A method and system for predicting battery energy storage lifetime based on multiple time scales

Abstract

This invention relates to the field of predictive model technology, specifically to a method and system for predicting battery storage lifespan based on multiple time scales. The method includes the following steps: deploying three sets of parallel data windows (short-term, medium-term, and long-term) to acquire battery capacity and internal resistance parameters; initializing parameters using a sliding window; synchronously calibrating the baseline; correcting parameters using Kalman filtering; extracting the magnitude and rate of change; outputting the mean capacity and variance of internal resistance; calculating the residual sequence; triggering anomalies by three consecutive internal resistance-related thresholds when the capacity residual exceeds a threshold; weighted fusion after anomalies; and outputting the remaining lifespan. In this invention, dynamic monitoring using parallel data windows across multiple time scales, synchronous baseline calibration using a sliding window, and dynamic parameter correction using Kalman filtering establish a linkage judgment mechanism between capacity and internal resistance residuals. A dual-attenuation coupling model integrates multi-dimensional attenuation features, enabling multi-scale collaborative analysis of battery aging. This overcomes the limitations of linear models, enhances the sensitivity and robustness of anomaly detection, improves the accuracy and timeliness of lifespan prediction, and reduces the risk of misjudgment.

Description

Technical Field

[0001] This invention relates to the field of prediction model technology, and in particular to a method and system for predicting battery energy storage lifetime based on multiple time scales. Background Technology

[0002] The field of predictive modeling technology involves using mathematical modeling and data analysis to predict system behavior trends and extrapolate outcomes. Its core lies in constructing a quantitative relationship network between variables and establishing a dynamic evolution mechanism. This field encompasses four technical levels: data feature extraction, model architecture design, parameter optimization strategies, and verification and evaluation systems. It establishes input-output mapping relationships through training with historical data to extrapolate future states. Traditional battery energy storage life prediction refers to a technical solution based on charge-discharge cycle records and voltage and current parameter analysis. It employs capacity decay curve fitting and empirical formula derivation, monitors the rate of change of battery internal resistance and the decreasing trend of energy density, and establishes a linear regression model using temperature stress accelerated aging experimental data. Finally, it uses cycle counting and capacity retention rate threshold determination to estimate remaining lifespan.

[0003] Traditional technologies rely on fitting capacity decay curves within a single time window, failing to distinguish decay characteristics at different time scales during battery aging. Linear regression models struggle to capture dynamic nonlinear decay patterns, and empirical formula derivations are limited by fixed parameter assumptions, resulting in insufficient correlation analysis between early capacity fluctuations and the mid-term accelerated aging stage. Static capacity retention threshold determination lacks a dynamic response mechanism to multi-parameter coupling effects, the co-evolutionary relationship between internal resistance changes and capacity decay is not effectively quantified, abnormal state detection is susceptible to short-term noise interference, and the time-varying characteristics of accelerated aging experimental data and real operating conditions are poorly matched, causing lag or deviation in remaining lifetime prediction under complex charge and discharge scenarios. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and to propose a method and system for predicting battery energy storage lifetime based on multiple time scales.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for predicting battery energy storage lifetime based on multiple time scales, comprising the following steps:

[0006] S1: Deploy three sets of parallel data windows for short-term, medium-term and long-term to obtain capacity and internal resistance parameters during battery operation. Use a sliding window algorithm to initialize the window length parameter and synchronize the current cycle capacity baseline value and internal resistance baseline value with the window configuration.

[0007] S2: Call the Kalman filter algorithm to correct the capacity baseline value and the internal resistance baseline value, extract the capacity variation amplitude parameter and the internal resistance change rate parameter based on the window length parameter, and output the corrected capacity mean and internal resistance variance.

[0008] S3: The sliding window algorithm is used to extract data within the window, and the residual between the mean capacity and the measured value is calculated to generate a capacity residual sequence. The residual between the variance of the internal resistance and the measured value is calculated simultaneously to generate an internal resistance residual sequence. A capacity fluctuation threshold is set based on the capacity variation amplitude parameter. When the capacity residual sequence exceeds the threshold three times in a row, the internal resistance residual sequence is linked to trigger an anomaly judgment.

[0009] S4: Based on the anomaly determination, call the weighted fusion function, input the capacity change magnitude parameter and the internal resistance change rate parameter into the dual attenuation coupling model, and output the remaining battery life.

[0010] As a further embodiment of the present invention, the window length parameter specifically includes a short-term window length parameter, a medium-term window length parameter, and a long-term window length parameter; the corrected mean capacity and internal resistance variance include a capacity filter correction value, an internal resistance dynamic variance value, a benchmark synchronization value, and a rate correlation value; the capacity fluctuation threshold includes a consecutive over-limit number threshold, a residual linkage condition threshold, and an abnormal triggering mechanism threshold; and the remaining battery life specifically refers to the capacity decay coefficient and the internal resistance coupling factor.

[0011] The three sets of parallel data windows establish a data pipeline through the benchmark synchronization value, and use a moving average algorithm to keep the parameters synchronized.

[0012] The dual-attenuation coupling model includes a capacity-internal resistance cross-validation structure.

[0013] As a further aspect of the present invention, the sliding window algorithm employs a dynamic adjustment mechanism when initializing the window length parameter: when the capacity change range parameter... With the internal resistance change rate parameter satisfy Automatically extend window length to ;

[0014] in, This represents the current cycle capacity change (unit: Ah / cycle). The average value representing the historical capacity variation (unit: Ah / cycle). Represents the rate of change of internal resistance in the current cycle (unit: mΩ / cycle). The mean value representing the rate of change of internal resistance throughout history (unit: mΩ / cycle). This represents the dimensionless shape parameter threshold determined by three times the standard deviation of historical data. Represents the initial window baseline length (unit: cycle). Representative capacity change influencing factors This represents the influencing factor of changes in internal resistance.

[0015] As a further aspect of the present invention, the Kalman filter algorithm correction process employs a two-state coupled equation: state equation ;

[0016] Observation equations ;

[0017] in, The joint state vector at time t (dimension: 2×1) contains the capacity filter correction value (unit: Ah) and the internal resistance dynamic variance value (unit: mΩ²). The diagonal elements represent the state transition matrix (dimension: 2×2). This represents the capacity decay inertia factor (characterizing the memory effect properties of capacity changes). The inertia factor representing the change in internal resistance (characterizing the hysteresis effect of the change in internal resistance). The control matrix (dimension: 2×1) has element values ​​determined by the ratio of rate correlation values ​​to baseline synchronization values. The capacity-internal resistance coupling coefficient is calculated by relating the window length parameter to historical data. Represents the process noise vector (unit: Ah², mΩ) 4 ), Represents the observation vector (dimension: 2×1). Represents the observation matrix (dimension: 2×2). Represents the observation noise vector (unit: Ah², mΩ) 4 ).

[0018] As a further aspect of the present invention, the dynamic adjustment method for the capacity fluctuation threshold includes: when the capacity residual sequence of three consecutive periods satisfies At that time, the residual linkage condition threshold is adjusted to ;

[0019] in, Represents the capacity residual at time t (unit: Ah). The standard deviation of the capacity residual sequence (unit: Ah). This represents the mean of the capacity residual sequence (unit: Ah). Represents the sensitivity coefficient. This represents the threshold value for the initial residual linkage condition. This represents the threshold adjustment factor, whose value satisfies... , The sliding standard deviation (unit: mΩ² / cycle) represents the variance of internal resistance, and the calculation window length is 10 cycles.

[0020] As a further aspect of the present invention, the dual-attenuation coupling model adopts a layered fusion structure: the first layer calculates the capacity attenuation-dominant lifetime. ;

[0021] The second layer calculates the dominant lifetime based on internal resistance decay. ;

[0022] The final lifetime weighted value ;

[0023] in, This represents the current measured battery capacity (unit: Ah). Represents the capacity lifetime end threshold (unit: Ah). This represents the capacity degradation correction coefficient obtained through regression analysis of battery aging experiments. This represents the current cycle capacity change (unit: Ah / cycle). This represents the current measured value of the battery's internal resistance (unit: mΩ). This represents the internal resistance lifetime termination threshold (unit: mΩ). This represents the internal resistance degradation correction coefficient obtained through regression analysis of battery aging experiments. Represents the rate of change of internal resistance in the current cycle (unit: mΩ / cycle). The representative weighting coefficient (range: 0.5-0.8) is determined by the severity level of the anomaly assessment result.

[0024] As a further aspect of the present invention, the anomaly determination result classification method includes: when the number of consecutive out-of-limit occurrences of the capacity residual sequence reaches 120% of the threshold of the anomaly triggering mechanism, the anomaly level is automatically increased and a second-order Kalman filter is activated.

[0025] The second-order Kalman filter adds an inertia correction term to the state equation. ,in, This represents the inertia correction factor, whose value is negatively correlated with the fluctuation amplitude of the internal resistance dynamic variance. The fluctuation amplitude is taken as the moving standard deviation of the internal resistance variance (unit: mΩ² / cycle), and the calculation window length is 10 cycles. Represents the joint state vector at time t-1. This represents the joint state vector at time t-2.

[0026] As a further aspect of the present invention, the weighted fusion function adopts an adaptive adjustment strategy: when the correlation coefficient between the capacity variation magnitude parameter and the internal resistance change rate parameter is... When using the product fusion method ;

[0027] when When using a linear weighting method ;

[0028] when When the maximum selection method is used. ;

[0029] in, The Pearson correlation coefficient, weighted by window length, Represents the capacity change weighting coefficient. The weighting coefficient representing the change in internal resistance is obtained through regression analysis of the window length parameter and the historical decay rate.

[0030] As a further aspect of the present invention, the collaborative mechanism of the three sets of parallel data windows includes: the long-term window outputs the baseline parameters to the medium-term window as initialization conditions;

[0031] The intermediate window outputs trend parameters to the short-term window as constraint boundaries;

[0032] The real-time correction parameters output by the short-term window are fed back to the long-term window for parameter calibration;

[0033] A data pipeline is established between the three windows using the aforementioned baseline synchronization value. An exponentially weighted moving average algorithm is employed to maintain parameter synchronization. The exponential weighting coefficient is dynamically adjusted based on the window length parameter, and the formula for calculating the weighting coefficient is as follows: ,in This represents the current window length parameter value.

[0034] A battery storage lifetime prediction system based on multiple time scales is provided. The system is used to implement the above-mentioned battery storage lifetime prediction method based on multiple time scales. The system includes: a multi-scale data acquisition module, which is used to deploy the three sets of parallel data windows and execute the sliding window algorithm.

[0035] The joint filter correction module is used to run the Kalman filter algorithm and generate the corrected mean capacity and variance of internal resistance.

[0036] The residual collaborative analysis module is used to calculate the capacity residual sequence and the internal resistance residual sequence, and to perform the anomaly determination.

[0037] The lifetime fusion prediction module is used to implement the calculation process of the weighted fusion function and the dual-attenuation coupling model;

[0038] The dynamic configuration management module is used to adjust the window length parameter and capacity fluctuation threshold in real time based on the anomaly determination result;

[0039] The output of the multi-scale data acquisition module is connected to the input of the joint filtering correction module;

[0040] The output of the joint filtering correction module is connected to the residual collaborative analysis module and the dynamic configuration management module, respectively.

[0041] The output of the residual collaborative analysis module is connected to the input of the weighting function of the lifetime fusion prediction module.

[0042] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0043] In this invention, dynamic monitoring is achieved by deploying parallel data windows across multiple time scales. A sliding window algorithm is used to simultaneously calibrate the capacity baseline value and the internal resistance baseline value. Combined with the Kalman filter algorithm, key parameters are dynamically corrected to eliminate errors caused by fluctuations in data across a single time dimension. A linkage judgment mechanism between the capacity residual sequence and the internal resistance residual sequence is established. By fusing multi-dimensional decay characteristics through a dual-decay coupling model, multi-scale collaborative analysis of the battery aging process is achieved. This overcomes the limitations of traditional linear regression models in representing dynamic nonlinear decay processes, enhances the sensitivity and robustness of abnormal state detection, solves the problem of insufficient adaptability of static threshold judgment under complex operating conditions, improves the timeliness and accuracy of remaining life prediction, and reduces the risk of misjudgment caused by sudden changes in local parameters or noise interference. Attached Figure Description

[0044] Figure 1 This is a flowchart of the battery energy storage lifetime prediction method based on multiple time scales of the present invention.

[0045] Figure 2 This is a flowchart of the data window deployment and parameter synchronization calibration process of the present invention;

[0046] Figure 3 This is a flowchart of the Kalman filter correction parameters of the present invention;

[0047] Figure 4 This is a flowchart of the anomaly detection and dynamic adjustment process of the present invention;

[0048] Figure 5 This is a flowchart of the remaining service life prediction and fusion process of the present invention. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this invention clearer, the software-based technical solution is described in detail below with reference to system architecture diagrams and embodiments. It should be understood that the specific embodiments described herein are only for explaining the technical solutions of this invention and do not constitute a limitation on the scope of protection.

[0050] In the description of this invention, the system architecture relationships or data processing flows indicated by terms such as "layer," "module," "interface," "data flow," "client," and "server" are all defined based on the architecture diagram or flowchart corresponding to the embodiments. This way of describing is only used to clearly illustrate the logical relationships between the elements in the technical solution, and not to limit the physical deployment form. The term "multiple" includes two or more technical units, including but not limited to multiple data nodes, processing threads, service instances, or functional components and other scalable elements. The specific number is determined according to the actual business scenario and needs to be specifically specified.

[0051] Please see Figure 1 and Figure 2 This invention provides a technical solution: a battery energy storage lifetime prediction method based on multiple time scales, comprising the following steps:

[0052] S1: By deploying three sets of parallel data windows (short-term, medium-term, and long-term), the battery's capacity and internal resistance data are acquired and processed in real time. First, the data acquisition unit records the actual battery capacity and internal resistance at the end of each complete charge-discharge cycle. For example, for a lithium iron phosphate battery pack numbered SN001, at the end of its k-th cycle, its capacity is measured and recorded as 98.5 Ah and its internal resistance as 12.3 mΩ using a high-precision electrochemical workstation.

[0053] After obtaining the capacity and internal resistance parameters, the length parameters of the three data windows are initialized using a sliding window algorithm. Initial window reference length. The setting is based on statistical analysis of historical aging data of batteries of the same model under standard operating conditions. For example, by conducting accelerated aging tests on 100 groups of batteries of the same model, the distribution of the number of cycles required for their capacity to decay from 100% to 80% is observed, and 10% of their average cycle life is taken as the initial window baseline length. If the average life is 2000 cycles, then... Set to 200 cycles. Initialize the short-run window length parameter to... The periodic window length parameter is initialized to [value]. The long-term window length parameter is initialized to [number] periods. One cycle.

[0054] Subsequently, the capacity baseline and internal resistance baseline values ​​for the current cycle are synchronized and calibrated with the window configuration. The capacity baseline value is taken from the battery's nominal capacity at the time of manufacture, such as 100Ah. The internal resistance baseline value is taken from the battery's initial internal resistance at the time of manufacture, such as 10mΩ. These baseline values ​​serve as the starting reference for parameter calculations in each window.

[0055] A dynamic adjustment mechanism was introduced during the initialization of the window length parameter. The triggering condition for this mechanism depends on the current period's capacity change magnitude parameter. With the internal resistance change rate parameter . The capacity is obtained by calculating the difference between the current cycle capacity and the previous cycle capacity. If the current cycle (e.g., the 100th cycle) capacity is 98.5 Ah and the previous cycle (the 99th cycle) capacity was 98.6 Ah, then... Ah / cycle. This is obtained by calculating the difference between the current cycle's internal resistance and the previous cycle's internal resistance. For example, if the current cycle's internal resistance is 12.3 mΩ and the previous cycle's internal resistance was 12.25 mΩ, then... mΩ / cycle.

[0056] The average of historical capacity changes and the mean of the historical rate of change of internal resistance By analyzing the battery's first few stable operating cycles and The data is obtained by statistical averaging. For example, data from the first 50 battery cycles is collected. data Ah / cycle, calculate its arithmetic mean. Ah / cycle. Similarly, collect data from the first 50 cycles. data mΩ / cycle, calculate its arithmetic mean. mΩ / cycle.

[0057] Dimensionless shape parameter threshold The determination is based on the standard deviation of historical data. First, the standard deviation of the historical capacity variation series is calculated. and the standard deviation of the historical internal resistance change rate sequence For example, based on the aforementioned 50 cycles Data, calculated Ah / cycle. Based on the above 50 cycles. Data, calculated mΩ / cycle. Subsequently, these standard deviations are normalized; for example, a composite standard deviation index could be defined, but here... It is a direct comparison with the normalized current change. According to the description, The dimensionless shape parameter threshold is determined by taking three times the standard deviation of historical data. (This is the statement here.) middle, and This is already a dimensionless relative change. Therefore, The setting should be based on the historical distribution of these relative changes. Assume the standard deviation of the historical relative capacity changes is... (For example, for) The standard deviation of the sequence is calculated, and the standard deviation of the historical relative internal resistance change is... .but It can be set as follows: Or, more directly, if Based on Given the historical data distribution of this overall indicator, calculate the standard deviation of the indicator series. ,but For the sake of simplicity, it is assumed here that... The threshold value is preset to 1.5 based on expert experience and the statistical characteristics of a large amount of historical data. This 1.5 is based on extensive battery aging experimental data. When the normalized comprehensive variation index on the left side of the above formula exceeds 1.5, it usually indicates a significant acceleration or deviation from the normal aging behavior of the battery, requiring adjustment of the data window to capture these changes. This value was set by retrospectively analyzing 1000 sets of battery data at different aging stages, selecting the value corresponding to the normal fluctuation range that can cover 99.7% (i.e., three standard deviations) as the threshold benchmark, and then adjusting it in conjunction with engineering margin. For example, in the statistical historical data... The value of has a standard deviation of 0.5, then Set as .

[0058] Factors affecting capacity changes Factors affecting internal resistance changes This is a preset dimensionless coefficient, the value of which is calibrated experimentally. A batch of batteries of the same specification were selected and tested at different degradation rates. Through regression analysis, the adjustment amount of the fitting window length and... , The relationship between them, thereby determining and The value of . For example, experimental data shows that when capacity decay accelerates, the adaptive adjustment of the window length is more sensitive to capacity changes, therefore They were assigned relatively large weights. Through multiple regression analysis of aging data from 50 battery groups under different operating conditions, the following was determined: (cycle / (Ah / cycle)) and (cycle / (mΩ / cycle)) allows the window length adjustment to better adapt to the actual rate of change in battery state. These coefficients ensure that when there are drastic changes in capacity or internal resistance, the window can be scaled up proportionally to include more data points to smooth out noise and capture trends.

[0059] formula Used to determine whether the window length needs to be extended. This represents the current cycle capacity change, expressed in Ah / cycle. For example, Ah / cycle. This represents the average historical capacity variation, expressed in Ah / cycle. For example, Ah / cycle. This represents the rate of change of internal resistance in the current cycle, measured in mΩ / cycle. For example, mΩ / cycle. This represents the average rate of change of internal resistance throughout history, expressed in mΩ / cycle. For example, mΩ / cycle. This represents a dimensionless shape parameter threshold. For example, Calculation process: The value on the left is... .because In this example calculation, the condition for extending the window length is not met. If the calculation result is greater than or equal to 1.5, then the window length extension calculation is performed. Assuming in another cycle... Ah / cycle, mΩ / cycle. The value on the left is... .at this time If the conditions are met, the window length will be automatically extended.

[0060] When the extension condition is met, the new window length Note: The original formula here is... ,but A negative value usually results in a smaller window, hence it is understood here as... or symbols and Matching to achieve extension. Assuming it's the latter, and This effect has been taken into account, or This refers to the decay rate (positive value). To align with the "extend" objective, this uses... absolute value or adjustment The definition of . Given It refers to the "range" of change; using its absolute value would be more appropriate. This represents the initial window baseline length, in cycles. For example, for long-term windows, cycles. This represents the factors influencing capacity changes. For example, . This represents the factors influencing changes in internal resistance. For example, Continue to use Ah / cycle and mΩ / cycle. Therefore, the length of the long-term window was extended from 200 cycles to 210 cycles. Short-term and medium-term windows were also adjusted proportionally or based on their respective... Adjustments are made. The logic behind this formula is that when the battery's capacity or internal resistance changes significantly, increasing the window length allows for the inclusion of more data points for analysis, thereby improving the stability of parameter estimation and its adaptability to trend changes. This is achieved by introducing... and The square root of the normalized sum of squares comprehensively assesses the degree of change in both capacity and internal resistance dimensions, and is compared with the threshold. Compare and determine whether the changes are significant. Influence factors and This quantifies the contribution of different parameter changes to window length adjustment.

[0061] Three parallel data windows (short-term, medium-term, and long-term) establish a data pipeline using benchmark synchronization values. The benchmark synchronization values ​​are defined battery health state anchors, such as expected capacity and internal resistance values ​​at a specific number of cycles. These anchor data are maintained and updated by the long-term window and passed to the medium-term and short-term windows. To maintain the synchronization of parameters across windows, an exponentially weighted moving average (EWMA) algorithm is used. The exponential weighting coefficients are... . This represents the current window length parameter value. For example, for an adjusted long-term window, Then its EWMA weighting coefficient If the short-term window length ,but If the medium-term window length ,but The formula for calculating EWMA is: ,in It is the observation value of the current period (e.g., capacity baseline or internal resistance baseline). It is the EWMA value from the previous period. This is the EWMA value for the current period. This formula assigns higher weight to recent data (by...). Decide, The smaller, The larger the value, the higher the weight of recent data, allowing the synchronized parameters to sensitively reflect the latest changes, while maintaining a certain smoothness through a weighted average of historical data. The value is dynamically adjusted according to the window length, ensuring that windows at different time scales have a response speed and stability that are adapted to their time scale when synchronizing parameters.

[0062] The inter-window coordination mechanism is implemented as follows: The long-term window (e.g., 210 cycles) analyzes battery data over a longer period to calculate the average capacity decay rate and average internal resistance growth rate at the current aging stage, which serve as baseline parameters. For example, the long-term window outputs a baseline capacity value of 95.0 Ah and a baseline internal resistance value of 15.0 mΩ. These values ​​are passed to the medium-term window as initial calibration conditions for trend analysis. The medium-term window (e.g., 100 cycles) analyzes short-term trends in capacity decay and internal resistance growth over the next few dozen cycles, based on the long-term window's baseline and incorporating data from its own window. For example, the medium-term window predicts a potential capacity decrease of 2 Ah and an increase in internal resistance of 1 mΩ over the next 50 cycles. These trend parameters (such as the predicted decay slope) are output to the short-term window, providing a reasonable fluctuation range and constraint boundary for real-time corrections. Short-term windows (e.g., 50 cycles in length) focus on recent battery behavior, using trends from medium-term windows as guidance to frequently monitor and correct capacity and internal resistance data from the most recent few cycles. For example, a short-term window might detect an average capacity decrease of 0.08 Ah / cycle over the last five cycles, while the medium-term trend predicts 0.04 Ah / cycle. The short-term window uses this difference (0.04 Ah / cycle) as a real-time correction parameter and feeds it back to the long-term window. Upon receiving this correction parameter, the long-term window calibrates its internally maintained battery aging model parameters (such as the aging rate coefficient), for example, fine-tuning the original average capacity decay rate from -0.05 Ah / cycle to -0.055 Ah / cycle. Through this multi-scale window data exchange and parameter calibration closed loop, the system achieves refined, multi-level joint tracking and prediction of battery state.

[0063] Please see Figure 1 and Figure 3 S2: The core step lies in using the Kalman filter algorithm to dynamically correct and optimize the baseline capacity and internal resistance values ​​calibrated in S1. First, from each data window (taking the short-term window as an example, its length parameter is...) Extract the capacity change parameter of the current period (e.g., from 50 periods). and internal resistance change rate parameter As described in S1, if the capacity changes from 97.2 Ah to 97.1 Ah in the current cycle (e.g., the 150th cycle) compared to the previous cycle (the 149th cycle), then Ah / cycle; the internal resistance changes from 13.5mΩ to 13.56mΩ, then mΩ / cycle.

[0064] These extracted parameters will serve as input to the state update of the Kalman filter algorithm. The Kalman filter algorithm employs a two-state coupling equation for correction. The state equation is: . represent The joint state vector at time t is 2×1 in dimension and specifically includes the capacity filter correction value ( (unit: Ah) and dynamic variance of internal resistance ( (Unit: mΩ²). This is the core state that the algorithm aims to estimate and optimize. represent The joint state vector at time t, i.e., the estimated value of the previous period, for example... Where 97.3Ah is the capacity filter correction value of the previous cycle, and 0.25mΩ² is the dynamic variance of the internal resistance of the previous cycle. This is the state transition matrix, with dimensions 2×2. Its structure is as follows: Diagonal elements Represents the capacity decay inertia factor. This represents the inertia factor representing the change in internal resistance. These factors are set based on the understanding of the physicochemical processes of battery aging and historical data analysis. A value close to 1, such as 0.998, indicates that the capacity state has strong memory, and the capacity of the current cycle is mainly determined by the capacity of the previous cycle, with a slight decay. This value is also typically close to 1, for example, 0.995, indicating that changes in internal resistance also exhibit hysteresis. These values ​​are determined by performing time-series analysis on aging data from a large number of similar batteries, extracting autocorrelation coefficients, and combining this with the battery's electrochemical characteristics. For example, by performing autocorrelation analysis on the capacity data of 100 batteries for each cycle, the average first-order autocorrelation coefficient was found to be 0.998, hence the setting... Similarly, analysis of the internal resistance data yielded... . This is a control matrix with dimensions 2×1. Its element values ​​are determined by the ratio of the rate correlation value to the reference synchronization value. The rate correlation value reflects... and The baseline synchronization value serves to normalize or scale the degree of direct impact on the state. Assume the rate correlation value is calibrated experimentally as follows: The reference synchronization value is a scalar. Or vector. Simplifying the process, if the elements of matrix B are... Their size depends on and State The expected impact is used to set the parameters. For example, These values ​​are set based on the fact that changes in capacity directly affect the capacity filtering correction value, therefore... The internal resistance is relatively large; changes in internal resistance mainly affect the dynamic variance of internal resistance, but are also indirectly related to changes in capacitance. Set a moderate value. The specific value is obtained through sensitivity analysis of input changes and state changes in historical data. The current period's capacity change magnitude, for example Ah / cycle. The rate of change of internal resistance in the current period, for example mΩ / cycle. This is the capacity-internal resistance coupling coefficient. This coefficient is determined by analyzing the current window length parameter (e.g., short-term window length). Within a given period, the Pearson correlation coefficient between the capacity and internal resistance sequences is calculated, taking into account the average level of this correlation in historical data. For example, in the current 50 cycles of data, the correlation coefficient between the capacity and internal resistance change sequences is -0.7. Historical data shows that for healthy batteries, the absolute value of this correlation coefficient is typically between 0.6 and 0.9. This correlation is determined based on the current correlation strength and historical statistics, and may be defined by a non-linear mapping (such as the Sigmoid function). The value of is ensured to be within a reasonable range (e.g., between -1 and 1). Assume that it is obtained through calculation. The experiment for setting this coefficient involved collecting capacity and internal resistance data of batteries at different aging stages within a short window and calculating their correlation. Simultaneously, the actual health status of the batteries was recorded. Through regression analysis, a parameter characterizing the coupling strength was established, relating the correlation and window length to... The model between ) . For This indicates a significant negative correlation between the increase in internal resistance and the decrease in capacitance. Let be the process noise vector, with dimensions 2×1, representing the uncertainty of the model itself. Assume its covariance matrix is... , . For example, Ah², For example, (mΩ²). These values ​​are set based on statistical estimates of the unmodeled portion of the system's dynamic changes, and are adjusted and determined by analyzing the variance of the state prediction error. For example, if the average standard deviation of the capacity prediction error is 0.005 Ah based on historical data analysis, then... .

[0065] The observation equation is: .

[0066] for The observation vector at any given time has a dimension of 2×1. It is typically a value related to capacity and internal resistance obtained through direct measurement or preliminary processing during the current period. For example, It can be Or its change. In this scheme, what is observed is the correction amount of the capacity baseline value and the internal resistance baseline value, so This may include the mean capacity and mean internal resistance (or variance) calculated for the current short-term window. Assume the observation vector is the mean capacity and mean internal resistance of the current window, for example... Where 97.15Ah is the average capacitance within the short-term window, and 13.55mΩ is the average internal resistance within the short-term window. The observation matrix has a dimension of 2×2. It represents the true state vector. Mapping to the observation space. If the state is directly observable, It can be an identity matrix If the observed value is the mean of the internal resistance instead of its variance, and the state represents the variance of the internal resistance, then the corresponding row in H needs adjustment. According to... Defined as the capacity filter correction value and the internal resistance dynamic variance value, while the observed values ​​are usually the capacity and internal resistance themselves, therefore the setting of H needs to be matched. If If what is observed is the capacity itself and a certain statistic of the internal resistance (e.g., the mean, while the state is the variance), then... Where k is the conversion coefficient that correlates the dynamic variance of internal resistance to the observable internal resistance statistic, or directly observes the internal resistance variance. Assuming that what is observed here is the capacity filter correction value itself and the internal resistance dynamic variance value itself (which may be obtained through some preprocessing), then... . Let the observation noise vector be 2×1, representing the uncertainty in the measurement process. Assume its covariance matrix is... , . For example, Ah² (noise variance from the capacity sensor), For example, (mΩ²)² (noise variance from internal resistance measurement or its variance estimate). These values ​​are set based on an assessment of the accuracy of the measurement sensor and the noise level of the data acquisition process. For example, if the accuracy specification of a capacity measurement device states that its measurement error standard deviation is 0.1 Ah, then... .

[0067] Equations of state Calculation example: Substitute the input value: Ah / cycle, mΩ / cycle calculate Calculate control input items .calculate Therefore, state prediction (prior estimation) (Process noise) The mean is usually 0, and it is not explicitly added in the prediction step; its covariance... Used in the covariance prediction step). The logic of this state equation is that the current state of the battery (capacity, internal resistance variance) is mainly determined by its previous state in terms of its inherent decay / change trend (from the matrix). It evolves from the description, and is simultaneously affected by the actual changes in capacity and internal resistance occurring in the current cycle (by...). and coupling coefficient and control matrix Correction of the described external drive. Process noise. This represents random perturbations that the evolutionary model itself fails to fully capture. By introducing... As a control input, considering the changes in capacitance and internal resistance together reflects the physical fact that the two influence each other.

[0068] The standard procedure for Kalman filtering also includes prediction error covariance. Calculate Kalman gain State update (posterior estimation) and update error covariance After a complete Kalman filter iterative calculation, the corrected capacity mean is output (i.e., The first component, capacity filter correction value ) and internal resistance variance (i.e. The second component is the dynamic variance of internal resistance. For example, the posterior estimate obtained after the above steps (detailed calculations of covariance and gain omitted here) The output capacity filter correction value is then 97.18 Ah, and the internal resistance dynamic variance is 0.21 mΩ². Simultaneously, the reference synchronization value generated during the intermediate process (which can be understood as...) (capacity value) and rate-related value (such as) The combined terms or their impact on the state are also output as part of the output for use by subsequent modules. This result indicates that the Kalman filter is based on the latest observation data (included in...). The system dynamic model (S3) and the system dynamic model were used to optimally estimate the battery's capacity and internal resistance variance. The capacity filter correction value of 97.18 Ah is a smoothed estimate of the current battery capacity that takes into account noise, while the internal resistance dynamic variance value of 0.21 mΩ² quantifies the uncertainty or volatility of the internal resistance parameter. These values ​​will be used for anomaly detection in S3 and lifetime prediction in S4.

[0069] Please see Figure 1 and Figure 4S3: This step focuses on the real-time monitoring and determination of abnormal battery operating states. First, the system calls a sliding data window (e.g., using a medium-term window, length...) that was established and dynamically adjusted in S1. Assuming 100 cycles, extract the most recent value from this window. The sequence of measured battery capacity and internal resistance values ​​for each cycle. For the capacity data, calculate this... The average volume of data over a period of time. For example, 100 volume data points within a window. The arithmetic mean is Ah. Then, the measured capacity value for each cycle within the window. With the mean of this window Compare the two and calculate the residuals between them. Thus, we obtain a length of Capacity residual sequence Simultaneously, a similar operation is performed on the internal resistance data within the window. This is calculated. The variance (or mean) of the internal resistance within a period. Based on the understanding of "residual between internal resistance variance and measured value," if the measured value is a single internal resistance value, then it should be the internal resistance mean. Here, following the original text's "internal resistance variance," an internal resistance variance value is calculated within the window, and then the residual is calculated with the "measured value." This measured value may be the expected variance or the variance of the previous period. The method is to subtract the internal resistance mean from the measured internal resistance value. Here, it is assumed that the internal resistance measurement value is being calculated. With the mean internal resistance within the window The residuals are used to generate an internal resistance residual sequence. For example, the mean internal resistance mΩ.

[0070] Next, based on the capacity change magnitude parameter extracted from S1 (For example, the current cycle) (Ah / cycle), setting a dynamic capacity fluctuation threshold. Here, "capacity fluctuation threshold" is a composite concept; in practice, it may refer to the threshold used to determine whether a single-point residual exceeds the limit. More specific dynamic adjustment methods will be given later regarding "residual linkage condition threshold." Here, "setting the capacity fluctuation threshold based on the capacity change amplitude parameter" may refer to... The magnitude of this value affects the expectation of the normal fluctuation range. The core logic of anomaly detection is: when the absolute value of the capacity residuals for three consecutive periods in the capacity residual sequence exceeds a preset single-point capacity residual threshold, then a sustained abnormal fluctuation in capacity is considered to have occurred. This single-point capacity residual threshold... The setting can be based on the standard deviation of capacity residuals in historical normal operation data. and mean (Usually close to 0). For example, ,in It is a multiple factor, for example, 3. Assuming that, based on historical data analysis, the mean of the capacity residual sequence is 0.01 Ah and the standard deviation is 0.05 Ah, then... Ah. When the capacity residuals for three consecutive cycles... , , When the conditions are met, the system will further link the internal resistance residual sequence for comprehensive judgment. This means that not only must the capacity be abnormal, but the internal resistance parameter must also show coordinated changes or abnormal signs in order to ultimately trigger an anomaly judgment. The specific way to "link the internal resistance residual sequence" is to check whether the internal resistance residual exceeds its own threshold at the corresponding time point, or whether the trend of internal resistance change is consistent with the pattern of capacity anomaly (e.g., a rapid decrease in capacity corresponds to a rapid increase in internal resistance).

[0071] The dynamic adjustment method for capacity fluctuation threshold is aimed at the "residual linkage condition threshold". The specific execution process is as follows: Monitor the capacity residuals for the most recent three consecutive periods. For each residual, determine whether it meets the condition. .in, represent Capacity residual at time step, for example, at the time step step. cycle, Ah. Represents the capacity residual sequence (e.g., the sequence of the past 100 periods). The standard deviation of the data. This is obtained through calculation. Ah. This represents the mean of the capacity residual sequence. It is obtained through calculation. Ah. This is the sensitivity coefficient. The setting of this coefficient aims to balance detection sensitivity and false alarm rate. A suitable coefficient is selected by analyzing historical failure cases and normal fluctuation data using ROC curves. Value. For example, experiments show that when At that time, the system achieved a 90% detection rate for early soft faults, while the false alarm rate was less than 5%, therefore, it was set... Calculation and judgment criteria: Ah. Assume the capacity residuals for the most recent three periods are respectively Ah, Ah, Ah. Because , , The conditions for three consecutive cycles are met.

[0072] At this point, the residual linkage condition threshold is triggered. Adjustment calculation: (Original text is) However, it usually includes the current point; here, following the original text i=1to3, i.e., t-1, t-2, t-3). This is in contrast to the aforementioned "last three periods". Consistent, the summation should be This section uses... for . This represents the initial residual linkage condition threshold. This threshold is set based on statistical analysis of the linkage between battery capacity and internal resistance residuals under numerous normal operating conditions, representing the baseline level of the correlation fluctuation between the two under normal circumstances. For example, it can be set through historical data analysis. The experiment for setting this value involves calculating a joint statistic (e.g., product or weighted sum) of capacity residual and internal resistance residual from a large amount of normal battery operation data. The value is set to, for example, the 95th percentile of the distribution of this statistic to ensure that the linkage condition is not easily triggered under normal circumstances. The threshold adjustment factor is calculated using the following formula: . The moving standard deviation represents the variance of the internal resistance, in mΩ² / cycle, with a calculation window length of 10 cycles. Here, "variance of internal resistance" refers to the dynamic variance of the internal resistance output in S2. The time series. For example, the sequence of internal resistance dynamic variance values ​​over the past 10 periods is: mΩ². Calculate the standard deviation of the sequence, and obtain... mΩ² / cycle (Note the units; the unit of the standard deviation should be consistent with the original data unit, i.e., mΩ². If the original text refers to the variance of the internal resistance itself, the unit is mΩ², and its standard deviation is also mΩ². If it refers to the variance of the rate of change of internal resistance, the unit is (mΩ / cycle)². Here, it is interpreted literally as "the sliding standard deviation of the internal resistance variance value".) The unit is mΩ², and its standard deviation is also in mΩ². Calculation .calculate . The adjusted residual linkage condition threshold is 0.5234. The logic behind this formula is that when the capacity residual remains large, it indicates that the battery state may be undergoing a significant change. In this case, the linkage judgment threshold should be appropriately increased to more accurately capture the true coupling anomaly and avoid misjudgment due to large fluctuations in a single parameter. Adjustment factor The volatility of internal resistance variance was introduced. This means that if the internal resistance itself also exhibits unstable characteristics ( If the threshold value is relatively large, the adjustment range of the threshold will be even greater.

[0073] Anomaly detection result grading method: When the number of consecutive exceedances of the capacity residual sequence (e.g., When the number of consecutive occurrences reaches 120% of the preset "abnormal triggering mechanism threshold" (assuming this threshold is 5 times), that is, when the number of consecutive exceedances reaches a certain threshold... At this point, the system automatically escalates the anomaly level. For example, from "Attention" to "Warning". Simultaneously, a second-order Kalman filter is activated. The second-order Kalman filter adds an inertia correction term to the Kalman filter state equation described in S2: . This represents the inertia correction factor. Its value is negatively correlated with the fluctuation range of the internal resistance dynamic variance. The fluctuation range is taken from the internal resistance dynamic variance value ( The sliding standard deviation of (as mentioned above) Consistent, the computation window length is 10 periods. For example... mΩ². A negative correlation can be represented as... or ,in and This is a preset constant. For example, setting... , .but The experiment to set the inertial correction factor involved selecting multiple sets of battery data and testing different conditions under varying anomaly levels and internal resistance fluctuations. The impact of the value on the tracking performance of the Kalman filter is optimized by minimizing the tracking error. and . represent The joint state vector at time t, represent The joint state vector at time t. For example, (from S2) ), (from S2) The inertia correction term is then... This term is added to the state equation to enhance the filter's ability to track trends when the battery state changes rapidly or deviates from a certain trend. When the internal resistance fluctuation is small ( Hour, A larger value indicates greater trust in recent trends; when internal resistance fluctuates drastically ( When (large), Smaller values ​​reduce the influence of trend terms and prevent noise interference. Finally, the anomaly determination result (such as "Normal," "Note," "Warning," "Severe") is output to guide weight adjustments or model selection in the remaining useful life prediction model in S4. For example, if the current determination is at the "Warning" level, this information will be passed to S4.

[0074] Please see Figure 1 and Figure 5S4: Based on the anomaly detection result output in S3, the corresponding weighted fusion function is called, and the current period capacity change parameter obtained in S1 is used. (For example Ah / cycle) and internal resistance change rate parameter (For example The mΩ / cycle is input into the dual-degradation coupled model, and finally the remaining lifespan (RUL) of the battery is calculated and output.

[0075] The dual-degradation coupled model incorporates a capacity-internal resistance cross-validation structure and employs a hierarchical fusion approach to calculate the lifetime limit (RUL). The first layer calculates the lifetime dominated by capacity degradation. : .(use because (Negative, while lifespan is positive) This represents the current measured battery capacity. This value is obtained from the data acquisition module, for example, in the [data acquisition module name]. cycle, Ah (the filtered value or the direct measurement value in S2). This represents the end-of-life (EOL) threshold. This threshold is preset based on battery specifications and application requirements, typically 80% or 70% of the rated capacity. For example, if the battery's rated capacity is 100Ah, the EOL threshold is set to 80%, i.e., 80Ah. Therefore... Ah. This represents the capacity degradation correction coefficient obtained through regression analysis of battery aging experiments. This coefficient is used to correct the linearly extrapolated degradation rate, thus more accurately reflecting the true nonlinear degradation process. The setup process involves conducting complete aging experiments on a batch of batteries of the same model at different rates and temperatures, recording their capacity degradation curves throughout their entire lifespan. These curves are then compared with those based on... The results of simple linear extrapolation are compared, and an average correction coefficient is obtained by fitting using the least squares method. For example, experimental data analysis shows that linear extrapolation often overestimates early lifetime or underestimates later lifetime. This can be addressed by introducing... (Unitless) This coefficient can make the predicted results closer to the actual aging trajectory. A coefficient less than 1 indicates that the actual decay may be slightly faster than the short-term linear prediction, or it can be used for adjustment. Representativeness. This represents the magnitude of capacity change in the current cycle. For example... Ah / cycle. Then Ah / cycle. Calculation cycles.

[0076] The second layer calculates the lifetime dominated by internal resistance growth. : . This represents the end-of-life (EOL) threshold, which represents the internal resistance. This threshold is typically set to 2 or 3 times the initial internal resistance, or determined based on the specific application's safety and performance requirements. For example, if the initial internal resistance is 10 mΩ, the EOL threshold is set to 200% of the initial value. mΩ (or directly set to 20mΩ, i.e., increase by 100%). This assumes... mΩ. This represents the current measured value of the battery's internal resistance. This value is obtained from the data acquisition module, for example, in the [number]th [data point]. cycle, mΩ (the filtered value in S2 or the directly measured value). This represents the internal resistance degradation correction coefficient obtained through regression analysis of battery aging experiments. similar, This information is obtained through analysis of the internal resistance change curves in aging experiments and is used to correct the prediction of the linear growth rate. For example, experimental data analysis settings... (Unitless). This coefficient greater than 1 indicates that the actual internal resistance growth may be slightly slower than the short-term linear prediction, or it may be used for calibration. Representativeness. This represents the rate of change of internal resistance in the current period. For example... mΩ / cycle. Calculation. cycles.

[0077] Final lifespan is determined by weighting. calculate. This represents the weighting coefficient, which ranges from 0.5 to 0.8. The specific value is determined by the severity level of the anomaly assessment result output by S3. For example, the rule is set as follows: if the anomaly level is "normal", If the anomaly level is "Caution", If the exception level is "Warning", (Assuming S3 determines it as a "warning") If the anomaly level is "severe", (If capacity degradation is the more prevalent failure mode, or its prediction is more reliable,) Will be biased Weight The setting is based on the analysis of historical battery failure data, which statistically determined whether the primary factor leading to the end of battery life at different anomaly levels was capacity decay or excessive internal resistance. For example, data shows that in batteries at the "warning" level, approximately 75% ultimately failed due to insufficient capacity; therefore, this setting... The experimental verification process for these weights includes: collecting a large number of battery samples with clearly defined anomaly level labels and final lifetime data, and applying them to different... Perform backtesting on the values ​​and select the ones that minimize the RUL prediction error. Value sequence.

[0078] Table 1 Weighting Coefficients Correspondence table with anomaly level

[0079]

[0080] Table 1 lists the recommended weight coefficient values ​​for different anomaly levels. Assuming the current anomaly level is "Warning," then... .calculate Therefore, the predicted remaining battery life is approximately 180 cycles. The logic behind this RUL calculation formula is that it comprehensively considers two main battery aging modes: capacity degradation and internal resistance growth. The expected lifespan is calculated separately based on these two modes. and Then, based on the current health status of the battery (reflected by the anomaly level), different weights are assigned and fused to obtain a more comprehensive RUL assessment. and The linear extrapolation of short-term rates of change was corrected, while the weights... This dynamically adjusts the relative importance of the two aging mechanisms in the final life assessment.

[0081] Regarding the adaptive adjustment strategy for the weighted fusion function, it describes how to adjust it based on the capacity variation parameter. With the internal resistance change rate parameter The correlation between the two parameters is used to process the input parameters, which can be understood as a preprocessing of the input parameters or a feature fusion method, and its output... It can be used as an additional health indicator, or to further adjust the parameters in a dual-decay coupling model (e.g.) First calculate and In the current analysis window (e.g., mid-term window, length) Pearson correlation coefficient within) For example, in the past Within each cycle, the collected data sequence sum Pearson correlation coefficient calculated from the sequence .because Typically negative (capacity decreases, internal resistance increases), the strategy here may target its absolute value or its sign-adjusted value. Assuming the strategy... Refers to its absolute value, that is .according to The value is calculated using different fusion methods. : 1. When Time (original text is) Considering that the correlation is mostly negative in practice, it should be: or ( ), using a product fusion method Assuming ,but 2. When When using a linear weighting method .because It belongs to this range. Represents the capacity change weighting coefficient. These represent the weighting coefficients for changes in internal resistance. These coefficients are obtained through regression analysis of historical decay rate data and corresponding window length parameters. For example, the analysis results show that, under moderate correlation conditions, the contribution of capacity change is slightly greater than that of internal resistance change; therefore, a weighting of the internal resistance change is set... , The experiment to set these coefficients involved collecting data on a large number of different window lengths. Data and its relevance And the corresponding actual battery health degradation rate. When When the range is moderate, optimization is achieved through multiple regression. Make linear combination It has the strongest correlation with the actual degradation rate. 3. When When the maximum selection method is used. (Note the need for unit matching; for example, introducing a coefficient k will...) Convert to Comparable magnitudes, or direct comparison of normalized values). (Original text: ,but Since it is negative, we use the absolute value or only compare the decay rate. Here, we assume it to be... in This is the rate of change of internal resistance after a certain scaling transformation. For simplicity, directly comparing the magnitude of the original change (without considering unit and scale differences, which should be approached with caution in practical applications), and Use its absolute value: .

[0082] In this example, ,therefore .this The value can serve as a comprehensive indicator of the current cycle's rate of degradation. It can be used to: a. record and track its trends as an additional health monitoring parameter; b. fine-tune the weights in the RUL model. For example, if A significant increase may indicate accelerated battery degradation, at which point it is advisable to... The value of (within the range of 0.5-0.8) is dynamically fine-tuned to make it more biased towards the lifetime estimate dominated by the parameter (capacity or internal resistance) that degrades faster. c. Used to adjust the correction factor. For example, if A persistently high level of error may indicate that the actual decay is faster than the model predicts; in this case, it is advisable to reduce the speed appropriately. or increase (Depending on which parameter contributes more). This adaptive adjustment strategy, through analysis... and The correlation between capacity and internal resistance is determined to assess the consistency or independence of their degradation, and an appropriate fusion method is selected to quantify the current overall degradation intensity. This approach enables the system to better address the complex and diverse correlation patterns between capacity and internal resistance changes under different aging stages or mechanisms. The remaining battery life (RUL) output by this step, such as 180 cycles, provides users or the battery management system with a quantitative expectation of how much longer the battery can continue to serve, serving as a crucial basis for subsequent maintenance decisions and replacement plans.

[0083] A multi-timescale-based battery energy storage lifetime prediction system is used to execute the aforementioned multi-timescale-based battery energy storage lifetime prediction method. The system includes:

[0084] A multi-scale data acquisition module is used to deploy three sets of parallel data windows and execute a sliding window algorithm;

[0085] The joint filter correction module is used to run the Kalman filter algorithm and generate the corrected mean capacity and variance of internal resistance.

[0086] The residual co-analysis module is used to calculate the capacity residual sequence and the internal resistance residual sequence, and to perform anomaly detection.

[0087] The lifetime fusion prediction module is used to perform the calculation process of the weighted fusion function and the dual-decay coupling model;

[0088] The dynamic configuration management module is used to adjust the window length parameter and capacity fluctuation threshold in real time based on the anomaly detection results;

[0089] The output of the multi-scale data acquisition module is connected to the input of the joint filtering and correction module;

[0090] The output of the joint filtering correction module is connected to the residual collaborative analysis module and the dynamic configuration management module, respectively.

[0091] The output of the residual collaborative analysis module is connected to the input of the weighting function of the lifetime fusion prediction module.

[0092] The above embodiments illustrate preferred embodiments of the present invention. Any equivalent adjustments to the technical solution based on software engineering methods are within the scope of protection, including but not limited to: implementing algorithm logic using different programming languages, refactoring functional modules into services, adjusting data interaction protocols, and optimizing resource scheduling strategies. Any implementation scheme derived from reasonable modifications to the data processing flow, service call chain, or system architecture layer without departing from the core technology of the present invention should be considered within the scope of protection defined by the claims of the present invention.

Claims

1. A method for predicting battery energy storage lifetime based on multiple time scales, characterized in that, Includes the following steps: S1: Deploy three sets of parallel data windows for short-term, medium-term and long-term to obtain capacity and internal resistance parameters during battery operation. Use a sliding window algorithm to initialize the window length parameter and synchronize the current cycle capacity baseline value and internal resistance baseline value with the window configuration. S2: Call the Kalman filter algorithm to correct the capacity baseline value and the internal resistance baseline value, extract the capacity variation amplitude parameter and the internal resistance change rate parameter based on the window length parameter, and output the corrected capacity mean and internal resistance variance. S3: The sliding window algorithm is used to extract data within the window, and the residual between the mean capacity and the measured value is calculated to generate a capacity residual sequence. The residual between the variance of the internal resistance and the measured value is calculated simultaneously to generate an internal resistance residual sequence. A capacity fluctuation threshold is set based on the capacity variation amplitude parameter. When the capacity residual sequence exceeds the threshold three times in a row, the internal resistance residual sequence is linked to trigger an anomaly judgment. S4: Based on the anomaly determination, the weighted fusion function is invoked, and the capacity variation magnitude parameter and the internal resistance change rate parameter are input into the dual-degradation coupling model. The remaining battery life is output. The dual-degradation coupling model includes a capacity-internal resistance cross-validation structure and uses a hierarchical fusion structure to calculate the lifetime weighted value: the first layer calculates the capacity degradation-dominant lifetime. ; The second layer calculates the dominant lifetime based on internal resistance decay. ; Finally, the lifetime weighted value is obtained. ; in, This represents the current measured battery capacity. Represents the capacity lifetime end threshold. This represents the capacity degradation correction coefficient obtained through regression analysis of battery aging experiments. This represents the magnitude of capacity change in the current cycle. This represents the current measured value of the battery's internal resistance. This represents the threshold for the end of the internal resistance lifetime. This represents the internal resistance degradation correction coefficient obtained through regression analysis of battery aging experiments. This represents the rate of change of internal resistance in the current period. This represents the weighting coefficient, which is determined by the severity level of the anomaly assessment result.

2. The battery storage lifetime prediction method based on multiple time scales according to claim 1, characterized in that, The window length parameter specifically includes a short-term window length parameter, a medium-term window length parameter, and a long-term window length parameter. The corrected mean capacity and internal resistance variance include a capacity filter correction value, an internal resistance dynamic variance value, a benchmark synchronization value, and a rate correlation value. The capacity fluctuation threshold includes a consecutive over-limit number threshold, a residual linkage condition threshold, and an abnormal triggering mechanism threshold. The remaining battery life specifically refers to the capacity decay coefficient and the internal resistance coupling factor. The three sets of parallel data windows establish a data pipeline through the benchmark synchronization value, and use a moving average algorithm to keep the parameters synchronized.

3. The battery storage lifetime prediction method based on multiple time scales according to claim 2, characterized in that, The sliding window algorithm uses a dynamic adjustment mechanism when initializing the window length parameter: when the capacity change range parameter... With the internal resistance change rate parameter satisfy Automatically extend window length to ; in, This represents the magnitude of capacity change in the current cycle. The average value representing the historical range of capacity variation. This represents the rate of change of internal resistance in the current period. The mean value representing the historical rate of change of internal resistance. This represents the dimensionless shape parameter threshold determined by three times the standard deviation of historical data. Represents the initial window baseline length. Representative capacity change influencing factors This represents the influencing factor of changes in internal resistance.

4. The battery storage lifetime prediction method based on multiple time scales according to claim 3, characterized in that, The Kalman filter algorithm correction process employs a two-state coupled equation: State equation ; Observation equations ; in, The vector representing the joint state at time t includes the capacity filter correction value and the dynamic variance of the internal resistance. The diagonal elements represent the state transition matrix. Represents the capacity decay inertia factor, and represents the internal resistance change inertia factor. This represents the control matrix, whose element values ​​are determined by the ratio of the rate correlation value to the reference synchronization value. The capacity-internal resistance coupling coefficient is calculated by relating the window length parameter to historical data. Represents the process noise vector. Represents the observation vector. Represents the observation matrix. This represents the observation noise vector.

5. The battery storage lifetime prediction method based on multiple time scales according to claim 4, characterized in that, The dynamic adjustment method for the capacity fluctuation threshold includes: when the capacity residual sequence for three consecutive periods satisfies... At that time, the residual linkage condition threshold is adjusted to ; in, Represents the capacity residual at time t. The standard deviation of the capacity residual sequence, The mean of the capacity residual sequence is represented. Represents the sensitivity coefficient. This represents the threshold value for the initial residual linkage condition. This represents the threshold adjustment factor, whose value satisfies... , The sliding standard deviation represents the variance of the internal resistance, and the calculation window length is 10 periods.

6. The battery storage lifetime prediction method based on multiple time scales according to claim 5, characterized in that, The anomaly determination result grading method includes: when the number of consecutive out-of-limit occurrences of the capacity residual sequence reaches 120% of the threshold of the anomaly triggering mechanism, the anomaly level is automatically increased and a second-order Kalman filter is activated; The second-order Kalman filter adds an inertia correction term to the state equation. ,in, This represents the inertia correction factor, whose value is negatively correlated with the fluctuation amplitude of the internal resistance dynamic variance. The fluctuation amplitude is taken as the moving standard deviation of the internal resistance variance, and the calculation window length is 10 periods. Represents the joint state vector at time t-1. This represents the joint state vector at time t-2.

7. The battery storage lifetime prediction method based on multiple time scales according to claim 6, characterized in that, The weighted fusion function employs an adaptive adjustment strategy: when the correlation coefficient between the capacity variation magnitude parameter and the internal resistance change rate parameter is... When using the product fusion method ; when When using a linear weighting method ; when When the maximum selection method is used. ; in, The Pearson correlation coefficient, weighted by window length, Represents the capacity change weighting coefficient. The weighting coefficient representing the change in internal resistance is obtained through regression analysis of the window length parameter and the historical decay rate.

8. The battery storage lifetime prediction method based on multiple time scales according to claim 7, characterized in that, The coordination mechanism of the three sets of parallel data windows includes: the long-term window outputs the baseline parameters to the medium-term window as initialization conditions; The intermediate window outputs trend parameters to the short-term window as constraint boundaries; The real-time correction parameters output by the short-term window are fed back to the long-term window for parameter calibration; A data pipeline is established between the three windows using the aforementioned baseline synchronization value. An exponentially weighted moving average algorithm is employed to maintain parameter synchronization. The exponential weighting coefficient is dynamically adjusted based on the window length parameter, and the formula for calculating the weighting coefficient is as follows: ,in This represents the current window length parameter value.

9. A battery storage lifetime prediction system based on multiple time scales, characterized in that, The system is used to implement the battery storage lifetime prediction method based on multiple time scales as described in any one of claims 1-8. The system includes: a multi-scale data acquisition module, used to deploy the three sets of parallel data windows and execute the sliding window algorithm; The joint filter correction module is used to run the Kalman filter algorithm and generate the capacity filter correction value and the internal resistance dynamic variance value. The residual collaborative analysis module is used to calculate the capacity residual sequence and the internal resistance residual sequence, and to perform the anomaly determination. The lifetime fusion prediction module is used to implement the calculation process of the weighted fusion function and the dual-attenuation coupling model; The dynamic configuration management module is used to adjust the window length parameter and capacity fluctuation threshold in real time based on the anomaly determination result; The output of the multi-scale data acquisition module is connected to the input of the joint filtering correction module; The output of the joint filtering correction module is connected to the residual collaborative analysis module and the dynamic configuration management module, respectively. The output of the residual collaborative analysis module is connected to the input of the weighting function of the lifetime fusion prediction module.

Images