Health state monitoring and life prediction method and system for energy storage battery pack

Abstract

The invention provides a health state monitoring and service life prediction method and system for an energy storage battery pack, relates to the field of energy storage batteries, and solves the problems that prediction models in the prior art mostly adopt a single machine learning algorithm and lack adaptability to a battery degradation mechanism and actual working conditions, and the prediction efficiency is poor. And the battery health state evaluation and residual life prediction precision is low. The method comprises the following steps: preprocessing an original data set, analyzing a multi-dimensional health feature vector, and constructing a health feature matrix; constructing a health state evaluation model based on the improved CNN-LSTM hybrid model and by fusing battery degradation physical mechanism constraints; inputting the health feature matrix into a health state evaluation model, and outputting a current SOH value; and predicting residual life information based on the current SOH value and the load fluctuation correction coefficient. The method is used in the process of health state evaluation and residual life prediction of the energy storage battery pack.

Description

Technical Field

[0001] This application relates to the field of energy storage batteries, and in particular to a method and system for monitoring the health status and predicting the lifespan of energy storage battery packs. Background Technology

[0002] As core equipment for peak-valley electricity price arbitrage, load shaving and valley filling, and emergency power supply, commercial and industrial energy storage systems typically operate their battery packs under complex conditions of high-rate charging and discharging, long-term cycling, and severe load fluctuations. The state of health (SOH) of the battery pack directly affects the operational safety and reliability of the energy storage system, while the remaining lifetime (RUL) prediction is a key basis for formulating operation and maintenance strategies and reducing operating costs.

[0003] Existing methods for predicting battery health status and remaining life often employ single machine learning algorithms, such as traditional LSTM and support vector machines. These methods lack adaptability to battery degradation mechanisms and actual operating conditions, which affects the accuracy of battery health status assessment and remaining life prediction results.

[0004] Therefore, this application provides a method and system for monitoring the health status and predicting the lifespan of an energy storage battery pack. Summary of the Invention

[0005] This application provides a method and system for monitoring the health status and predicting the lifespan of energy storage battery packs, which solves the technical problem that existing prediction models mostly use a single machine learning algorithm and lack adaptability to battery degradation mechanisms and actual operating conditions, resulting in low accuracy in battery health status assessment and remaining lifespan prediction.

[0006] To achieve the above objectives, this application adopts the following technical solution: Firstly, a method for monitoring the health status and predicting the lifespan of an energy storage battery pack is provided, including: Collect the battery pack's electrical parameters, environmental parameters, and operating condition parameters to construct the original dataset; The original dataset is preprocessed, and multi-dimensional health feature vectors are analyzed to construct a health feature matrix; A health status assessment model is constructed based on an improved CNN-LSTM hybrid model and incorporating constraints from the physical mechanism of battery degradation; the health feature matrix is ​​input into the health status assessment model, and the current SOH value is output. Based on the current SOH value and load fluctuation correction factor, the remaining life information is predicted; the remaining life information includes the actual remaining cycle time and the actual remaining service time; the load fluctuation correction factor is used to quantify the accelerating effect of high load operation on battery aging.

[0007] Based on the above technical solutions, current predictions of the State of Health (SOH) and remaining life of energy storage batteries generally face three major technical bottlenecks: First, relying on a single parameter or simple model makes it difficult to characterize the nonlinear, multi-factor coupled aging behavior of batteries under complex operating environments; second, purely data-driven methods lack physical interpretability and have poor generalization ability when operating conditions change abruptly or data distribution shifts; and third, the failure to effectively integrate the accelerated impact of actual operating strategies (such as high load and peak-valley scheduling) on ​​aging leads to life predictions deviating from engineering reality. To address this, this solution proposes an integrated intelligent assessment method of "multi-source sensing—feature fusion—physical constraints—operating condition correction": A high-quality raw dataset is constructed by simultaneously collecting electrical, environmental, and operating condition parameters, and multi-dimensional health features covering the time domain, frequency domain, entropy value, and operating condition correlation are extracted to comprehensively characterize the battery degradation state; an improved CNN-LSTM hybrid model is adopted to enhance feature extraction capabilities while embedding physical constraints based on the Arrhenius equation and the Peck equation, ensuring that the prediction results conform to the electrochemical aging mechanism; and a load fluctuation correction coefficient is introduced to dynamically quantify the accelerated effect of high load on aging, correcting the theoretical remaining cycle count to the actual service time. This scheme significantly improves the accuracy of SOH assessment and the robustness, interpretability, and engineering applicability of remaining lifetime prediction, providing key technical support for the safe, economical, and long-life operation of energy storage systems.

[0008] In conjunction with the first aspect above, in one possible implementation, the construction of the health feature matrix includes: The time-domain, frequency-domain, and entropy characteristics of the preprocessed electrical and environmental parameters are calculated respectively, and the operating condition correlation characteristics are analyzed based on the operating condition parameters. The time-domain, frequency-domain, and entropy characteristics and the operating condition correlation characteristics are combined within a preset time window to obtain a multi-dimensional health feature vector. Multiple multi-dimensional health feature vectors are stacked in chronological order to obtain a health feature matrix. Among them, the time domain features include voltage fluctuation rate, temperature gradient change rate, internal resistance decay rate, and inflection point offset; the frequency domain features are the peak frequency of voltage or current; the entropy features include the approximate entropy of the internal resistance time series and the sample entropy of the temperature time series; the operating condition correlation features include the cumulative number of charge-discharge cycles, the ratio of high load duration to total daily operating time, and the ratio of charge-discharge duration during peak and valley electricity price periods.

[0009] In conjunction with the first aspect above, in one possible implementation, the voltage fluctuation rate is the ratio of the standard deviation to the mean of the individual voltage series; The rate of change of the temperature gradient is the moving average of the absolute values ​​of the temperature differences between adjacent sampling points in the temperature sequence. The inflection point offset is the position offset of the first derivative minimum point of the state-of-charge-voltage curve relative to the initial state. The peak frequency is the frequency corresponding to the peak value of the power spectral density of the voltage or current signal.

[0010] In conjunction with the first aspect above, in one possible implementation, the construction of the health status assessment model includes: The health feature matrix and the corresponding SOH labels at each time point are integrated into several sets of training data and test data; An improved CNN-LSTM hybrid model is constructed, and a joint loss function is designed. The joint loss function includes a data fitting error term between the model's predicted SOH and the actual SOH, and a physical constraint term constructed based on the Arrhenius equation and the Peck equation. Using the joint loss function as the optimization objective, the improved CNN-LSTM hybrid model is trained end-to-end using training data, and tested using test data. Based on the test results, the improved CNN-LSTM hybrid model is adjusted to obtain a health status assessment model with a health feature matrix as input and a maximum available capacity label as output.

[0011] In conjunction with the first aspect above, in one possible implementation, the improved CNN-LSTM hybrid model includes a CNN feature extraction layer, an LSTM temporal modeling layer, and an attention mechanism layer; The CNN feature extraction layer uses multiple one-dimensional convolution kernels of different sizes to perform convolution operations on the input health feature matrix to extract local key degradation features; The LSTM timing modeling layer contains at least two LSTM units and is used to learn the long-term temporal dependencies in the battery capacity decay process. The attention mechanism layer weights the temporal features of the LSTM output to enhance the contribution weights of key features such as voltage fluctuation rate, internal resistance decay rate, and temperature gradient change rate.

[0012] In conjunction with the first aspect above, in one possible implementation, the physical constraint term includes: The actual degradation rate is obtained by differentiating the SOH sequence output by the health status assessment model with respect to the number of charge-discharge cycles; the physical constraint term is obtained by calculating the square of the difference between the actual degradation rate and the theoretical degradation rate; the theoretical degradation rate is calculated based on the Arrhenius equation and the Peck equation.

[0013] In conjunction with the first aspect above, in one possible implementation, the predicted remaining lifetime information includes: Based on the current SOH in the preset aging range, the corresponding piecewise function is used to fit the battery capacity degradation curve; Based on the fitted degradation curve, calculate the theoretical remaining number of cycles required for the battery to degrade from its current state to a preset scrap threshold under no operating disturbance conditions. Calculate the load fluctuation correction factor K = 1 + × Duration of high load / Total daily runtime; where... To correct the proportional coefficient, high load is an operating state where the discharge rate is greater than or equal to a preset threshold; Calculate the ratio of the theoretical remaining cycle count to the load fluctuation correction factor K to obtain the corrected actual remaining cycle count; calculate the ratio of the corrected actual remaining cycle count to the daily or annual average charge-discharge cycle count of the energy storage system to obtain the actual remaining service time.

[0014] In conjunction with the first aspect above, in one possible implementation, the fitting of the battery capacity degradation curve includes: If the current SOH is greater than or equal to the right end of the preset aging range, a linear function is used to fit the battery capacity degradation curve; if the current SOH is within the preset aging range, an exponential function is used to fit the battery capacity degradation curve; if the current SOH is less than or equal to the left end of the preset aging range, a quadratic polynomial function is used to fit the battery capacity degradation curve.

[0015] In conjunction with the first aspect above, in one possible implementation, the construction of the original dataset includes: The energy storage battery pack's electrical parameters, environmental parameters, and operating condition parameters are collected synchronously through a multi-dimensional sensor array. The electrical parameters include the voltage of each individual battery cell, the total current of the battery pack, the internal resistance of each individual battery cell, and the state of charge. The environmental parameters include the temperature at multiple locations, the ambient humidity, and the operating status of the heat dissipation system. The operating condition parameters include the cumulative number of charge-discharge cycles, the current charge-discharge rate, the charging and discharging duration during peak and off-peak electricity price periods, and the duration of high load. Furthermore, all sensors transmit data to the data processing unit via the CAN bus with a unified timestamp, and the parameters are integrated into structured data samples according to the sampling time to form a raw dataset with time series index.

[0016] Secondly, this application provides a health status monitoring and lifespan prediction system for energy storage battery packs, comprising: a data acquisition module, a health assessment module, and a lifespan prediction module; wherein, the data acquisition module is used to acquire electrical parameters, environmental parameters, and operating condition parameters of the battery pack to construct a raw dataset; the health assessment module is used to preprocess the raw dataset and analyze multi-dimensional health feature vectors to construct a health feature matrix; a health status assessment model is constructed based on an improved CNN-LSTM hybrid model and incorporating constraints from battery degradation physical mechanisms; the health feature matrix is ​​input into the health status assessment model to output the current SOH value; and the lifespan prediction module is used to predict remaining lifespan information based on the current SOH value and a load fluctuation correction coefficient.

[0017] This application provides a method and system for monitoring the health status and predicting the remaining life of energy storage battery packs. It effectively addresses the problems of inaccurate SOH assessment and large deviations in remaining life prediction caused by single data dimensions, lack of physical constraints in the model, and poor adaptability to operating conditions in existing technologies. By simultaneously collecting electrical, environmental, and operating condition parameters from multi-dimensional sensors, a highly consistent original dataset is constructed. This dataset is then integrated with time-domain, frequency-domain, entropy values, and operating condition-related features to comprehensively characterize the battery degradation state. An improved CNN-LSTM hybrid model is employed, incorporating multi-scale convolution, attention mechanisms, and residual connections to significantly enhance feature extraction and temporal modeling capabilities. Simultaneously, physical constraint terms based on the Arrhenius and Peck equations are embedded in the loss function to ensure that the prediction results conform to the electrochemical aging mechanism. Furthermore, a load fluctuation correction coefficient is incorporated to dynamically correct the accelerated aging effect of high-load operation, achieving a precise mapping from theoretical cycle counts to actual service time. This method combines high precision, strong generalization, and engineering deployability, providing reliable support for intelligent operation and maintenance, economic dispatch, and safety early warning of energy storage systems.

[0018] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0019] Figure 1A flowchart illustrating a method for monitoring the health status and predicting the lifespan of an energy storage battery pack, provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of the health status assessment model provided in the embodiments of this application; Figure 3 This is a schematic diagram showing the comparison between the predicted and measured SOH values ​​provided in the embodiments of this application. Figure 4 This is a schematic diagram comparing the RUL prediction results with the actual remaining lifetime provided in the embodiments of this application. Detailed Implementation

[0020] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and 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] The health status monitoring and lifespan prediction method for an energy storage battery pack provided in this application embodiment can be applied to a health status monitoring and lifespan prediction system for an energy storage battery pack. The system includes: a data acquisition module, a health assessment module, and a lifespan prediction module. The data acquisition module is used to collect the battery pack's electrical parameters, environmental parameters, and operating condition parameters to build the raw dataset. The health assessment module is used to preprocess the original dataset, analyze multi-dimensional health feature vectors, and construct a health feature matrix; based on an improved CNN-LSTM hybrid model and incorporating constraints from the physical mechanism of battery degradation, a health status assessment model is constructed; the health feature matrix is ​​input into the health status assessment model, and the current SOH value is output. The lifespan prediction module is used to predict remaining lifespan information based on the current SOH value and load fluctuation correction factor.

[0022] To address the technical problems in existing technologies, this application provides a method for monitoring the health status and predicting the lifespan of energy storage battery packs. The method includes: collecting electrical parameters, environmental parameters, and operating condition parameters of the battery pack to construct a raw dataset; preprocessing the raw dataset and analyzing multi-dimensional health feature vectors to construct a health feature matrix; constructing a health status assessment model based on an improved CNN-LSTM hybrid model and incorporating constraints from battery degradation physical mechanisms; inputting the health feature matrix into the health status assessment model to output the current state of health (SOH) value; and predicting the remaining lifespan based on the current SOH value and a load fluctuation correction coefficient. Based on this, the technical solution effectively solves the core problems of data fragmentation, insufficient feature representation, lack of physical interpretability of the model, and poor adaptability to operating conditions in traditional battery health status (SOH) and remaining lifespan prediction by constructing an integrated intelligent assessment framework of "perception-feature-modeling-prediction". Simultaneous acquisition of three types of parameters—electrical, environmental, and operating conditions—and construction of a time-indexed raw dataset ensures the spatiotemporal consistency of multi-source information. Through systematic preprocessing and extraction of multi-dimensional health features (time domain, frequency domain, entropy value, and operating condition-related features), the degradation signals of batteries, from macroscopic performance to microscopic complexity, are comprehensively captured. Furthermore, an improved CNN-LSTM model incorporating attention mechanisms, multi-scale convolution, and residual structures is employed, with physical constraints based on the Arrhenius and Peck equations embedded in the loss function, enabling the model to possess both strong data fitting capabilities and consistency with electrochemical mechanisms. Combined with current SOH and load fluctuation correction coefficients, the remaining lifetime is dynamically corrected, achieving a precise mapping from "theoretical cycle" to "actual service time." This not only significantly improves prediction accuracy but also possesses strong generalization capabilities and engineering deployability, providing reliable technical support for the safe operation, economical dispatch, and preventative maintenance of energy storage systems.

[0023] like Figure 1 As shown in the embodiment of this application, a method for monitoring the health status and predicting the lifespan of an energy storage battery pack includes: S201. Collect the electrical parameters, environmental parameters, and operating condition parameters of the battery pack to construct the original dataset.

[0024] Among them, the energy storage battery pack's electrical parameters, environmental parameters, and operating condition parameters are collected synchronously through a multi-dimensional sensor array during operation. The electrical parameters include the voltage of each individual battery cell, the total current of the battery pack, the internal resistance of each individual battery cell, and the state of charge. The environmental parameters include the temperature at multiple locations, the ambient humidity, and the operating status of the heat dissipation system. The operating condition parameters include the cumulative number of charge-discharge cycles, the current charge-discharge rate, the charging and discharging duration during peak and off-peak electricity price periods, and the duration of high load. Furthermore, all sensors transmit data to the data processing unit via the CAN bus with a unified timestamp, and the parameters are integrated into structured data samples according to the sampling time to form a raw dataset with time series index.

[0025] It should be noted that the multi-dimensional sensor array includes a distributed voltage acquisition module, a Hall current sensor, an AC internal resistance online monitoring unit, a multi-point temperature probe, a humidity sensor, a heat dissipation system status feedback interface, and operating condition scheduling information from the energy management system (EMS).

[0026] For example, all sensors report data via the CAN bus at a frequency of 10Hz. Each frame of the message contains a unified timestamp distributed by the main controller. The data processing unit aligns the parameters according to the timestamp, aggregates them every 10 seconds, and generates a structured data sample containing electrical parameters, environmental parameters, and operating condition parameters, forming the original dataset with time series index.

[0027] S202. Preprocess the original dataset and analyze the multi-dimensional health feature vectors to construct a health feature matrix.

[0028] The data preprocessing includes removing outlier data using the Grubbs criterion, denoising the data using wavelet transform, and filling in missing data using linear interpolation; the health features include the time-domain features, frequency-domain features, and entropy features of electrical and environmental parameters.

[0029] S203. Based on the improved CNN-LSTM hybrid model and incorporating the constraints of battery degradation physical mechanisms, a health status assessment model is constructed; the health feature matrix is ​​input into the health status assessment model, and the current SOH value is output; SOH = current maximum available capacity / rated capacity × 100%.

[0030] S204. Based on the current SOH value and load fluctuation correction factor, predict the remaining life information; the remaining life information includes the actual remaining cycle time and the actual remaining service time; the load fluctuation correction factor is used to quantify the accelerated effect of high load operation on battery aging.

[0031] Based on the above technical solutions, this application provides a method for monitoring the health status and predicting the lifespan of energy storage battery packs. It constructs a comprehensive, multi-dimensional, and physical information fusion intelligent evaluation system. This system synchronously collects electrical, environmental, and operating parameters via a multi-sensor array and CAN bus, forming a highly consistent raw dataset. High-quality preprocessing is achieved by combining the Grubbs criterion, wavelet denoising, and interpolation, extracting multi-dimensional health features such as time-domain, frequency-domain, and entropy values ​​to comprehensively characterize the battery degradation state. Furthermore, an improved CNN-LSTM model incorporating multi-scale convolution, attention mechanisms, and residual connections is adopted, embedding physical constraints based on the Arrhenius and Peck equations to ensure that the prediction results both fit the data and conform to the electrochemical mechanism. Finally, the remaining lifespan is dynamically corrected by combining segmented degradation modeling and load fluctuation correction coefficients. This solution achieves a closed loop of "high-precision perception - strong characterization modeling - physically reliable prediction", significantly improving the accuracy, robustness and engineering practicality of SOH assessment and life prediction. It solves the problem that existing energy storage battery state of health (SOH) and remaining life prediction methods mostly rely on single parameters or pure data-driven models, which are difficult to accurately reflect the nonlinear and multi-factor coupled aging behavior of batteries under complex operating conditions, resulting in low prediction accuracy and poor generalization ability, and failing to meet the high reliability operation and maintenance requirements of industrial and commercial energy storage systems.

[0032] In one possible implementation of this application embodiment, the above-mentioned S202 can be specifically implemented by the following S301 and S302, which are described in detail below: S301. Calculate the time-domain characteristics, frequency-domain characteristics, and entropy characteristics of the preprocessed electrical and environmental parameters, and analyze the correlation characteristics of the operating conditions based on the operating condition parameters. S302. Combine time-domain features, frequency-domain features, entropy features, and working condition-related features within a preset time window to obtain a multi-dimensional health feature vector. Stack multiple multi-dimensional health feature vectors in chronological order to obtain a health feature matrix. Among them, the time-domain features include voltage fluctuation rate, temperature gradient change rate, internal resistance decay rate, and inflection point offset; the frequency-domain features are the peak frequencies of voltage or current; and the entropy features include the approximate entropy of the internal resistance time series and the sample entropy of the temperature time series.

[0033] The approximate entropy of the internal resistance time series is calculated as follows: For internal resistance time series, based on a pre-defined embedding dimension m and a similarity tolerance of r times the standard deviation, the approximate entropy formula is used. Calculate; where, The mean log similarity for dimension m is the log average of the probability that all vector pairs are "similar" when the embedding dimension is m. Let be the average logarithmic similarity with dimension m+1.

[0034] The sample entropy of the temperature time series is calculated as follows: For temperature time series, under the same parameters, according to the sample entropy formula Calculate; where A and B are the number of (m+1)-dimensional and m-dimensional non-self-matching vector pairs that satisfy the similarity condition, respectively.

[0035] It should be noted that voltage volatility is the ratio of the standard deviation to the mean of a single voltage series.

[0036] The rate of change of the temperature gradient, i.e., the rate at which the battery temperature changes over time, is used to characterize thermal management performance or the risk of localized overheating. Its calculation is based on the temperature difference between adjacent sampling points and the time interval, and noise is smoothed using a moving average to improve robustness.

[0037] The inflection point offset is the position of the minimum point of the first derivative of the state-of-charge-voltage curve relative to the initial state; for example, if the inflection point of a new battery is at SOC=35%, and the current SOC=42%, then the offset is +7%SOC.

[0038] The peak frequency is the frequency corresponding to the peak value of the power spectral density of a voltage or current signal.

[0039] Operating condition-related characteristics include the cumulative number of charge-discharge cycles, the ratio of high load duration to total daily operating time, and the ratio of charge-discharge duration during peak and off-peak electricity price periods.

[0040] For example, suppose the preset time window is 1 day; 18-dimensional features are extracted each day (4 time domain + 1 frequency domain + 2 entropy values ​​+ 3 operating conditions + others); continuous monitoring is performed for 30 days. Then, an 18-dimensional health feature vector is obtained each day; the 30 days are stacked to form a 30×18 health feature matrix; this health feature matrix is ​​the input to the SOH assessment model.

[0041] Based on the above technical solutions, in commercial and industrial energy storage systems, battery aging is affected by multiple factors coupled together, including electrochemistry, thermodynamics, and complex operating strategies. Traditional single parameters (such as capacity or internal resistance alone) are insufficient to comprehensively characterize its health status, resulting in low accuracy and poor generalization ability in remaining lifetime prediction. To address this, this technical solution proposes constructing a multi-dimensional health feature matrix: on the one hand, it extracts time-domain features (reflecting macroscopic degradation trends), frequency-domain features (capturing dynamic response anomalies), and entropy features (quantifying sequence complexity and sensitive to early, subtle degradation) from preprocessed electrical and environmental parameters; on the other hand, it integrates operating condition-related features, incorporating actual usage patterns such as cumulative cycles, high load ratios, and peak-valley scheduling into the evaluation system. By integrating four types of features within the same time window to form a high-dimensional vector and stacking them over time to form a health feature matrix, it retains temporal evolution information and achieves a deep integration of "data-driven + operating condition awareness." This solution significantly improves the accuracy and robustness of SOH assessment, and is particularly suitable for energy storage scenarios with varying operating conditions, multiple temperatures, and high scheduling complexity, providing a reliable basis for precise operation and maintenance and lifespan management.

[0042] In one possible implementation of the embodiments of this application, such as Figure 2 As shown, the above S203 can be specifically implemented through the following S401, S402 and S403, which are explained in detail below: S401. Integrate the health feature matrix and the corresponding SOH labels at each time point into several sets of training data and test data; S402. Construct an improved CNN-LSTM hybrid model and design a joint loss function; The joint loss function includes a data fitting error term between the model's predicted SOH and the actual SOH, and a physical constraint term constructed based on the Arrhenius equation and the Peck equation.

[0043] The physical constraint term is derived by differentiating the SOH sequence output by the health status assessment model with respect to the number of charge-discharge cycles to obtain the actual degradation rate. The square of the difference between the actual degradation rate and the theoretical degradation rate is then calculated to obtain the physical constraint term. The theoretical degradation rate is calculated based on the Arrhenius equation and the Peck equation. Where A is the pre-factor, Let R be the activation energy, R be the gas constant, and T be the average temperature. The cumulative number of charge-discharge cycles is represented by b, which is the aging index.

[0044] S403. Using the joint loss function as the optimization objective, the improved CNN-LSTM hybrid model is trained end-to-end using training data, and the improved CNN-LSTM hybrid model is tested using test data. Based on the test results, the improved CNN-LSTM hybrid model is adjusted to obtain a health status assessment model with a health feature matrix as input and a maximum available capacity label as output.

[0045] It should be noted that the traditional CNN-LSTM model is composed of two common deep learning architectures: CNN (Convolutional Neural Network) and LSTM (Long Short-Term Memory Network), each adept at handling different types of data. CNN excels at capturing spatial local correlations in input data and is widely used in fields such as image recognition. LSTM is used to process time series data, effectively capturing long-term dependencies and is suitable for natural language processing, time series prediction, and other applications.

[0046] In battery state of health (SOH) assessment, traditional CNN-LSTM hybrid models typically employ the following structure: an input layer for extracting input features (such as voltage, current, and temperature); a CNN feature extraction layer for extracting local features; an LSTM layer that takes the features extracted by the CNN as input and further learns the patterns of these features changing over time to predict future state of health or remaining lifetime (RUL); and an output layer for outputting the prediction results.

[0047] However, this infrastructure has some limitations, as follows: Limited ability to represent complex patterns: For complex degradation patterns, simple CNNs may not be able to fully extract all the useful information.

[0048] Lack of attention to key features: In battery health status assessment, some features (such as internal resistance growth and voltage fluctuations) may be more important than others, but traditional models do not have a mechanism to automatically increase the weight of these key features.

[0049] Insufficient generalization performance: The model's performance may decline, especially when faced with different operating conditions (such as high and low temperature environments and changes in charge and discharge rates).

[0050] To address the aforementioned issues, the improved CNN-LSTM hybrid model incorporates the following optimizations: The CNN feature extraction layer employs multi-scale convolutional kernels, introducing convolutional kernels of different sizes (e.g., 3D). 1,5 1,7 1) To capture local patterns in the feature matrix more comprehensively and improve the model's adaptability to complex degradation patterns.

[0051] Attention mechanism: An attention layer was added, which allows the model to dynamically adjust the attention given to each feature, especially emphasizing those features that have a greater impact on health status, such as changes in internal resistance under high load conditions.

[0052] In summary, the improved CNN-LSTM hybrid model enhances the model's ability to understand complex degradation patterns, improves the recognition accuracy of key features, and ensures the physical consistency of prediction results by making the above improvements to the traditional architecture, thereby achieving a more accurate and reliable assessment of battery health status.

[0053] Based on the above technical solutions, battery state of health (SOH) assessment in industrial and commercial energy storage systems faces challenges such as high-dimensional data, nonlinear degradation, and variable operating conditions. While traditional CNN-LSTM models can integrate spatial and temporal features, they suffer from weak representation capabilities, neglect of key features, and poor generalization, making it difficult to meet the requirements of high-precision and robust lifespan management. To address this, this solution proposes an improved CNN-LSTM hybrid model: it enhances the ability to extract local features from complex degradation patterns by introducing multi-scale convolutional kernels; it embeds an attention mechanism to dynamically focus on sensitive indicators such as internal resistance and voltage fluctuations; and it uses residual connections to improve the training stability of deep networks. More importantly, a joint loss function is designed to minimize the error between predicted and measured capacity while embedding physical constraint terms based on the Arrhenius and Peck equations, forcing the model output to conform to the degradation trajectory of electrochemical aging mechanisms. This technical solution not only improves the model's data fitting ability, but also significantly enhances its generalization performance and interpretability under different temperatures, charge / discharge rates and scheduling strategies through a dual-drive approach of "data-driven + physical priors", providing a highly reliable and deployable SOH assessment tool for energy storage systems.

[0054] In one possible implementation of this application embodiment, the above-mentioned S204 can be specifically implemented by the following S501, S502, S503 and S504, which are described in detail below: S501. Based on the preset aging range where the current SOH is located, the corresponding piecewise function is used to fit the battery capacity degradation curve; If the current SOH is greater than or equal to the right end of the preset aging range, a linear function is used to fit the battery capacity degradation curve; if the current SOH is within the preset aging range, an exponential function is used to fit the battery capacity degradation curve; if the current SOH is less than or equal to the left end of the preset aging range, a quadratic polynomial function is used to fit the battery capacity degradation curve.

[0055] For example, based on the current aging stage range of SOH [60%, 80%], the corresponding function is selected to fit the degradation curve: If SOH ≥ 80% → adopt a linear function; if 60% < SOH < 80% → adopt an exponential function; if SOH ≤ 60% → adopt a quadratic polynomial.

[0056] S502. Calculate the theoretical remaining cycle count required for the battery to decay from the current state to the preset scrapping threshold under the condition of no operating condition disturbance according to the obtained degradation curve.

[0057] It should be noted that it is assumed that the battery operates under standard constant operating conditions (e.g., constant current 0.5C charge and discharge, 25°C environment, no high-rate impact); the degradation curve under this condition only reflects the natural aging trend and does not include the acceleration factors in actual operation; the purpose is to first obtain a benchmark prediction value (theoretical remaining cycle count), and then introduce the influence of actual operating conditions through the load fluctuation correction coefficient K in S503.

[0058] S503. Introduce the load fluctuation correction coefficient K = 1 + × high-load duration / total daily operating duration; where is the correction proportionality coefficient, an empirical coefficient calibrated according to the battery chemical system and historical operating data, usually taking a value of 0.1; high load is the operating state where the discharge rate is greater than or equal to the preset threshold.

[0059] S504. Calculate the ratio of the theoretical remaining cycle count to the load fluctuation correction coefficient K to obtain the corrected actual remaining cycle count; calculate the ratio of the corrected actual remaining cycle count to the daily or annual average charge and discharge cycle count of the energy storage system to obtain the actual remaining service time.

[0060] It should be noted that applying K to the degradation rate is mathematically equivalent to dividing the theoretical remaining cycle count by K, because if the degradation rate increases by K times, the number of cycles required to reach the same capacity decay is reduced to 1 / K of the original.

[0061] Based on the above technical solutions, in commercial and industrial energy storage systems, relying solely on historical data extrapolation or fixed degradation models for battery remaining life prediction makes it difficult to consider the differences in degradation mechanisms at different aging stages and the impact of complex actual operating conditions, easily leading to prediction results deviating from the actual service state. Therefore, this solution proposes a remaining life prediction method combining piecewise function fitting and operating condition correction: Based on the current aging stage of SOH (initial stability, mid-term acceleration, and final collapse), linear, exponential, or quadratic polynomial functions are used to fit the degradation curve, accurately characterizing the capacity decay characteristics of each stage; the theoretical remaining cycle count is calculated under standard operating conditions as a benchmark, and a load fluctuation correction coefficient is introduced to quantify the accelerating effect of high-load operation on aging; the actual remaining cycle count is obtained by dividing the theoretical cycle count by the load fluctuation correction coefficient, and then converted into maintainable service time. This solution respects the electrochemical aging laws of batteries while integrating actual operating strategies, achieving synergistic optimization of "mechanism modeling + operating condition perception," significantly improving the accuracy, adaptability, and engineering practicality of remaining life prediction, and providing a reliable decision-making basis for the economic dispatch and preventative maintenance of energy storage systems.

[0062] For example, taking a 1MWh industrial and commercial energy storage system in an industrial park as an example, the battery pack consists of 20 modules connected in series. Each module contains 16 individual cells with a rated capacity of 280Ah and a rated voltage of 3.2V.

[0063] A multi-dimensional sensor array is constructed according to step S201 of this invention: a voltage sensor (accuracy ±0.01V) is installed at the positive and negative terminals of each individual battery cell, and a current sensor (range 0-500A, accuracy ±1%) is installed in the main circuit of the battery pack; two temperature sensors are arranged inside each module, and four temperature sensors (accuracy ±0.5℃) are arranged in the air duct of the energy storage cabinet; an AC internal resistance tester (accuracy ±0.1mΩ) is used to collect the internal resistance of the individual battery cells; a humidity sensor (range 0-95%RH, accuracy ±3%RH) is installed inside the energy storage cabinet; and a working condition recorder records parameters such as the number of charge-discharge cycles and the charge-discharge rate.

[0064] Data preprocessing and feature extraction: 100 days of operational data were collected. Twelve abnormal voltage data were removed using the Grubbs criterion. After wavelet transform denoising, the signal-to-noise ratio of temperature data was improved by 30%. Linear interpolation was used to complete eight missing internal resistance data. Eighteen health features were extracted, with the average voltage fluctuation rate being 0.02 and the average internal resistance approximate entropy being 1.2.

[0065] SOH Evaluation: The feature matrix was input into the improved CNN-LSTM hybrid model, which was trained for 200 iterations with a learning rate of 0.001. Testing showed that the average error between the SOH predictions and the measured values ​​was 2.8%, lower than the 6.5% error of existing single-model approaches. The evaluation data performance is as follows: Figure 3As shown.

[0066] RUL Prediction: The current battery pack has completed 300 cycles, with a measured SOH of 92%. Based on the degradation curve and a load fluctuation correction factor K=1.08 (average high load duration of 2 hours per day, total daily operating time of 12 hours), the remaining cycle count is predicted to be 1700 cycles, with a remaining service life of approximately 4.7 years (calculated based on an average of 360 cycles per year). The evaluation data is as follows... Figure 4 As shown.

[0067] Dynamic calibration: After 10 cycles, the capacity calibration test showed that the measured SOH value was 91.5%, the model prediction value was 91.7%, and the deviation was 0.2%, so no parameter adjustment was required; after 50 cycles, the deviation was 3.2%, and the model parameters were updated by gradient descent, and the deviation was reduced to 2.1% after calibration.

[0068] Some of the data in the above formula are calculated by removing dimensions and taking their numerical values. The formula is the closest to the real situation obtained by software simulation of a large amount of collected data. The preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained through simulation of a large amount of data.

Claims

1. A method for monitoring the health status and predicting the lifespan of an energy storage battery pack, characterized in that, include: Collect the battery pack's electrical parameters, environmental parameters, and operating condition parameters to construct the original dataset; The original dataset is preprocessed, and multi-dimensional health feature vectors are analyzed to construct a health feature matrix; A health status assessment model is constructed based on an improved CNN-LSTM hybrid model and incorporating constraints from the physical mechanism of battery degradation; the health feature matrix is ​​input into the health status assessment model, and the current SOH value is output. Based on the current SOH value and load fluctuation correction factor, the remaining service life information is predicted; the remaining service life information includes the actual remaining number of cycles and the actual remaining service time. The load fluctuation correction coefficient is used to quantify the accelerating effect of high-load operation on battery aging.

2. The method for monitoring the health status and predicting the lifespan of an energy storage battery pack according to claim 1, characterized in that, The construction of the health feature matrix includes: The time-domain, frequency-domain, and entropy characteristics of the preprocessed electrical and environmental parameters are calculated respectively, and the operating condition correlation characteristics are analyzed based on the operating condition parameters. The time-domain, frequency-domain, and entropy characteristics and the operating condition correlation characteristics are combined within a preset time window to obtain a multi-dimensional health feature vector. Multiple multi-dimensional health feature vectors are stacked in chronological order to obtain a health feature matrix. Among them, the time domain features include voltage fluctuation rate, temperature gradient change rate, internal resistance decay rate, and inflection point offset; the frequency domain features are the peak frequency of voltage or current; the entropy features include the approximate entropy of the internal resistance time series and the sample entropy of the temperature time series; the operating condition correlation features include the cumulative number of charge-discharge cycles, the ratio of high load duration to total daily operating time, and the ratio of charge-discharge duration during peak and valley electricity price periods.

3. The method for monitoring the health status and predicting the lifespan of an energy storage battery pack according to claim 2, characterized in that, The voltage fluctuation rate is the ratio of the standard deviation to the mean of the individual voltage series. The rate of change of the temperature gradient is the moving average of the absolute values ​​of the temperature differences between adjacent sampling points in the temperature sequence. The inflection point offset is the position offset of the first derivative minimum point of the state-of-charge-voltage curve relative to the initial state. The peak frequency is the frequency corresponding to the peak value of the power spectral density of the voltage or current signal.

4. The method for monitoring the health status and predicting the lifespan of an energy storage battery pack according to claim 1, characterized in that, The construction of the health status assessment model includes: The health feature matrix and the corresponding SOH labels at each time point are integrated into several sets of training data and test data; An improved CNN-LSTM hybrid model is constructed, and a joint loss function is designed. The joint loss function includes a data fitting error term between the model's predicted SOH and the actual SOH, and a physical constraint term constructed based on the Arrhenius equation and the Peck equation. Using the joint loss function as the optimization objective, the improved CNN-LSTM hybrid model is trained end-to-end using training data, and tested using test data. Based on the test results, the improved CNN-LSTM hybrid model is adjusted to obtain a health status assessment model with a health feature matrix as input and a maximum available capacity label as output.

5. The method for monitoring the health status and predicting the lifespan of an energy storage battery pack according to claim 1, characterized in that, The improved CNN-LSTM hybrid model includes a CNN feature extraction layer, an LSTM temporal modeling layer, and an attention mechanism layer; The CNN feature extraction layer uses multiple one-dimensional convolution kernels of different sizes to perform convolution operations on the input health feature matrix to extract local key degradation features; The LSTM timing modeling layer contains at least two LSTM units and is used to learn the long-term temporal dependencies in the battery capacity decay process. The attention mechanism layer weights the temporal features of the LSTM output to enhance the contribution weights of key features such as voltage fluctuation rate, internal resistance decay rate, and temperature gradient change rate.

6. The method for monitoring the health status and predicting the lifespan of an energy storage battery pack according to claim 1, characterized in that, The physical constraints include: The actual degradation rate is obtained by differentiating the SOH sequence output by the health status assessment model with respect to the number of charge-discharge cycles; the physical constraint term is obtained by calculating the square of the difference between the actual degradation rate and the theoretical degradation rate; the theoretical degradation rate is calculated based on the Arrhenius equation and the Peck equation.

7. The method for monitoring the health status and predicting the lifespan of an energy storage battery pack according to claim 1, characterized in that, The predicted remaining lifespan information includes: Based on the current SOH in the preset aging range, the corresponding piecewise function is used to fit the battery capacity degradation curve; Based on the fitted degradation curve, calculate the theoretical remaining number of cycles required for the battery to degrade from its current state to a preset scrap threshold under no operating disturbance conditions. Calculate the load fluctuation correction factor K = 1 + × Duration of high load / Total daily runtime; where... To correct the proportional coefficient, high load is an operating state where the discharge rate is greater than or equal to a preset threshold; Calculate the ratio of the theoretical remaining cycle count to the load fluctuation correction factor K to obtain the corrected actual remaining cycle count; calculate the ratio of the corrected actual remaining cycle count to the daily or annual average charge-discharge cycle count of the energy storage system to obtain the actual remaining service time.

8. The method for monitoring the health status and predicting the lifespan of an energy storage battery pack according to claim 7, characterized in that, The fitted battery capacity degradation curve includes: If the current SOH is greater than or equal to the right end of the preset aging range, a linear function is used to fit the battery capacity degradation curve; if the current SOH is within the preset aging range, an exponential function is used to fit the battery capacity degradation curve; if the current SOH is less than or equal to the left end of the preset aging range, a quadratic polynomial function is used to fit the battery capacity degradation curve.

9. The method for monitoring the health status and predicting the lifespan of an energy storage battery pack according to claim 1, characterized in that, The construction of the original dataset includes: The energy storage battery pack's electrical parameters, environmental parameters, and operating condition parameters are collected synchronously through a multi-dimensional sensor array. The electrical parameters include the voltage of each individual battery cell, the total current of the battery pack, the internal resistance of each individual battery cell, and the state of charge. The environmental parameters include the temperature at multiple locations, the ambient humidity, and the operating status of the heat dissipation system. The operating condition parameters include the cumulative number of charge-discharge cycles, the current charge-discharge rate, the charging and discharging duration during peak and off-peak electricity price periods, and the duration of high load. Furthermore, all sensors transmit data to the data processing unit via the CAN bus with a unified timestamp, and the parameters are integrated into structured data samples according to the sampling time to form a raw dataset with time series index.

10. A health status monitoring and lifespan prediction system for an energy storage battery pack, operating based on the health status monitoring and lifespan prediction method for an energy storage battery pack according to any one of claims 1-9, characterized in that, It includes a data acquisition module, a health assessment module, and a lifespan prediction module; The acquisition module is used to collect the battery pack's electrical parameters, environmental parameters, and operating condition parameters to construct the original dataset; The health assessment module is used to preprocess the original dataset, analyze multi-dimensional health feature vectors, and construct a health feature matrix. A health status assessment model is constructed based on an improved CNN-LSTM hybrid model and incorporating constraints from the physical mechanism of battery degradation; the health feature matrix is ​​input into the health status assessment model, and the current SOH value is output. The lifespan prediction module is used to predict remaining lifespan information based on the current SOH value and load fluctuation correction coefficient.

Images