect dimension quantifies the performance of each candidate instruction in actual application by analyzing error correction records within a preset time window. The system counts the number of successful corrections for each instruction, records the magnitude of error reduction, and associates these data with the change trend of instruction priority. These raw indicators are normalized to convert them into comparable initial screening weights. The processing considers the effect differences under different battery operating conditions to ensure the fairness of weight allocation. The initial weight calculation module uses a sliding time window strategy, giving relatively higher consideration to newer correction records, allowing the system to adapt to changes in battery degradation characteristics.

[0081] The electrochemical compatibility dimension extracts parameter constraints from the baseline template, which reflect the basic physical and chemical laws in the battery operation process. The system analyzes the key parameter fields in the electrochemical conversion rules and establishes a constraint relationship network between the parameters. The candidate reconstruction instructions are mapped into this network, and the matching degree of their parameter adjustment scheme with the constraint field is calculated. The compatibility evaluation not only checks whether the parameter values are within the allowed range, but also analyzes whether the parameter combination violates the inherent mutual relationship. The evaluation results are converted into compatibility constraint weights through a nonlinear mapping function, which sets a higher sensitivity to key parameters and significantly reduces their weights when detecting adjustment schemes that may endanger battery safety.

[0082] The error neighborhood table analysis module monitors the data distribution characteristics in real time, especially the proportion change of low correlation data. This module uses a density clustering algorithm to identify sparse areas in the data distribution, which usually correspond to high-risk states predicted by the model. The system calculates the error risk level according to the aggregation degree of low correlation data, and the higher the risk level, the greater the uncertainty of the current prediction. The risk level calculation uses a multi-section function to process, dividing the continuous density index into several risk intervals, each corresponding to a different decision strategy. The historical moving average serves as a stability benchmark to determine whether the current risk level is within the normal fluctuation range.

[0083] The dynamic overlay factor generator calculates the weight adjustment parameters based on the error risk level and stability index. This generator uses a threshold segmentation adjustment mechanism to set different adjustment sensitivities in different risk intervals. When the system is in a low-risk state, the overlay factor remains relatively stable and mainly relies on the initial screening weight for decision-making; as the risk level rises, the overlay factor gradually enhances the influence of compatibility constraints, and in extreme cases, it can completely switch to a conservative decision-making mode. The stability factor controls the fluctuation amplitude of the overlay factor, preventing decision oscillation caused by short-term data fluctuations. The deviation degree of the overlay factor from the historical average is continuously monitored, and the stability factor's effectiveness is automatically adjusted when the deviation exceeds the preset threshold.

[0084] The weight fusion process uses a dynamic mixing algorithm to combine the initial screening weight and the compatibility constraint weight according to the proportion determined by the overlay factor. The fusion algorithm considers the nonlinear interaction between the weights and determines the optimal mixing method through cross-validation. In some special cases, when there is a serious conflict between the two weight indicators, the system will start a conflict resolution protocol to prioritize the parameter constraints that are crucial to battery safety. The fusion result generates the comprehensive screening probability of each candidate instruction, which drives the final selection operation after standardization processing.

[0085] The instruction selection operation adopts a strategy of combining probabilistic sampling and deterministic checking. The system first randomly pre-selects several candidate instructions according to the comprehensive screening probability distribution, and then performs detailed compatibility verification on these instructions. The verification process checks the interaction between the pre-selected instructions and the core parameters of the reference template, especially those complex adjustment schemes involving multiple physical field coupling. When parameter conflicts are found, the system records the conflict type and severity, and automatically switches to the backup instruction queue. The instructions in the backup queue have been pre-screened and have a higher basic compatibility guarantee. After the selection operation is completed, the selected target reconstruction instruction enters the execution queue, and the historical selection record of the related instruction is updated.

[0086] The conflict detection and processing subsystem continuously monitors the parameter state during instruction execution. The system maintains a parameter interaction knowledge base that stores side effect information for various parameter combinations. When a potential conflict is detected, the system performs pattern matching based on the cases in the knowledge base to quickly identify the conflict type and take appropriate mitigation measures. For new conflict patterns, the system initiates a learning process to supplement the knowledge base with new conflict characteristics and their handling schemes. Conflict handling adopts a gradual strategy, from warning, parameter fine-tuning to instruction rollback, gradually upgrading response measures according to the severity of the conflict.

[0087] The backup instruction queue management module is responsible for maintaining a pre-screened candidate instruction set. This module regularly extracts effective instructions from historical successful cases and performs compatibility enhancement processing. Newly generated candidate instructions need to be observed and verified in the backup queue before entering the main selection pool. The instructions in the queue are prioritized, and the sorting criteria include historical success rate, compatibility score, and recent usage frequency. When the main selection pool cannot provide suitable instructions due to conflicts or other reasons, the system will select alternative solutions from the backup queue according to priority.

[0088] A complete instruction life cycle management mechanism is established during the implementation of the system. Newly generated candidate instructions first enter the observation period, during which they can only be used for low-risk prediction tasks; after sufficient verification, they enter the regular use stage; when the correction effect of the instructions continues to decline or compatibility problems occur, they are gradually downgraded until they are eliminated. This life cycle management ensures the dynamic updating of the instruction library, constantly adapting to changes in battery degradation characteristics.

[0089] The data tracking and diagnosis subsystem records the complete context information of all screening decisions, including the model state at the time, input parameters, weight calculation process, and final selection results. These data are used for subsequent decision quality analysis and algorithm improvement. The diagnosis tool can reproduce historical decision scenarios to help understand the reasons for specific screening results. When systematic bias or decision defects are found, the system initiates a calibration process to adjust key parameters in the weight calculation.

[0090] The visualization interface provides an interpretable demonstration of the decision-making process for the operator. The evaluation indicators of candidate instructions, weight distribution ratios, and the decision-making path of the final screening results are presented in a graphical manner. The interface supports interactive exploration, allowing for in-depth review of the detailed evaluation data and historical performance records of specific instructions. These visualization tools enhance the transparency of the system, facilitating human supervision and intervention.

[0091] Embodiment 3: Refer to Figure 4 , focusing on the construction of the model adaptive evaluation function and the generation of dynamic correction factors, it describes how to achieve accurate evaluation of battery degradation state through spatio-temporal correlation analysis and multi-dimensional data fusion. This implementation establishes a complete parameter correlation calculation system, organically integrating battery operating data, physical field coupling parameters, and degradation characteristics to form an adaptive evaluation framework.

[0092] The system first analyzes the distribution characteristics of feature weight parameters, using a clustering algorithm based on kernel density estimation to identify high-density regions in the parameter space. These regions correspond to the typical state of the battery under certain working conditions, and each cluster center is considered as a reference benchmark point. Real-time operating data is projected into this parameter space, and its multidimensional distance from each benchmark point is calculated to form a spatio-temporal correlation matrix. Each element in the matrix represents the correlation strength between a specific dimension parameter and the battery degradation rate, taking into account the time delay effect and spatial distribution characteristics. The correlation calculation uses an improved grey correlation analysis method, which can handle nonlinear and non-stationary parameter relationships.

[0093] The generation process of dynamic correction factors is based on the analysis of electrochemical parameter fluctuations. The system monitors the data flow in the high attenuation rate region and the temperature anomaly region, and aligns the change trajectories of the two regions using the dynamic time warping algorithm. Key turning points and trends are identified during trajectory alignment, and the parameter difference between the two regions is quantified. The difference calculation not only considers the instantaneous value difference, but also analyzes the matching degree of the change rate and the persistence of the trend. The initial correction factor is generated based on the difference size and direction, and its calculation process follows the following relationship:

[0094]

[0095] where, represents the initial correction factor, is the size of the analysis window, is the temperature gradient change of the th sampling point, corresponds to the capacity attenuation gradient, is the local fluctuation amplitude, is a small constant to prevent division by zero. This formula quantifies the dynamic coupling relationship between temperature anomalies and capacity fade, and normalizes the factor to fall within the standard range.

[0096] Historical records of physical field coupling parameters are organized into a time series database, which is used to construct a decay control model. This model analyzes the decay patterns of the correction factor under different operating conditions and establishes a prediction curve of the factor's effectiveness over time. The decay control utilizes a dynamic damping mechanism, automatically adjusting the damping coefficient based on the current battery state and the model's prediction error. When the prediction error is large, the system reduces the damping strength to allow the correction factor to take greater effect. When the model stabilizes, the damping is appropriately increased to prevent oscillations caused by overcorrection.

[0097] The evaluation function is constructed using a multi-layered fusion strategy. The base layer integrates feature weight parameters and spatiotemporal correlation to generate a preliminary evaluation value. The middle layer introduces dynamic correction factors to adapt the preliminary evaluation value to the working conditions. The top layer incorporates penalty terms related to deviations in physical field coupling parameters to constrain the evaluation results to a reasonable range. The penalty term is calculated using a piecewise function, with different sensitivities for different types of parameter deviations. Minor deviations only result in a linear penalty, while severe deviations in key parameters trigger exponentially increasing penalties, forcing the evaluation results to reflect the significance of the parameter anomaly.

[0098] Analysis of high decay rate regions pays special attention to parameter sensitivity characteristics. The system uses micro-perturbation testing to identify the contribution of each parameter to the decay rate and establishes a sensitivity ranking list. Parameters with high sensitivity receive greater weight in the evaluation function, and changes in their sensitivity can significantly impact the final evaluation results. Sensitivity analysis is regularly updated to adapt to changes in parameter importance during battery aging. Detection of temperature anomalies utilizes an adaptive threshold method, with the threshold dynamically adjusted based on historical battery operating data to avoid false positives or negatives caused by fixed thresholds.

[0099] Electrochemical response delay is modeled using a transfer function approach. The system performs frequency domain analysis on input and output data during different charge and discharge phases to identify the delay characteristics and amplitude-frequency characteristics of the battery response. These characteristics are quantified as delay parameters, which are used to modify the alignment process of the dynamic time warping algorithm. Delay modeling specifically focuses on the phase relationship between temperature change and capacity decay, accurately capturing the hysteresis characteristics of thermal effects on battery performance.

[0100] Post-processing of the assessment results includes smoothing filtering and confidence annotation. Smoothing filtering utilizes a physically constrained algorithm to eliminate noise while preserving key characteristic points of the assessment curve. Confidence calculations consider factors such as data integrity, model fit, and parameter reliability, assigning a quality label to each assessment point. When the confidence level falls below a threshold, the system automatically triggers a data review process, pausing assessment output if necessary until data reliability is confirmed.

[0101] The real-time monitoring subsystem continuously tracks the running state of the evaluation function. It records the trajectory of evaluation values, detects abnormal fluctuations and trend reversals. When it finds that the evaluation values continuously deviate from the normal range, it starts the root cause analysis process to trace back the parameter change chain that leads to the deviation. The monitoring results are used to optimize the internal parameters of the evaluation function, forming a closed-loop system for continuous improvement.

[0102] The visualization interface converts complex evaluation processes into intuitive graphical displays. The three-dimensional surface graph presents the correspondence between evaluation values and key parameters, the dynamic heat map displays the state changes in different regions, and the trend comparison curve juxtaposes the evaluation results with actual measurement values. The interface supports interactive queries, allowing in-depth viewing of evaluation details and parameter composition at specific time points.

[0103] Example 4: Refer to Figure 5 , the generation and application process of the feedback matrix, through specific examples to show how the system handles the coupling relationship between prediction errors and validation results. The core of this implementation is to establish a multi-source data fusion mechanism, organically combining model output, measured data and historical cases to form a feedback system with self-correcting ability.

[0104] Taking a typical lithium-ion battery pack life prediction scenario as an example, the following key parameters are collected during system operation:

[0105] Table 1: Battery pack operating parameters and prediction error records.

[0106]

[0107] The system first analyzes the coupling relationship between the error index and the compatibility compliance rate. At time t1, although the error index is low (0.006), the compatibility compliance rate has shown a downward trend (92%). The system finds through historical case matching that similar situations in past data often indicate an impending increase in error. The coupling coefficient calculation module quantifies this early warning signal as a risk indicator, marking it as an attention area in the initial feedback matrix.

[0108] When running to time t3, the temperature rises to 31.8°C accompanied by an increase in current ripple to 7.2%, and the error index jumps to 0.015. The system retrieves the error neighborhood table and finds three historical cases similar to this state: case A ultimately leads to prediction failure, case B recovers stability through parameter adjustment, and case C enters accelerated decay. The feedback matrix generator analyzes the key differences between these three cases and finds that the synchronization of voltage decay rate and temperature change is the decisive factor. The synchronization of voltage decay and temperature rise in the current data is closer to the characteristics of case A, so the system marks a high-risk warning in the feedback matrix.

[0109] The verification data packet check process found that the measured capacity decay rate in the current charge and discharge mode exceeded the expected range of the baseline template. The compatibility check module decomposed this deviation into three components: temperature influence factor, current distribution factor, and aging acceleration factor. By comparing the operating parameters in the verification data packet with historical records, the system identified that the abnormal current distribution was the main cause. This finding was encoded as a modification instruction for the feedback matrix, requiring adjustment of the weight coefficient of the current ripple parameter.

[0110] The update intensity control of the feedback matrix uses a case learning mechanism. The system analyzes the intervention effects in similar scenarios in the historical modification records and finds that for compatibility decline caused by abnormal current distribution, moderate parameter adjustment is most effective. Overly strong modification can lead to overfitting of the current anomaly, while overly weak modification cannot prevent error expansion. Based on this understanding, the system sets the update intensity of the current feedback matrix to the middle level, corresponding to a parameter weight adjustment amplitude of 20%-30%.

[0111] The construction process of the multi-dimensional feedback model demonstrates the intelligent integration capability of the system. At time t4, although the temperature has fallen, the error index continues to rise to 0.017. Instead of simply relying on the temperature parameter for judgment, the system comprehensively analyzes the second derivative change of the voltage decay curve and the spectral characteristics of the current ripple. These analyses find that the voltage decay is forming a new acceleration trend, and high-frequency components appear in the current ripple. The feedback model matches these two characteristics with the historical case library and identifies a typical pattern of battery separator aging, thereby adding a material-level degradation marker to the feedback matrix.

[0112] The disposal strategy for high-error areas reflects the hierarchical decision-making ability of the system. When the error index exceeds the 0.015 threshold, the system automatically activates a three-level response mechanism: the primary response adjusts the model parameter weights, the intermediate response modifies the feature extraction algorithm, and the advanced response triggers the baseline template update. In the current example, the system first attempts to improve the prediction by redistributing the weights of the temperature and current parameters, while monitoring the error trend of subsequent data points. This gradual response avoids overreaction while effectively controlling error expansion.

[0113] The visualization of the feedback matrix uses multi-dimensional projection technology. The operation interface displays a three-dimensional coordinate space, with the three axes representing the prediction error, compatibility compliance rate, and parameter sensitivity, respectively. Each data point forms a dynamic cloud cluster in the space, with the color depth representing the risk level and the shape change reflecting the trend characteristics. The data at times t1 to t4 forms a clear evolution path in the space, helping the operator understand the migration process of the system state.

[0114] The conflict resolution mechanism plays a key role in actual operation. When the correction suggestion in the feedback matrix conflicts with the core constraints of the benchmark template, the system initiates a consultative decision-making process. For example, a correction suggestion is to increase the temperature weight to reduce the error, but the safety regulation constraints of the benchmark template prohibit increasing the weight influence in the high temperature interval. The system finds a feasible solution through a compromise algorithm, moderately adjusts the temperature weight within the allowed range, and enhances the compensation effect of the cooling system parameters. The optimization under such constraints reflects the engineering practicability of the system.

[0115] The system maintenance module ensures the long-term reliability of the feedback mechanism. The diagnostic process, which is executed regularly, checks the health status of each element in the feedback matrix, including the timeliness of historical data, the applicability of association rules, and the reasonableness of weight coefficients. When the diagnostic process finds that a parameter association rule is invalid due to battery process improvement, it automatically starts the rule update process to establish a new association model from the latest data. This self-maintenance function ensures that the system can adapt to the iterative development of battery technology.

[0116] The design of the exception handling process considers various boundary conditions in actual operation. When a sensor failure causes data loss, the system can generate a reasonable substitute value based on the characteristics of adjacent data points and the overall state of the battery pack, and mark the data quality label. For sudden abnormal events such as current surges, the system temporarily freezes the update of the feedback matrix, and decides the processing strategy after confirming the nature of the event. These design details ensure the stable operation of the system under non-ideal conditions.

[0117] Integration with external systems is achieved through standardized interfaces. The output format of the feedback matrix is compatible with common battery management system protocols, supporting collaboration with charging and discharging control devices, thermal management units, and other subsystems. In the example scenario, when the system detects a high temperature accompanied by a high error state, it not only adjusts the internal prediction model parameters, but also sends optimization instructions to the cooling system through a digital interface, forming a cross-system optimization scheme.

[0118] Embodiment 5: Around the battery degradation state quantification model and topology network analysis, it is described in detail how to realize the accurate matching of battery operating state and historical degradation mode through dynamic time warping algorithm. This implementation establishes a complete feature space analysis method, which converts the degradation process of the whole life cycle of the battery into a calculable topology structure, providing deep state recognition ability for the prediction model.

[0119] The system initialization phase constructs the basic framework of the degradation feature space, which organizes historical degradation data in a hierarchical network form. Each node in the network represents a degradation state and contains the complete feature vector set in that state, covering multi-dimensional parameters such as temperature distribution, current characteristics, and voltage response. The connection edges between nodes encode the state transition probability, reflecting the typical path of the battery from one degradation phase to another. During network construction, an adaptive clustering algorithm is used to automatically determine the number and location of nodes, ensuring coverage of various typical degradation modes under various working conditions.

[0120] The data preprocessing module divides the real-time collected operation data into sliding windows, with the window size dynamically adjusted according to the battery's charge and discharge cycle. Each data window is assigned a unique timestamp index to record its position information in the battery life cycle. The window overlap rate is set as a adjustable parameter to balance between data processing accuracy and computational efficiency. The segmented data segments are standardized to eliminate differences caused by different sensor dimensions and sampling frequencies, establishing a unified benchmark for subsequent feature comparison.

[0121] The electrochemical feature extraction process uses a multi-resolution analysis method to characterize the intrinsic properties of data segments from both time and frequency domains. Time domain analysis focuses on parameter trend changes, identifying key turning points and stable intervals; frequency domain analysis reveals periodic fluctuations and abnormal harmonic components. The feature vector construction considers the phase relationship between different parameters, capturing the pattern of coordinated changes in multiple physical quantities. After dimension reduction, the extracted feature vectors form standardized state descriptors, facilitating similarity calculation with the topological network nodes.

[0122] The implementation of the dynamic time warping algorithm is specifically optimized for battery data characteristics. The core of the algorithm is to establish the optimal matching path between the current data segment and the historical node features, allowing flexible alignment of similar patterns on the time axis. The path search uses a dynamic programming method with constraints to limit the maximum time stretching range and prevent forced matching that does not conform to physical meaning. The similarity calculation introduces a composite measure of shape distance and trend distance, considering both the closeness of instantaneous values and the consistency of change direction.

[0123] The construction of the similarity matrix uses an incremental updating strategy, gradually improving as new data arrives. Each element in the matrix records the matching degree of a specific data segment with the topological node, and the matching quality distribution is visually displayed through color coding. When extracting the optimal path from the matrix using the backtracking algorithm, both local matching degree and global continuity are considered to avoid selecting paths that may have good local matching but are not coherent overall. Path smoothing eliminates minor fluctuations caused by measurement noise, highlighting the true degradation trend.

[0124] The generation of model revision commands is based on the quantitative analysis of path deviation degree. The system continuously monitors the deviation of the current optimal path from the reference path, which comes from the typical trajectories of the same degradation stage in historical data. The deviation degree calculation distinguishes between systematic shifts and random fluctuations, the former of which may indicate a substantial change in battery performance, and the latter of which usually reflects temporary disturbances. When the systematic shift exceeds a pre-set threshold, different levels of revision commands are triggered, from parameter tuning to model structure updating, forming a hierarchical response mechanism.

[0125] Dynamic maintenance of the topological network is a key step in the implementation process. The system regularly assesses the impact of new data on the network structure, and when it finds that new degradation patterns cannot be adequately represented by existing nodes, it initiates the network expansion process. The insertion position of new nodes is determined by the density distribution of existing nodes, with sparse areas being prioritized for filling. The node merging mechanism handles similar nodes that become redundant as data accumulates, maintaining the compactness and efficiency of the network. The weights of the connecting edges are dynamically updated based on the latest data, reflecting the probability changes of degradation paths during battery aging.

[0126] Visualization of state recognition results uses multi-dimensional projection technology to map key dimensions of high-dimensional feature space to two-dimensional or three-dimensional display planes. The projection algorithm preserves the topological relationships in the original space, so that similar states remain clustered after projection. The position changes of the current data segment in the projection map form dynamic trajectories, allowing operators to visually observe the migration process of battery states in the feature space. The trajectory playback function supports backtracking analysis, helping to understand the causes and trends of specific states.

[0127] The abnormal processing mechanism covers the entire process of data collection, feature extraction, and state matching. The sensor fault detection module identifies abnormal readings and attempts to compensate for them through adjacent data and model prediction. Robust statistical methods are used to handle outliers in the feature extraction process, reducing their interference with overall analysis. Low-quality data segments are specially marked in the state matching stage to avoid false state determination due to data problems. The system maintains an abnormal case knowledge base to quickly identify and respond to recurring problem patterns.

[0128] Computational efficiency optimization is reflected in multiple aspects. Incremental calculation is used in feature extraction, with only the changed part being processed when new data arrives. Hierarchical indexing technology is applied in topological network search to quickly locate the subset of possible matching nodes. Dynamic time warping implements various acceleration strategies, including early termination, approximate search, and parallel computing. Intelligent caching mechanisms are used in memory management to optimize storage layout based on data access patterns. These optimization measures enable the system to run in real-time on resource-constrained hardware platforms.

[0129] The interface design with the prediction model supports bidirectional data flow. State identification results are passed as key input features to the prediction model, enhancing its understanding of the current battery degradation stage. Simultaneously, the prediction model's output feeds back into the state identification process to help validate and refine the boundaries of the feature space. This tight integration enables the two modules to co-evolve and adapt to changes in battery performance as they age.

[0130] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0131] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the remaining life of a battery in an electrochemical energy storage system, characterized in that: include: Collect multi-dimensional operating data of the target battery pack during the charge and discharge cycle, including temperature distribution data, current ripple data, and voltage decay curve, and generate feature weight parameters based on the matching results of the real-time operating conditions and the historical operation database; Extracting degradation characteristic patterns of the target battery pack over its entire life cycle, the degradation characteristic patterns are generated by analyzing capacity decay trajectories at different charge and discharge rates, and using the degradation characteristic patterns as a reference template for establishing a dynamic degradation feature space; Constructing a multi-scale coupled life prediction model, driving the model parameters to iteratively update in the direction of minimizing the prediction error, generating feature reconstruction instructions for each iteration and updating the physical field coupling parameters of the benchmark template, and constructing a model adaptive evaluation function based on the correlation between feature weight parameters and real-time operation data; triggering cross-cycle data verification based on the amplitude of the feature reconstruction instruction, wherein the cross-cycle data verification extracts a verification data packet matching the current charge and discharge mode from a historical operation database, and performs a working condition compatibility check on the verification data packet using the electrochemical conversion rules in the reference template; A feedback matrix is ​​generated by combining the prediction error index output by the model adaptive evaluation function with the verification result of the verification data packet, and the mapping relationship between the feature weight parameter and the reference template is dynamically adjusted.

2. The method for predicting the remaining life of a battery in an electrochemical energy storage system according to claim 1, characterized in that: The multi-scale coupled life prediction model is constructed to drive the iterative update of model parameters in the direction of minimizing the prediction error, including: The degenerate characteristic pattern is mapped into a high-dimensional tensor, where each dimension corresponds to the adjustable range of the physical field coupling parameter, and the correlation coefficient of the electrochemical parameters is calculated based on the current charge and discharge mode; constructing an error neighborhood table based on the correlation coefficient, recording historical failure cases associated with the current degradation characteristic mode and differences in their physical field coupling parameters, and generating a candidate reconstruction instruction set; A dual-constraint screening mechanism is used to select a target reconstruction instruction from the candidate reconstruction instruction set, wherein the constraint conditions include an error reduction rate of the candidate reconstruction instruction in historical correction and a parameter compatibility threshold with the electrochemical conversion rule; Applying the target reconstruction instruction to the degenerate characteristic pattern, dynamically updating the parameter value of the corresponding dimension in the high-dimensional tensor, and adjusting the screening threshold of the error neighborhood table to trigger the elimination of low-correlation data; The matching degree between the updated degradation feature pattern and the working condition of the verification data package is recalculated, and the change in the matching degree is fed back to the adjustment process of the feature weight parameters.

3. The method for predicting the remaining life of a battery in an electrochemical energy storage system according to claim 2, wherein: The adopting of a dual-constraint screening mechanism to select a target reconstruction instruction from the candidate reconstruction instruction set includes: Count the error reduction rate of each candidate reconstruction instruction within the preset time window, and normalize the ratio of the number of historical successful corrections to the change in the reconstruction instruction priority as the initial screening weight; Extract the parameter constraint fields corresponding to the electrochemical conversion rules in the benchmark template, calculate the coverage ratio between the parameter dimensions corresponding to the candidate reconstruction instructions and the preset constraint fields, and map them into compatibility constraint weights; The initial screening weight and the compatibility constraint weight are dynamically superimposed, and the superposition coefficient is adjusted according to the distribution density of low-correlation data in the error neighborhood table, so that the compatibility constraint weight obtains a higher proportion in the high-risk area of ​​prediction error; The instruction selection operation is performed based on the comprehensive screening probability to verify the compatibility status of the candidate reconstruction instructions with the core parameters of the benchmark template. If there is a conflict, the backup reconstruction instruction queue is enabled.

4. The method for predicting the remaining life of a battery in an electrochemical energy storage system according to claim 3, wherein: The dynamic superposition of the initial screening weight and the compatibility constraint weight includes: Based on the comparison between the proportion of low-correlation data in the error neighborhood table and the historical maximum data capacity, the error risk level is calculated using a multi-segment function; Generate a dynamic superposition factor based on the error risk level, adopt a threshold segmented adjustment mechanism, and combine the historical sliding mean as a stability factor to control the fluctuation range of the superposition factor; The initial screening weight is correlated with the dynamic superposition factor, and the correlation result is integrated with the compatibility constraint weight; The intensity of the stability factor is adjusted according to the degree of deviation between the superposition factor and the historical sliding mean.

5. The method for predicting the remaining life of a battery in an electrochemical energy storage system according to claim 1, wherein: The method of constructing a model adaptive evaluation function based on the correlation between the feature weight parameter and the real-time operation data includes: Based on the distribution characteristics of the feature weight parameters, the spatiotemporal correlation between the parameters of each dimension in the real-time operation data and the battery degradation rate is calculated; Generate a dynamic correction factor based on spatiotemporal correlation, associate the parameter change trends of high attenuation rate areas with those of temperature anomalies, and constrain the scope of the correction factor through the physical field coupling parameters of the benchmark template; The feature weight parameters, dynamic correction factors and spatiotemporal correlation are integrated to construct a model adaptive evaluation function, and a penalty term related to the deviation of the physical field coupling parameters is introduced.

6. The method for predicting the remaining life of a battery in an electrochemical energy storage system according to claim 5, characterized in that: Generating a dynamic correction factor based on the spatiotemporal correlation includes: Based on the spatiotemporal correlation and the inverse correlation between electrochemical parameter fluctuations and capacity decay gradients, the difference in parameter changes between the high decay rate region and the temperature anomaly region is calculated. Generate an initial correction factor based on the difference in parameter changes, associate the parameter sensitivity in the high decay rate region with the electrochemical response delay in the temperature anomaly region, and constrain the rate of change of the correction factor through the physical field coupling parameters of the benchmark template; Based on the historical correction record of the physics coupling parameters, a decay control factor is applied to the initial correction factor.

7. The method for predicting the remaining life of a battery in an electrochemical energy storage system according to claim 1, wherein: The feedback matrix is ​​generated by combining the prediction error index output by the model adaptive evaluation function and the verification data packet verification result, including: Calculate the error and compatibility coupling coefficient based on the prediction error index and the compatibility compliance rate in the verification data packet check result; An initial feedback matrix is ​​generated based on the error and compatibility coupling coefficient, the compatibility loss in high-error areas is associated with feature reconstruction instructions, and the update strength of the feedback matrix is ​​constrained by the historical correction record of the benchmark template. A multi-dimensional feedback model is constructed by integrating the error with the compatibility coupling coefficient, the prediction error index and the compatibility compliance rate.

8. The method for predicting the remaining life of a battery in an electrochemical energy storage system according to claim 1, wherein: Also includes: Establish a quantitative model for battery pack degradation status and analyze the primary and secondary characteristic quantities in multi-dimensional operating data; Construct a topological network of degradation feature space and encode the historical degradation patterns of different charge and discharge stages into topological nodes; Calculate the similarity path between the current running data and the topological nodes through the dynamic time warping algorithm; When the similarity path exceeds the preset threshold, the prediction model parameter reset mechanism is triggered.

9. The method for predicting the remaining life of a battery in an electrochemical energy storage system according to claim 8, characterized in that: The calculation of the similarity path between the current running data and the topological node by the dynamic time warping algorithm includes: Perform sliding window segmentation on the current running data to generate multiple data segments and their timestamp indexes; Extract the electrochemical eigenvector of each data segment and calculate the dynamic bending path corresponding to the topological node eigenvector; Generate a similarity matrix based on the dynamic curved path and locate the optimal matching path through the backtracking algorithm; Generate model correction instructions based on the degree of deviation between the optimal matching path and the preset threshold.

10. A system for predicting the remaining life of a battery in an electrochemical energy storage system, used to implement the method for predicting the remaining life of a battery in an electrochemical energy storage system according to any one of claims 1 to 9, characterized in that: The system comprises: The data acquisition module collects multi-dimensional operating data of the target battery pack during the charge and discharge cycle, and generates feature weight parameters based on the matching results of the real-time operating conditions and the historical operation database; A template construction module extracts degradation characteristic patterns of the target battery pack over its entire life cycle and uses the degradation characteristic patterns as reference templates for establishing a dynamic degradation feature space; The model iteration module builds a multi-scale coupled life prediction model, drives the iterative update of model parameters in the direction of minimizing prediction errors, and constructs a model adaptive evaluation function based on the correlation between feature weight parameters and real-time operation data; The verification trigger module triggers cross-cycle data verification based on the amplitude of the feature reconstruction instruction and verifies the working condition compatibility of the verification data packet using the electrochemical conversion rules in the reference template; The feedback generation module generates a feedback matrix through the prediction error index output by the model adaptive evaluation function and the verification result of the verification data packet, and dynamically adjusts the mapping relationship between the feature weight parameters and the benchmark template.

Images