Battery energy storage life prediction method and system based on multiple time scales

Abstract

The invention relates to the technical field of prediction models, in particular to a battery energy storage life prediction method and system based on multiple time scales, and the method comprises the following steps: deploying three groups of short-term, medium-term and long-term parallel data windows, obtaining battery capacity and internal resistance parameters, sliding window initialization parameters, synchronous baseline calibration, and Kalman filtering correction parameters, and extracting a variation amplitude and rate, an output capacity mean value and an internal resistance variance, calculating a residual sequence, performing three-time linkage on the capacity residual exceeding a threshold to trigger an abnormity, performing weighted fusion after the abnormity, and outputting the residual life. According to the invention, through multi-time scale parallel data window dynamic monitoring, sliding window synchronous calibration base line, Kalman filtering dynamic correction parameters, capacity and internal resistance residual error linkage determination mechanism establishment, dual-attenuation coupling model fusion multi-dimensional attenuation characteristics and multi-scale collaborative analysis of battery aging, the limitation of a linear model is broken through; the anomaly detection sensitivity and robustness are enhanced, the life prediction accuracy and timeliness are improved, and the misjudgment risk is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of prediction models, in particular to a battery energy storage life prediction method and system based on multiple time scales. BACKGROUND

[0002] The technical field of prediction models involves predicting trends and results of system behavior through mathematical modeling and data analysis means. The core lies in constructing a quantitative relationship network between variables and establishing a dynamic evolution mechanism. This field includes four technical levels: data feature extraction, model architecture design, parameter optimization strategy, and verification evaluation system. By training historical data to establish input-output mapping relationships, future state extrapolation is achieved. Among them, the traditional battery energy storage life prediction refers to a technical solution that records the number of charge and discharge cycles and analyzes voltage and current parameters, uses capacity decay curve fitting and empirical formula derivation methods, monitors the change rate of battery internal resistance and the downward trend of energy density, establishes a linear regression model combined with temperature stress accelerated aging test data, and uses cycle period counting method and capacity retention rate threshold determination to achieve remaining useful life estimation.

[0003] The traditional technology relies on single time window capacity decay curve fitting and fails to distinguish the decay characteristics of different time scales in the battery aging process. The linear regression model is difficult to capture dynamic nonlinear decay rules, and the empirical formula derivation is limited by fixed parameter assumptions, resulting in insufficient correlation analysis of early capacity fluctuations and mid-term accelerated aging stages. The static capacity retention rate threshold determination lacks a dynamic response mechanism for multi-parameter coupling effects. The collaborative evolution relationship between internal resistance change and capacity decay is not effectively quantified. Abnormal state detection is easily disturbed by short-term noise. The time-varying characteristics matching degree of accelerated aging test data and real working conditions is insufficient, causing the remaining life prediction to lag or deviate in complex charge and discharge scenarios. SUMMARY

[0004] The purpose of the present application is to solve the problems existing in the prior art and to provide a battery energy storage life prediction method and system based on multiple time scales.

[0005] In order to achieve the above purpose, the present application adopts the following technical scheme: a battery energy storage life prediction method based on multiple time scales, comprising the following steps: S1: deploying three groups of parallel data windows of short-term, medium-term and long-term, obtaining capacity parameters and internal resistance parameters during battery operation, initializing window length parameters using a sliding window algorithm, synchronously calibrating current cycle capacity baseline value and internal resistance reference value with window configuration; S2: calling a Kalman filter algorithm to correct the capacity baseline value and the internal resistance reference value, extracting capacity variation amplitude parameters and internal resistance change rate parameters based on the window length parameters, and outputting corrected capacity mean and internal resistance variance; S3: intercepting data in the window by using a sliding window algorithm, calculating the residual of the capacity mean value and the measured value to generate a capacity residual sequence, synchronously calculating the residual of the internal resistance variance and the measured value to generate an internal resistance residual sequence, setting a capacity fluctuation threshold based on the capacity variation amplitude parameter, when the capacity residual sequence exceeds the threshold for three times in succession, linking the internal resistance residual sequence to trigger an abnormality determination; S4: according to the abnormality determination, calling a weighted fusion function, inputting the capacity variation amplitude parameter and the internal resistance change rate parameter into a double attenuation coupling model, and outputting a battery remaining service life.

[0006] As a further scheme of the present application, the window length parameter is specifically a short-term window length parameter, a medium-term window length parameter, and a long-term window length parameter, the corrected capacity mean value and internal resistance variance include a capacity filtering correction value, an internal resistance dynamic variance value, a reference synchronization value, and a rate correlation value, the capacity fluctuation threshold includes a continuous over-limit number threshold, a residual linkage condition threshold, and an abnormality triggering mechanism threshold, and the battery remaining service life specifically refers to a capacity attenuation coefficient and an internal resistance coupling factor; The three sets of parallel data windows establish a data pipeline through the reference synchronization value, and a sliding average algorithm is used to keep the parameters synchronized; The double attenuation coupling model includes a capacity-internal resistance cross-validation structure.

[0007] As a further scheme of the present application, when initializing the window length parameter by using the sliding window algorithm, a dynamic adjustment mechanism is adopted: when the capacity variation amplitude parameter and the internal resistance change rate parameter satisfy , the window length is automatically extended to . Wherein, represents the current cycle capacity variation amplitude (unit: Ah / cycle), represents the mean value of the historical capacity variation amplitude (unit: Ah / cycle), represents the current cycle internal resistance change rate (unit: mΩ / cycle), represents the mean value of the historical internal resistance change rate (unit: mΩ / cycle), represents a dimensionless shape parameter threshold determined by three times of the standard deviation of historical data, represents an initial window reference length (unit: cycle), represents a capacity variation influence factor, represents an internal resistance change influence factor.

[0008] As a further scheme of the present application, the Kalman filtering algorithm correction process adopts a double-state coupling equation: a state equation ; Observation equation ; where, represents the joint state vector at time t (dimension: 2x1), containing the capacity filtering correction value (unit: Ah) and the internal resistance dynamic variance value (unit: mΩ²), represents the state transition matrix (dimension: 2x2), whose diagonal elements represents the capacity attenuation inertia factor (characterizing the memory effect characteristic of capacity change), represents the internal resistance change inertia factor (characterizing the hysteresis effect characteristic of internal resistance change), represents the control matrix (dimension: 2x1), whose element value is determined by the ratio of the rate correlation value and the reference synchronization value, represents the capacity-internal resistance coupling coefficient, which is calculated by the window length parameter and the historical data correlation degree, represents the process noise vector (unit: Ah², mΩ 4 ), represents the observation vector (dimension: 2x1), represents the observation matrix (dimension: 2x2), represents the observation noise vector (unit: Ah², mΩ 4 ).

[0009] As a further scheme of the present application, the dynamic adjustment method of the capacity fluctuation threshold value comprises: when the capacity residual sequence of the continuous three periods satisfies , the residual linkage condition threshold value is adjusted to ; where, represents the capacity residual at time t (unit: Ah), represents the standard deviation of the capacity residual sequence (unit: Ah), represents the mean of the capacity residual sequence (unit: Ah), represents the sensitivity coefficient, represents the initial residual linkage condition threshold value, represents the threshold adjustment factor, whose value satisfies , represents the sliding standard deviation of the internal resistance variance value (unit: mΩ² / cycle), and the calculation window length is 10 periods.

[0010] As a further scheme of the present application, the double attenuation coupling model adopts a hierarchical fusion structure: the first layer calculates the capacity attenuation dominant life ; The second layer calculates the internal resistance attenuation dominant life ; Finally, the life weighting value ; wherein, represents the current battery capacity measured value (unit: Ah), represents the capacity life termination threshold (unit: Ah), represents the capacity attenuation correction coefficient obtained by regression analysis of battery aging experiment, represents the current cycle capacity variation range (unit: Ah / cycle), represents the current battery internal resistance measured value (unit: mΩ), represents the internal resistance life termination threshold (unit: mΩ), represents the internal resistance attenuation correction coefficient obtained by regression analysis of battery aging experiment, represents the current cycle internal resistance change rate (unit: mΩ / cycle), represents the weight coefficient (value range: 0.5-0.8), which is determined by the severity level of the abnormality determination result.

[0011] As a further scheme of the present application, the abnormality determination result grading method comprises: when the number of consecutive over-limit times of the capacity residual sequence reaches 120% of the abnormality triggering mechanism threshold, automatically upgrading the abnormality level and activating the second-order Kalman filter; The second-order Kalman filter adds an inertial correction term in the state equation wherein, represents the inertial correction factor, whose value is negatively related to the fluctuation range of the internal resistance dynamic variance value, and the fluctuation range is the sliding standard deviation (unit: mΩ² / cycle) of the internal resistance variance value, and the calculation window length is 10 cycles, represents the joint state vector at t-1 time, represents the joint state vector at t-2 time.

[0012] As a further scheme of the present application, the weighted fusion function adopts an adaptive adjustment strategy: when the Pearson correlation coefficient of the capacity variation range parameter and the internal resistance change rate parameter is greater than 0.5, ; when is less than 0.5, ; when is less than 0.5, ; wherein, represents the window length weighted Pearson correlation coefficient, represents the capacity variation weight coefficient, represents the internal resistance change weight coefficient, which is obtained by regression analysis of the window length parameter and the historical attenuation rate.

[0013] As a further scheme of the present application, the cooperative mechanism of the three sets of parallel data windows comprises: outputting a reference parameter of the long-term window to the medium-term window as an initialization condition; outputting a trend parameter of the medium-term window to the short-term window as a constraint boundary; outputting a real-time correction parameter of the short-term window to the long-term window for parameter calibration; establishing a data pipeline between the three windows through the reference synchronization value, and maintaining parameter synchronization by using an exponential weighted moving average algorithm, wherein the exponential weighting coefficient is dynamically adjusted according to the window length parameter, and the weighted coefficient calculation formula is wherein represents the current window length parameter value.

[0014] The battery energy storage life prediction system based on multiple time scales is used to implement the battery energy storage life prediction method based on multiple time scales, and comprises: a multi-scale data acquisition module for deploying the three sets of parallel data windows and executing the sliding window algorithm; a joint filtering correction module for running the Kalman filtering algorithm and generating a corrected capacity mean value and an internal resistance variance; a residual cooperative analysis module for calculating the capacity residual sequence and the internal resistance residual sequence, and executing the abnormality determination; a life fusion prediction module for implementing the calculation process of the weighted fusion function and the double-decay coupled model; a dynamic configuration management module for adjusting the window length parameter and the capacity fluctuation threshold in real time according to the abnormality determination result; The output end of the multi-scale data acquisition module is connected to the input end of the joint filtering correction module; The output end of the joint filtering correction module is connected to the residual cooperative analysis module and the dynamic configuration management module, respectively; The output end of the residual cooperative analysis module is connected to the weighted function input end of the life fusion prediction module.

[0015] Compared with the prior art, the present application has the advantages and positive effects that: In the present application, dynamic monitoring is achieved by deploying multiple time scale parallel data windows, synchronous calibration of capacity baseline value and internal resistance reference value is achieved by using sliding window algorithm, dynamic correction of key parameters is achieved by combining Kalman filter algorithm, errors caused by single time dimension data fluctuation are eliminated, linkage determination mechanism of capacity residual sequence and internal resistance residual sequence is established, multi-dimensional attenuation characteristics are fused through double attenuation coupling model, multi-scale collaborative analysis of battery aging process is achieved, the limitation of traditional linear regression model on dynamic nonlinear attenuation process is broken through, the sensitivity and robustness of abnormal state detection are enhanced, the adaptability problem of static threshold determination under complex working conditions is solved, the timeliness and accuracy of remaining life prediction are improved, and the misjudgment risk caused by local parameter mutation or noise interference is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The present application is a multi-time scale based battery energy storage life prediction method flow chart; Figure 2 The present application is a data window deployment and parameter synchronous calibration flow chart; Figure 3 The present application is a Kalman filter parameter correction flow chart; Figure 4 The present application is an abnormality determination and dynamic adjustment flow chart; Figure 5 The present application is a remaining useful life prediction and fusion flow chart. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical scheme and advantages of the present application clearer, the technical scheme realized by software will be described in detail below in combination with system architecture diagram and embodiment. It should be understood that the specific embodiments described herein are only used to explain the technical scheme of the present application, and do not constitute a limitation on the scope of protection.

[0018] In the description of the present application, the system architecture relationship or data processing flow indicated by the terms "level", "module", "interface", "data flow", "client", "server" and the like are defined based on the corresponding architecture diagram or flow chart of the embodiment. This way of expression is only used to clearly explain the logical relationship of each element in the technical scheme, and is not limited to the physical deployment form. The "multiple" contains two or more technical units, including but not limited to multiple data nodes, processing threads, service instances or functional components, etc. Extensible elements, the specific number is determined according to the actual business scenario.

[0019] Please refer to Figure 1 and Figure 2 The present application provides a technical scheme: a multi-time scale based battery energy storage life prediction method, comprising the following steps: S1: Real-time collection and processing of battery capacity and internal resistance data by deploying three parallel data windows of short, medium and long term. First, the data collection unit records the actual capacity value and internal resistance measurement value of the battery at the end of each complete charge-discharge cycle. For example, for a lithium iron phosphate battery package numbered SN001, at the end of its kth cycle, its capacity is measured and recorded as 98.5 Ah and its internal resistance is measured and recorded as 12.3 mΩ by a high-precision electrochemical workstation.

[0020] After obtaining the capacity parameter and the internal resistance parameter, a sliding window algorithm is used to initialize the length parameters of the three data windows. The initial window reference length is set based on statistical analysis of historical aging data of the same type of battery under standard working conditions. For example, by observing the cycle number distribution required for 100 groups of the same type of battery to decay from 100% to 80% capacity, 10% of the average cycle life is taken as the initial window reference length. If the average life is 2000 cycles, 200 cycles are set as the initial window reference length. The short-term window length parameter is initialized to the medium-term window length parameter is initialized to and the long-term window length parameter is initialized to .

[0021] Subsequently, the capacity baseline value and the internal resistance reference value of the current cycle are synchronized and calibrated with the window configuration. The capacity baseline value is taken from the nominal capacity of the battery at the time of factory shipment, for example 100 Ah. The internal resistance reference value is taken from the initial internal resistance of the battery at the time of factory shipment, for example 10 mΩ. These reference values are used as the starting reference for parameter calculation of each window.

[0022] During the initialization of the window length parameter, a dynamic adjustment mechanism is introduced. The triggering condition of this mechanism depends on the current cycle capacity variation amplitude parameter and the internal resistance change rate parameter . The current cycle capacity value is obtained by calculating the difference between the current cycle capacity value and the last cycle capacity value. If the current cycle (e.g. the 100th cycle) capacity is 98.5 Ah and the last cycle (e.g. the 99th cycle) capacity is 98.6 Ah, then Ah / cycle. The internal resistance value is obtained by calculating the difference between the current cycle internal resistance value and the last cycle internal resistance value. If the current cycle internal resistance is 12.3 mΩ and the last cycle internal resistance is 12.25 mΩ, then mΩ / cycle.

[0023] The average value of the historical capacity variation amplitude and the average value of the historical internal resistance change rate The data is obtained by statistical averaging of the first several stable operation cycles of the battery and For example, the data of the first 50 cycles of the battery is collected Ah / cycle, and the arithmetic mean value thereof is calculated as Ah / cycle. Similarly, the data of the first 50 cycles of the battery is collected mΩ / cycle, and the arithmetic mean value thereof is calculated as mΩ / cycle.

[0024] The determination of the dimensionless shape parameter threshold is based on the standard deviation of historical data. First, the standard deviation of the historical capacity variation amplitude sequence and the standard deviation of the historical internal resistance change rate sequence are calculated. For example, based on the above-mentioned 50-cycle data, the value of Ah / cycle is calculated. Based on the above-mentioned 50-cycle data, the value of mΩ / cycle is calculated. Subsequently, these standard deviations are normalized, for example, a comprehensive standard deviation index can be defined, but here is directly compared with the normalized current variation. According to the description, is a dimensionless shape parameter threshold determined by three times the standard deviation of historical data. The expression here, and has already been a relative change. Therefore, should be set based on the historical distribution of these relative changes. Assuming that the standard deviation of the historical relative capacity change is (for example, the standard deviation is calculated for the sequence), and the standard deviation of the historical relative internal resistance change is . Then can be set to, for example, . Or, more directly, if is based on the historical data distribution of the overall index , then the standard deviation of the sequence of this index is calculated, and . To simplify the example, it is assumed here that The value of 1.5 is set according to a large number of battery aging experiment data. When the normalized comprehensive variation index on the left side of the above formula exceeds 1.5, it usually indicates that the battery aging behavior accelerates significantly or deviates from the normal path, and the data window needs to be adjusted to capture these changes. The setting of this value is obtained by backtracking analysis of 1000 groups of battery data in different aging stages, selecting the value corresponding to the normal fluctuation range covering 99.7% (i.e. three standard deviations) as the threshold reference, and then adjusting it combined with the engineering margin. For example, the value of 0.5 is statistically analyzed in historical data, and the value of 1.5 is set to .

[0025] Capacity variation influence factor and internal resistance change influence factor are preset dimensionless coefficients, and their values are calibrated by experiments. A batch of batteries of the same specification are tested under different degradation rates, and the relationship between the window length adjustment amount and , is fitted by regression analysis to determine the values of and . For example, experimental data shows that when the capacity attenuation accelerates, the adaptability adjustment of the window length has higher sensitivity to the capacity change, so is given a relatively large weight. Through multivariate regression analysis of the aging data of 50 groups of batteries under different working conditions, the values of (cycle / (Ah / cycle)) and (cycle / (mΩ / cycle)) are determined, so that the window length adjustment can better adapt to the actual change rate of the battery state. These coefficients ensure that when the capacity or internal resistance changes dramatically, the window can be scaled up to include more data points to smooth noise and capture trends.

[0026] The formula is used to determine whether the window length needs to be extended. represents the current cycle capacity variation amplitude, with the unit of Ah / cycle. For example, Ah / cycle. represents the average of the historical capacity variation amplitude, with the unit of Ah / cycle. For example, Ah / cycle. represents the current cycle internal resistance change rate, with the unit of mΩ / cycle. For example, mΩ / cycle. represents the average of the historical internal resistance change rate, with the unit of mΩ / cycle. For example, mΩ / cycle.​​ represents a dimensionless shape parameter threshold. For example, . Calculation process: left side value is . Since 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, the window length extension calculation is performed. Suppose in another cycle, Ah / cycle, mΩ / cycle. Then the left side value is At this time , the condition is met, and the window length is automatically extended.

[0027] When the extension condition is met, the new window length . Note: the original formula here is , but is usually negative, which will cause the window to shrink, so the solution here is or the sign of matches to achieve extension. Suppose it is the latter, and has taken this effect into account, or is for the decay rate (positive value). To be consistent with the "extension" goal, here we use absolute value or adjust the definition of . Given that is the variable "amplitude", it is more reasonable to use its absolute value. represents the initial window reference length, with units of cycles. For example, for the long-term window, cycles. represents the capacity variation influence factor. For example, . represents the internal resistance change influence factor. For example, . The Ah / cycle and mΩ / cycle are carried over. cycles. Therefore, the length of the long-term window is extended from 200 cycles to 210 cycles. The short-term and medium-term windows are also adjusted in proportion or based on their respective . The logic of this formula is that when the capacity or internal resistance of the battery changes significantly, by increasing the window length, more data points can be included for analysis, thereby improving the stability of parameter estimation and adaptability to trend changes. By introducing and the square root of the normalized sum of squares, the degree of change in both capacity and internal resistance is comprehensively evaluated, and compared with the threshold to determine whether the change is significant. The influence factors and wherein the contribution of different parameter variations to the window length adjustment is quantified.

[0028] Three sets of parallel data windows (short-term, medium-term, long-term) establish data pipelines through reference synchronization values. Reference synchronization values refer to defined battery state-of-health anchor points, such as expected capacity values and internal resistance values at certain cycle counts. These anchor point data are maintained and updated by the long-term window and passed to the medium-term and short-term windows. To maintain synchronization of parameters among the windows, an exponential weighted moving average (EWMA) algorithm is employed. The exponential weighting coefficient . represents the current window length parameter value. For example, for the adjusted long-term window, , its EWMA weighting coefficient . If the short-term window length , then . If the medium-term window length , then . The calculation formula of EWMA is: , wherein is the observation value (e.g., capacity baseline or internal resistance reference) of the current cycle, is the EWMA value of the last cycle, is the EWMA value of the current cycle. This formula enables the synchronized parameters to sensitively reflect the latest changes by giving higher weights to recent data (determined by , the smaller is, the larger is, and the higher the weight of recent data is), while maintaining a certain smoothness through the weighted average of historical data. The values are dynamically adjusted according to the window length, ensuring that the windows of different time scales have response speed and stability suitable for their time scales when synchronizing parameters.

[0029] The coordination mechanism between windows is implemented as follows: the long-term window (e.g., window length 210 cycles) analyzes the battery data over a long period of time in the past, calculates the average capacity decay rate and average internal resistance growth rate of the battery at the current aging stage, which are used as baseline parameters. For example, the long-term window outputs a capacity baseline value of 95.0 Ah and an internal resistance baseline value of 15.0 mΩ. These values are passed to the medium-term window as the initial calibration conditions for the medium-term window to perform trend analysis. The medium-term window (e.g., window length 100 cycles) analyzes the short-term trends of capacity decay and internal resistance growth in the next few dozen cycles based on the baseline of the long-term window and in combination with the data within its own window. For example, the medium-term window predicts that in the next 50 cycles, the capacity may decrease by 2 Ah and the internal resistance may increase by 1 mΩ. These trend parameters (such as the predicted decay slope) are output to the short-term window to provide a reasonable fluctuation range and constraint boundary for real-time correction of the short-term window. The short-term window (e.g., window length 50 cycles) focuses on the most recent battery behavior, using the trends provided by the medium-term window as a guide to monitor and correct the latest few cycles of actual observed capacity and internal resistance data at a high frequency. For example, the short-term window monitors that the actual capacity has decreased by an average of 0.08 Ah / cycle in the last 5 cycles, while the medium-term trend predicts 0.04 Ah / cycle. The short-term window takes this difference (0.04 Ah / cycle) as a real-time correction parameter and feeds it back to the long-term window. After receiving this correction parameter, the long-term window calibrates the battery aging model parameters (such as the aging rate coefficient) maintained within it, 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 fine, multi-level joint tracking and prediction of the battery state.

[0030] See Figure 1 and Figure 3 , S2: The core of the step is to dynamically correct and optimize the capacity baseline value and internal resistance baseline value calibrated in S1 using the Kalman filter algorithm. First, the capacity variation amplitude parameter and the internal resistance change rate parameter of the current cycle are extracted from each data window (for example, the short-term window, with a length parameter of , such as 50 cycles). As described in S1, if the capacity changes from 97.2 Ah to 97.1 Ah in the current cycle (e.g., cycle 150) compared to the last cycle (cycle 149), then Ah / cycle; and if the internal resistance changes from 13.5 mΩ to 13.56 mΩ, then mΩ / cycle.

[0031] These extracted parameters will serve as inputs to the state update of the Kalman filter algorithm. The Kalman filter algorithm employs a dual-state coupled equation for correction. The state equation is: . represents the joint state vector at time t, with a dimension of 2x1, specifically containing the capacity filter correction value (Q, in Ah) and the internal resistance dynamic variance value (R, in mΩ²). This is the core state that the algorithm aims to estimate and optimize. . . . represents the joint state vector at time t-1, which is the estimated value of the previous period, for example: . , where 97.3 Ah is the capacity filter correction value of the previous period, and 0.25 mΩ² is the internal resistance dynamic variance value of the previous period. is the state transition matrix, with a dimension of 2x2. Its structure is . The diagonal elements represent the capacity decay inertia factor, represent the internal resistance change inertia factor. The setting of these factors is based on the understanding of the physical and chemical processes of battery aging and historical data analysis. is usually close to 1, for example, 0.998, indicating that the capacity state has strong memory, and the current period's capacity is mainly determined by the capacity of the previous period, with slight decay. is also usually close to 1, for example, 0.995, indicating that the change of internal resistance also has hysteresis. These values are set by time series analysis of a large number of aging data of the same type of battery, extracting autocorrelation coefficients, and combining the electrochemical characteristics of the battery. For example, by performing autocorrelation analysis on the capacity data of 100 batteries per period, the average value of the first-order autocorrelation coefficient is 0.998, so is set. Similarly, the analysis of internal resistance data gives . is the control matrix, with a dimension of 2x1. Its element values are determined by the ratio of the rate correlation value to the reference synchronization value. The rate correlation value reflects the direct influence of and on the state, and the reference synchronization value serves as a normalization or scaling function. Assuming that the rate correlation value is calibrated through experiments as , and the reference synchronization value is a scalar or a vector. For simplicity, if the elements of the B matrix are , their sizes are set according to the expected influence of and on the state . For example, . The basis for setting these values is: the variation of capacity directly affects the filter correction value of capacity, so Larger; the variation of internal resistance mainly affects the dynamic variance of internal resistance, but also indirectly relates to the capacity change, so A medium value is set. The specific value is obtained by analyzing the sensitivity of input change and state change in historical data. is the capacity variation range in the current cycle, for example Ah / cycle. is the internal resistance change rate in the current cycle, for example mΩ / cycle. is the capacity-internal resistance coupling coefficient. This coefficient is calculated by analyzing the Pearson correlation coefficient of the capacity sequence and the internal resistance sequence within the current window length parameter (such as the short-term window length ), and combining the average level of this correlation in historical data. For example, in the current 50-cycle data, the correlation coefficient of the capacity change sequence and the internal resistance change sequence is -0.7. Historical data shows that for healthy batteries, the absolute value of this correlation coefficient is usually between 0.6-0.9. According to the current correlation strength and historical statistics, and possibly through a nonlinear mapping (such as a Sigmoid function) to determine the value of , ensure that it is within a reasonable range (for example, between -1 and 1). Suppose is calculated. The setting experiment of this coefficient: collect the capacity and internal resistance data of batteries with different aging degrees within the short-term window, calculate their correlation. At the same time, record the actual state of health of the battery. Through regression analysis, a model is established between correlation, window length, and a parameter representing coupling strength (i.e. ). For , it means that there is a significant negative correlation between the increase of internal resistance and the decrease of capacity. is the process noise vector, with a dimension of 2x1, representing the uncertainty of the model itself. Its covariance matrix is assumed to be , . For example, it is Ah², for example (mΩ²). The setting of these values is based on the statistical estimation of the unmodeled part of the system dynamic change, adjusted and determined by analyzing the variance of state prediction error. For example, through historical data analysis, the average error standard deviation of capacity prediction is 0.005 Ah, then .

[0032] The observation equation is: .

[0033] is the observation vector at time , with a dimension of 2x1. It is usually the capacity and internal resistance related values obtained directly by measurement or preliminary processing in the current cycle. For example, may be or its variation. In this scheme, the correction of the capacity baseline and the resistance reference is observed, so may contain the mean and variance of the capacity and resistance in the current short-term window. Assuming the observation vector is the mean of the capacity and the mean of the resistance in the current window, for example where 97.15 Ah is the average capacity in the short-term window and 13.55 mΩ is the average resistance in the short-term window. is the observation matrix, with dimension 2x2. It maps the true state vector to the observation space. If the state is directly observable, may be the identity matrix . If the observation is the mean of the resistance instead of the variance, and the state is the variance of the resistance, the corresponding row of H needs to be adjusted. According to , the capacity filter correction and the resistance dynamic variance are defined, while the observation is usually the capacity and resistance itself, so the setting of H needs to be matched. If the observation is the capacity itself and some statistic of the resistance (for example, the mean, while the state is the variance), then where k is the conversion coefficient that relates the resistance dynamic variance to the observable resistance statistic, or directly observes the resistance variance, in which case . Assuming that the observation here is the capacity filter correction itself and the resistance dynamic variance value itself (possibly obtained through some preprocessing), then . is the observation noise vector, with dimension 2x1, representing the uncertainty in the measurement process. Assuming its covariance matrix is , . For example, is Ah² (noise variance from the capacity sensor), for example, is (mΩ²)² (noise variance from the resistance measurement or its variance estimate). The setting of these values is based on the evaluation of the measurement sensor accuracy and the noise level of the data acquisition process. For example, the accuracy specification of the capacity measurement device indicates that its measurement error standard deviation is 0.1 Ah, then .

[0034] State equation Example of calculation: input values into: Ah / cycle, mΩ / cycle Calculate . Calculate the control input term . Calculate . Therefore, the state prediction (prior estimate) (process noise The mean of the process noise is usually 0 and is not explicitly added in the prediction step, whose covariance is 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 evolved from its state at the last time step under the intrinsic decay / variation trend (described by matrix ), while being modified by the actual variation amount of capacity and internal resistance (described by and coupling coefficient ) and control matrix The process noise then represents the random disturbance that this evolution model itself fails to fully capture. By introducing as a control input, the variation of capacity and internal resistance is coupled to consider the physical fact of mutual influence between the two.

[0035] The standard procedure of Kalman filtering also includes prediction error covariance , calculation of Kalman gain , state update (posterior estimate) , and update of error covariance . After a complete Kalman filtering iteration calculation, the output is the modified capacity mean (i.e., the first component of , the capacity filtering correction value ) and the internal resistance variance (i.e., the second component of , the internal resistance dynamic variance value ). For example, after the above steps (the detailed covariance and gain calculations are omitted here), the posterior estimate is obtained. The output capacity filtering correction value is 97.18 Ah, and the internal resistance dynamic variance value is 0.21 mΩ². At the same time, the reference synchronization value (which can be understood as the capacity value in ) and the rate correlation value (such as the combined term of or its impact evaluation on the state) generated in the intermediate process are also output as outputs for subsequent modules. This result shows that the Kalman filter optimally estimates the capacity and internal resistance variance state of the battery according to the latest observation data (included in ) and the system dynamic model. The capacity filtering correction value 97.18 Ah is a smooth and noise-considered estimate of the current battery capacity, while the internal resistance dynamic variance value 0.21 mΩ² quantifies the uncertainty or volatility of the internal resistance parameter. These values will be used for anomaly detection in S3 and life prediction in S4.

[0036] Please refer to Figure 1 and Figure 4,S3: Step focuses on real-time monitoring and determination of abnormal battery operation status. First, the system calls the sliding data window established in S1 and dynamically adjusts the length (for example, using a medium-term window, length Assuming 100 cycles), the most recent The measured value sequence of battery capacity and internal resistance for each cycle. For example, the average value of the capacity data within the window is 100 The arithmetic mean is Ah. Then, the measured capacity value of each cycle in the window and the window mean Compare and calculate the residual between the two . In this way, we get a length of The capacity residual sequence of Simultaneously, similar operations are performed on the internal resistance data within the window. The variance of the internal resistance within a period (or mean, according to the understanding of "the residual of the internal resistance variance and the measured value", if the measured value is a single internal resistance value, it should be the internal resistance mean. Here, according to the original text "internal resistance variance", an internal resistance variance value is calculated within the window, and then the residual 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 internal resistance measured value. Here, it is assumed that the internal resistance measurement value is calculated. and the average internal resistance within the window The residual of the internal resistance residual sequence is generated For example, the mean internal resistance mΩ.

[0037] Next, based on the capacity variation parameters extracted from S1 (For example, the current cycle Ah / cycle), set a dynamic capacity fluctuation threshold. The "capacity fluctuation threshold" here is a complex concept. In actual operation, it may refer to the threshold used to determine whether a single point residual exceeds the limit. A more specific dynamic adjustment method is given later for the "residual linkage condition threshold". Here, "setting the capacity fluctuation threshold based on the capacity change amplitude parameter" may refer to The size of will affect the expectation of normal fluctuation range. The core logic of abnormal judgment is: when the absolute value of the capacity residual for three consecutive cycles in the capacity residual sequence exceeds the preset single-point capacity residual threshold, it is considered that the capacity has sustained abnormal fluctuations. 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 is a multiple factor, for example, 3. Assuming that the mean of the capacity residual series is 0.01 Ah and the standard deviation is 0.05 Ah by analyzing historical data, then Ah. When the capacity residual of the last three consecutive periods , , When the condition is met, the system will further link the internal resistance residual series for comprehensive judgment. This means that not only the capacity is abnormal, but also the internal resistance parameter also shows a coordinated change or abnormal sign, which can finally trigger the abnormal judgment. The specific way of this “linkage internal resistance residual series” can be to check whether the internal resistance residual exceeds its own threshold at the corresponding time point, or whether the change trend of the internal resistance is consistent with the mode of capacity abnormality (for example, rapid capacity decline corresponds to rapid internal resistance rise).

[0038] The dynamic adjustment method of the capacity fluctuation threshold is carried out for the “residual linkage condition threshold” . The specific execution process is as follows: monitor the capacity residual of the last three consecutive periods. For each residual, judge whether it meets the condition . Among them, represents the capacity residual at time , for example, in the period, Ah. represents the standard deviation of the capacity residual series (for example, the last 100 periods of data). Through calculation, we get Ah. represents the mean of the capacity residual series. Through calculation, we get Ah. is the sensitivity coefficient. The setting of this coefficient aims to balance the sensitivity and false positive rate of detection. Through ROC curve analysis of real failure cases and normal fluctuation data in historical data, a suitable value is selected. For example, experiments show that when , the detection rate of the system for early soft failure reaches 90%, while the false positive rate is less than 5%, so is set. Calculate the judgment condition: Ah. Assuming that the capacity residuals of the last three periods are Ah, Ah, Ah. Since , , , the condition of the last three consecutive periods is met.

[0039] At this time, the adjustment calculation of the residual linkage condition threshold is triggered: (Original text , but usually contains the current point, here i=1 to 3, i.e. t-1, t-2, t-3). To be consistent with the aforementioned "last three cycles" , the summation should be . Here we use for . represents the initial residual linkage condition threshold. This threshold is set based on statistical analysis of the residual linkage of battery capacity and internal resistance under a large number of normal conditions, representing the baseline level of the correlation fluctuation of the two under normal conditions. For example, through historical data analysis, set . The setting experiment: in a large number of normal operation data of the battery, calculate a certain joint statistic (such as product or weighted sum) of the capacity residual and the internal resistance residual, is set to the 95th percentile of the distribution of this statistic, ensuring that under normal conditions, the linkage condition is not easily triggered. represents the threshold adjustment factor, whose calculation formula is . represents the sliding standard deviation of the internal resistance variance value, with a unit of mΩ² / cycle, and the calculation window length is 10 cycles. Here, "internal resistance variance value" should refer to the time series of internal resistance dynamic variance value output in S2. For example, the internal resistance dynamic variance value sequence of the past 10 cycles is mΩ². Calculate the standard deviation of this sequence to get mΩ² / cycle (note the unit, the unit of the standard deviation should be consistent with the original data unit, i.e. mΩ². If the original refers to the variance of the internal resistance itself, the unit is mΩ², and the standard deviation is also mΩ². If it refers to the variance of the internal resistance change rate, the unit is (mΩ / cycle)². Here, according to the literal "sliding standard deviation of internal resistance variance value", the unit of the standard deviation is also mΩ²). Calculate . Calculate . . The adjusted residual linkage condition threshold is 0.5234. The logic of this formula is: when the capacity residual is consistently large, it indicates that the battery state may be undergoing significant changes, at which point the threshold for linkage determination should be appropriately increased to more accurately capture the true coupling anomaly and avoid false positives due to large fluctuations in a single parameter. The adjustment factor introduces the volatility of the internal resistance variance , meaning that if the internal resistance itself also exhibits unstable characteristics , the adjustment range of the threshold will be larger.

[0040] Abnormality determination result grading method: when the number of consecutive overruns of the capacity residual sequence (for example, When the number of consecutive occurrences of the abnormal trigger mechanism reaches 120% of the preset "abnormal trigger mechanism threshold" (assuming this threshold is 5 times), that is, the number of consecutive violations reaches When the system automatically raises the abnormality level, for example, from "Caution" to "Warning". At the same time, the 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: . Represents the inertia correction factor. Its value is negatively correlated with the fluctuation range of the dynamic variance value of the internal resistance. The fluctuation range is the dynamic variance value of the internal resistance ( ) of the sliding standard deviation (similar to the above consistent, the calculation window length is 10 cycles). For example, mΩ². The negative correlation can be expressed as or ,in and is a preset constant. For example, , .but The setting experiment of the inertia correction factor: select multiple sets of battery data, test different The impact of the value on the Kalman filter tracking performance is optimized by minimizing the tracking error and . represent The joint state vector at time , represent The joint state vector at time . For example, (From S2 ), (From S2 ). Then the inertia correction term is This term is added to the state equation to enhance the filter's ability to track the trend when the battery state changes rapidly or deviates from the trend. Hour, Larger, indicating more trust in the recent trend of change; when the internal resistance fluctuates violently ( When The smaller the value, the lessening the impact of the trend term and preventing noise interference. Ultimately, the abnormality determination result (such as "Normal," "Caution," "Warning," or "Critical") is output to guide weight adjustment 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 transmitted to S4.

[0041] See also Figure 1 and Figure 5S4: Step according to the abnormality determination result output in S3, call the corresponding weighted fusion function, and input the current cycle capacity variation amplitude parameter obtained in S1 (e.g. Ah / cycle) and the internal resistance change rate parameter (e.g. mΩ / cycle) into the double attenuation coupling model, and finally calculate and output the remaining useful life (RUL) of the battery.

[0042] The double attenuation coupling model contains a capacity-internal resistance cross-verification structure, and calculates the RUL in a hierarchical fusion manner. In the first layer, the capacity attenuation dominated life : is calculated (using Because is negative and life is positive) represents the current battery capacity measured value. This value is obtained from the data acquisition module, for example, in the first cycle, Ah (the filtered value in S2 or the directly measured value). represents the capacity life termination threshold. This threshold is preset according to the battery specifications and application requirements, and is usually 80% or 70% of the rated capacity. For example, if the rated capacity of the battery is 100 Ah, the EOL threshold is set to 80 Ah, i.e. 80 Ah. Therefore Ah. represents the capacity attenuation correction coefficient obtained by regression analysis of battery aging experiments. This coefficient is used to correct the linear extrapolation attenuation rate to more accurately reflect the real nonlinear attenuation process. Its setting process is as follows: a batch of same type batteries are subjected to complete aging experiments at different rates and different temperatures, and the capacity attenuation curves of the whole life cycle are recorded. Compare these curves with the simple linear extrapolation results based on , and obtain an average correction coefficient by least squares fitting. For example, experimental data analysis shows that linear extrapolation often overestimates early life or underestimates late life, and by introducing (unitless), the prediction result can be closer to the actual aging trajectory. This coefficient less than 1 indicates that the actual attenuation may be slightly faster than the short-term linear prediction, or is used to adjust the representativeness of . represents the current cycle capacity variation amplitude. For example Ah / cycle. Then Ah / cycle. Calculate cycles.

[0043] In the second layer, the internal resistance growth dominated life : is calculated. represents the EOL threshold of internal resistance. This threshold is usually set as 2 or 3 times of the initial internal resistance, or determined according to the safety and performance requirements of specific applications. For example, if the initial internal resistance is 10 mΩ, the EOL threshold is set as 200% of the initial value, i.e. 20 mΩ (or directly set as 20 mΩ, i.e. increase by 100%). Here we assume 10 mΩ. represents the measured value of the current battery internal resistance. This value is obtained from the data acquisition module, for example, in the first cycle, 20 mΩ (filtered value in S2 or directly measured value). represents the internal resistance decay correction coefficient obtained through regression analysis of battery aging experiments. Similar to , obtained by analyzing the internal resistance change curve in the aging experiment, used to correct the prediction of linear growth rate. For example, experimental data analysis sets (unitless). This coefficient greater than 1 indicates that the actual internal resistance growth may be slightly slower than the short-term linear prediction, or used to calibrate representative. represents the internal resistance change rate in the current cycle. For example mΩ / cycle. Calculate cycles.

[0044] The final life is calculated by the weighted value . represents the weight coefficient, which ranges from 0.5 to 0.8. The specific value of this weight coefficient is determined by the severity level of the abnormality determination result output by S3. For example, set the rules as follows: if the abnormality level is "normal", . If the abnormality level is "attention", . If the abnormality level is "warning", . (Assuming S3 determines "warning") If the abnormality level is "serious", . (If the capacity decay is a more major failure mode, or its prediction is more reliable, will be biased ). The setting of the weight is based on: by analyzing historical failed battery data, statistics show that under different abnormality levels, the main factor leading to the end of life is the capacity decay or the internal resistance exceeding the limit. For example, data shows that in the "warning" level battery, about 75% of the cases eventually fail due to insufficient capacity, so at this time . The experimental verification process of these weights includes: collecting a large number of battery samples with clear abnormality level labels and final life data, and analyzing the main factors leading to the end of life under different Backtest the values ​​and select the one that minimizes the RUL prediction error. A sequence of values.

[0045] Table 1 Weight coefficients Correspondence table with abnormal level

[0046] As shown in Table 1, the recommended weight coefficient values ​​for different abnormality levels are listed. Assuming that the current abnormality level is "warning", then .calculate cycles. Therefore, the predicted remaining service life of the battery is about 180 cycles. The logic of the RUL calculation formula is that it takes into account the two main battery aging modes of capacity decay and internal resistance growth. By calculating the life expectancy based on these two modes separately and , and then different weights are assigned according to the current health status of the battery (reflected by the abnormality level) for fusion to obtain a more comprehensive RUL assessment. and Corrected the linear extrapolation of short-term rate of change, while the weight The relative importance of the two aging mechanisms in the final lifespan assessment is dynamically adjusted.

[0047] Regarding the adaptive adjustment strategy of the weighted fusion function, it describes how to adjust the capacity variation parameter and internal resistance change rate parameters The correlation between them is used to process these two parameters, which can be understood as a preprocessing of the input parameters or a feature fusion method. It can be used as an additional health indicator or to further adjust the parameters in the double decay coupling model (e.g. ). First calculate and In the current analysis window (e.g., mid-term window, length ) within the Pearson correlation coefficient For example, in the past During the period, the collected sequence sum Pearson correlation coefficient calculated from the sequence .because It is usually negative (capacity decreases, internal resistance increases), where the strategy may target its absolute value or sign adjusted value. Refers to its absolute value, that is .according to The value of is calculated using different fusion methods :1. When When (original text is , considering that in practice the correlation is mostly negative, should be or ), the product fusion method is adopted . Assuming , then . 2. When , the linear weighted method is adopted . Since belongs to this interval. represents the capacity variation weight coefficient, represents the internal resistance variation weight coefficient. These coefficients are obtained by regression analysis on historical decay rate data and corresponding window length parameters. For example, the analysis results show that under the condition of moderate correlation, the contribution of capacity variation is slightly greater than that of internal resistance variation, and is set to , . The setting experiment of these coefficients: collect a large amount of data and its correlation under different window lengths, and the corresponding actual battery health degradation rate. When is in the medium range, the multivariate regression is optimized so that the linear combination has the maximum correlation with the actual degradation rate. Then . 3. When , the maximum selection method is adopted (Note that the unit is matched, for example, introduce a coefficient k to convert to a comparable order of magnitude with , or directly compare the normalized values). The original text is , but is negative, so use the absolute value or only compare the decay rate. Here it is assumed that where is the internal resistance variation rate after a certain scale transformation. To simplify, if the original variation is directly compared (without considering unit and scale differences, which needs to be cautious in practical application), and use its absolute value: .

[0048] In this example, , so . This value can be used as a comprehensive current cycle degradation rate indicator. It can be used: a. Record and track its trend of change as an additional health state monitoring parameter. b. Used to fine-tune the weight in the RUL model. For example, if significantly increases, it may indicate that the battery is accelerating, at which time the value of the parameter (in the range of 0.5-0.8) is dynamically fine-tuned to make it more biased towards the life estimation dominated by the parameter (capacity or internal resistance) that degrades faster.c. Adjustment correction coefficient For example, if continues to be high, it may mean that the actual decay is faster than expected by the model, in which case or may be appropriately reduced (depending on which parameter contributes more). This adaptive adjustment strategy, by analyzing the correlation between and judges the consistency or independence of the degradation of the two, and accordingly selects the appropriate fusion method to quantify the current comprehensive degradation intensity. This way makes the system better cope with the complex and diverse correlation patterns between capacity and internal resistance changes under different aging stages or different aging mechanisms. The output of this step, the battery remaining useful life (RUL), for example, 180 cycles, provides users or battery management systems with a quantitative expectation of how long the battery can continue to serve, which is a key basis for subsequent maintenance decisions and replacement plans.

[0049] The multi-time-scale-based battery energy storage life prediction system is used to execute the multi-time-scale-based battery energy storage life prediction method described above, and the system comprises: a multi-scale data acquisition module for deploying three parallel data windows and executing a sliding window algorithm; a joint filtering correction module for running a Kalman filtering algorithm and generating corrected capacity mean and internal resistance variance; a residual cooperative analysis module for calculating capacity residual sequence and internal resistance residual sequence, and performing abnormality judgment; a life fusion prediction module for implementing a weighted fusion function and a double-decay coupling model calculation process; a dynamic configuration management module for adjusting window length parameters and capacity fluctuation thresholds in real time according to the abnormality judgment result; The output end of the multi-scale data acquisition module is connected to the input end of the joint filtering correction module; The output end of the joint filtering correction module is respectively connected to the residual cooperative analysis module and the dynamic configuration management module; The output end of the residual cooperative analysis module is connected to the weighted function input end of the life fusion prediction module.

[0050] The above embodiments demonstrate the preferred implementation of the present application, and any equivalent adjustment of the technical solutions based on software engineering methods falls within the protection scope, including but not limited to: implementing algorithm logic in different programming languages, service reconstruction of functional modules, adjustment of data interaction protocols, optimization of resource scheduling strategies, etc. Any implementation derived through reasonable modification of the data processing flow, service calling link or system architecture level without deviating from the technical core of the present application shall be considered within the protection scope defined by the claims of the present application.

Claims

1. A battery energy storage life prediction method based on multiple time scales, characterized by: The following steps are involved: S1: Deploy three sets of parallel data windows: short-term, medium-term, and long-term, to obtain the capacity parameters and internal resistance parameters of the battery during operation. Use the sliding window algorithm to initialize the window length parameters, and synchronize the capacity baseline value and internal resistance reference value of the current cycle with the window configuration. S2: Calling a Kalman filter algorithm to correct the capacity baseline value and the internal resistance reference value, extracting a capacity variation amplitude parameter and an internal resistance variation rate parameter based on the window length parameter, and outputting a corrected capacity mean value and internal resistance variance; S3: Using a sliding window algorithm to intercept data within the window, the residual between the capacity mean and the measured value is calculated to generate a capacity residual sequence. Simultaneously, the residual between the internal resistance variance and the measured value is calculated 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 abnormality determination. S4: calling a weighted fusion function according to the abnormality determination, inputting the capacity variation amplitude parameter and the internal resistance variation rate parameter into a double-attenuation coupling model, and outputting the remaining battery life.

2. The battery energy storage life prediction method based on multiple time scales according to claim 1 is 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 capacity mean and internal resistance variance include a capacity filter correction value, an internal resistance dynamic variance value, a reference synchronization value, and a rate correlation value; the capacity fluctuation threshold includes a continuous over-limit times threshold, a residual linkage condition threshold, and an abnormal trigger mechanism threshold; the remaining battery life specifically refers to a capacity attenuation coefficient and an internal resistance coupling factor; A data pipeline is established between the three groups of parallel data windows through the reference synchronization value, and a sliding average algorithm is used to maintain parameter synchronization; The dual-attenuation coupling model includes a capacity-internal resistance cross-validation structure.

3. The battery energy storage life prediction method based on multiple time scales according to claim 2 is characterized in that: When the sliding window algorithm initializes the window length parameter, a dynamic adjustment mechanism is adopted: when the capacity variation parameter The internal resistance change rate parameter satisfy When the window length is automatically extended to ; in, Represents the capacity change range of the current cycle, Represents the mean of historical capacity changes, Represents the rate of change of internal resistance in the current cycle, Represents the average value of the historical internal resistance change rate, represents the dimensionless shape parameter threshold determined by three times the standard deviation of the historical data, Represents the initial window base length, represents the capacity change influencing factor, Represents the internal resistance change influencing factor.

4. The battery energy storage life prediction method based on multiple time scales according to claim 3 is characterized in that: The Kalman filter algorithm correction process adopts a dual-state coupling equation: state equation ; Observation equation ; in, Represents the joint state vector at time t, including the capacity filter correction value and the dynamic variance value of the internal resistance, Represents the state transfer matrix, whose diagonal elements represents the capacity decay inertia factor, Represents the inertia factor of internal resistance change, represents the control matrix, whose element value is determined by the ratio of the rate-related value to the reference synchronization value. Represents the capacity-internal resistance coupling coefficient, which is calculated by the correlation between the window length parameter and the historical data. represents the process noise vector, represents the observation vector, represents the observation matrix, represents the observation noise vector.

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

6. The battery energy storage life prediction method based on multiple time scales according to claim 5 is characterized in that: The dual-attenuation coupling model adopts a layered fusion structure: the first layer calculates the capacity attenuation-dominated lifetime. ; The second layer calculates the internal resistance decay dominant life ; The final life weighted value ; in, Represents the current measured value of battery capacity. represents the capacity life end threshold, Represents the capacity attenuation correction coefficient obtained through regression analysis of battery aging experiments, Represents the capacity change range of the current cycle, Represents the current measured value of the battery internal resistance, Represents the end-of-life threshold of internal resistance, Represents the internal resistance attenuation correction coefficient obtained through regression analysis of battery aging experiments, Represents the rate of change of internal resistance in the current cycle, Represents the weight coefficient, which is determined by the severity level of the abnormal judgment result.

7. The battery energy storage life prediction method based on multiple time scales according to claim 6 is characterized in that: The abnormality determination result grading method includes: when the number of consecutive exceeding limits of the capacity residual sequence reaches 120% of the abnormality triggering mechanism threshold, automatically raising the abnormality level and activating the second-order Kalman filter; The second-order Kalman filter adds an inertia correction term to the state equation ,in, Represents the inertia correction factor, and its value is negatively correlated with the fluctuation amplitude of the dynamic variance value of the internal resistance. The fluctuation amplitude is the sliding standard deviation of the internal resistance variance value, and the calculation window length is 10 cycles. represents the joint state vector at time t-1, Represents the joint state vector at time t-2.

8. The battery energy storage life prediction method based on multiple time scales according to claim 7 is characterized in that: The weighted fusion function adopts an adaptive adjustment strategy: when the correlation coefficient between the capacity variation amplitude parameter and the internal resistance variation rate parameter When using product fusion ; when When using linear weighting ; when When the maximum selection method is used ; in, represents the window length-weighted Pearson correlation coefficient, represents the capacity change weight coefficient, Represents the weight coefficient of internal resistance change, which is obtained through regression analysis of window length parameter and historical decay rate.

9. The battery energy storage life prediction method based on multiple time scales according to claim 8 is characterized in that: The coordination mechanism of the three sets of parallel data windows includes: the long-term window outputs the benchmark parameters to the medium-term window as initialization conditions; The mid-term 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; The data pipeline is established between the three windows through the reference synchronization value, and the exponential weighted sliding average algorithm is used to maintain parameter synchronization. The exponential weighting coefficient is dynamically adjusted according to the window length parameter. The weighting coefficient calculation formula is: ,in Represents the current window length parameter value.

10. A battery energy storage life prediction system based on multiple time scales, characterized by: The system is used to implement the battery energy storage life prediction method based on multiple time scales according to any one of claims 1 to 9, and 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; A joint filtering correction module, configured to run the Kalman filtering algorithm and generate a corrected capacity mean and internal resistance variance; A residual collaborative analysis module, configured to calculate the capacity residual sequence and the internal resistance residual sequence, and perform the abnormality determination; A life fusion prediction module, used for implementing the calculation process of the weighted fusion function and the double decay coupling model; A dynamic configuration management module, configured to adjust the window length parameter and the capacity fluctuation threshold in real time according to the abnormality determination result; The output end of the multi-scale data acquisition module is connected to the input end of the joint filtering correction module; The output ends of the joint filtering and correction module are respectively connected to the residual collaborative analysis module and the dynamic configuration management module; The output end of the residual collaborative analysis module is connected to the weighted function input end of the life fusion prediction module.

Images