Lithium ion battery health state prediction method and system based on multi-model collaborative estimation

Abstract

The invention relates to the technical field of health prediction of an energy storage system, and provides a lithium ion battery health state prediction method and system based on multi-model collaborative estimation, and the method comprises the steps: extracting health factors of a lithium ion battery, and carrying out the preprocessing; health factors are screened through correlation coefficients, it is determined that the health factors have multiple collinearity through variance expansion factors, a ridge regression health state estimation model is constructed through a ridge regression algorithm, the optimal solution of the ridge regression health state estimation model is solved, and the optimal ridge regression health state estimation model is obtained; and removing multiple collinearity through the optimal ridge regression state-of-health estimation model, and performing prediction through the improved lithium ion battery state-of-health prediction model to obtain a lithium ion battery state-of-health prediction value. According to the method, the prediction precision is remarkably improved, the health state of the lithium ion battery can be accurately predicted, and high robustness is achieved; the system can be deployed in energy storage power stations with different topological structures such as a centralized type, a string type and the like, and targeted modification is not needed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of health prediction of energy storage systems, and particularly relates to a lithium ion battery health state prediction method and system based on multi-model collaborative estimation. BACKGROUND

[0002] Electrochemical energy storage power stations are the core support of new power systems due to their flexible adjustment characteristics, but the increase in operation and maintenance costs caused by battery degradation, and the inability to accurately predict the health state may lead to early retirement or overrunning of the battery. At present, energy storage power stations mainly rely on the basic monitoring of the battery management system (BMS), but the traditional BMS can only output real-time parameters such as voltage and current, and cannot realize long-term health prediction, so high-precision health prediction technology is urgently needed. The current battery health state prediction technology for energy storage power stations has obvious defects. Model-driven methods are based on the construction of physical models of battery electrochemical mechanisms (such as SEI film growth and lithium dendrite formation), and estimate the health state by solving complex equation sets, which has large simplification errors and ignores nonlinear factors in the battery aging process, such as temperature fluctuations and changes in charge and discharge rates, resulting in large estimation errors; the calculation is complex, and high-precision models require a long time to solve, which cannot meet the real-time requirements.

[0003] Single data-driven methods are based on historical operation data, and use machine learning or deep learning models such as LSTM and Informer to establish a mapping relationship between parameters and health state, but cannot handle the multicollinearity between health factors (HI), and the long-period prediction accuracy is low; the selection of health factors depends on experience and lacks quantitative screening standards, resulting in poor model generalization ability and inability to adapt to different types of energy storage power stations; the health state (State of Health, SOH) data of energy storage systems has time sequence characteristics, and strong feature extraction capability is needed for data feature extraction to improve its robustness and generalization ability. SUMMARY

[0004] The present application aims to at least solve one of the technical problems existing in the related art. To this end, the present application provides a lithium ion battery health state prediction method and system based on multi-model collaborative estimation, which screens strong correlation health factors through the Spearman correlation coefficient, excludes weakly correlated interference terms, and solves the problem of strong subjectivity and dependence on experience in traditional health factor selection. Through ridge regression, the problem of multicollinearity among health factors is solved, and then by combining the sliding window strategy, time convolution network (TCN) and GRU model, the long-term time sequence dependence of the health state is captured, realizing the full-cycle coverage of "current health state estimation + future degradation trend prediction", and filling the technical gap of traditional data-driven models in "multicollinearity processing + long-term prediction accuracy". The present application significantly improves the prediction accuracy and can accurately predict the health state of lithium ion batteries. Through ridge regression to process multicollinearity, the health factor is screened by the Spearman coefficient, combined with the sliding window strategy and time convolution network (TCN), the improved lithium ion battery health state prediction model has high robustness; it can be deployed in energy storage power stations with different topological structures such as centralized and group string, without the need for targeted modification.

[0005] The present application provides a lithium ion battery health state prediction method based on multi-model collaborative estimation, comprising: S1: extracting health factors of lithium ion batteries and preprocessing the health factors; S2: calculating the correlation coefficient of the preprocessed health factors, and selecting health factors according to the correlation coefficient to obtain strong correlation health factors; S3: calculating the variance inflation factor of the strong correlation health factors and setting a variance inflation factor threshold, if the variance inflation factor is greater than the variance inflation factor threshold, the strong correlation health factors have multicollinearity; S4: using a ridge regression algorithm to construct a ridge regression health state estimation model, solving the optimal solution of the ridge regression health state estimation model, and obtaining an optimal ridge regression health state estimation model; S5: inputting the strong correlation health factors with multicollinearity into the optimal ridge regression health state estimation model to remove the multicollinearity of the strong correlation health factors, and obtaining a lithium ion battery health state estimation value; S6: predicting the lithium ion battery health state estimation value through the improved lithium ion battery health state prediction model to obtain a lithium ion battery health state prediction value.

[0006] Further, the health factors include direct health factors and indirect health factors, the direct health factors include internal resistance, maximum temperature, average temperature, minimum temperature, charging duration and temperature range, the indirect health factors include the time difference of equal discharge voltage interval in constant current charging and discharging working stage, the time difference of equal charging voltage interval, the discharge capacity difference of equal voltage interval, the capacity difference of equal charging voltage interval, the time of temperature reaching peak value, the time of voltage reaching lower limit cut-off potential, the first intersection point ordinate of capacity increment curve and curve , the second intersection point ordinate of capacity increment curve and curve , the peak height of capacity increment curve and the peak position of capacity increment curve, the area surrounded by capacity increment curve and curve , the slope above curve peak and the slope below curve peak.

[0007] Further, the direct health factors are obtained by measuring the actual operation of lithium ion batteries in energy storage power stations, and the indirect health factors are extracted from voltage, current and temperature.

[0008] Further, the preprocessing adopts the quartile range method to detect outliers of health factors, selects health factor data points below lower limit or above upper limit as outliers, and fills in the average values of two points before and after the outliers after removing the outliers.

[0009] Further, in the S2 step, the correlation coefficient of two health factors is calculated by the Spearman correlation coefficient method.

[0010] Further, the calculation expression of the variance inflation factor is: wherein, is the determination coefficient of least squares regression on other independent variables, is the sample size, is the model prediction value of the th observation value, is the true dependent variable value of the th observation value, is the mean value of dependent variables of all observation values, is the variance inflation factor of the th observation value.

[0011] Further, the ridge regression health state estimation model is: wherein, is the weight coefficient of the ridge regression health state estimation model, is the bias term of the ridge regression health state estimation model.​ As a strongly associated health factor, This indicates the health status of the lithium-ion battery.

[0012] Furthermore, in step S4, solving for the optimal solution of the ridge regression health state estimation model includes: S41: Add the L2 norm of the weight coefficients to the loss function of the ridge regression health status estimation model and multiply it by the regularization term of the regularization coefficient to obtain the regularized loss function. S42: Set the regularization coefficient to 1. When the regularization loss function is minimized, the optimal solution of the weight coefficients and the optimal solution of the bias term of the ridge regression health state estimation model are obtained.

[0013] Furthermore, the improved lithium-ion battery health status prediction model includes a sliding window layer, a temporal convolutional network layer, and two GRU layers. The temporal convolutional network layer includes causal convolution, dilated convolution, and residual connections connected in sequence. The input data sequence is divided based on a sliding window layer. Key time series features of the data sequence processed by the sliding window layer are extracted through a temporal convolutional network layer. The extracted key time series features are input into two GRU layers. The GRU layers use a gating mechanism to effectively capture long-term dependencies in the time series data and obtain the predicted value of the health status of lithium-ion batteries.

[0014] This invention also provides a multi-model collaborative estimation lithium-ion battery health state prediction system for performing the aforementioned multi-model collaborative estimation lithium-ion battery health state prediction method, comprising: An extraction module extracts health factors from lithium-ion batteries and preprocesses these health factors. The first calculation module calculates the correlation coefficient of the preprocessed health factors and selects health factors based on the correlation coefficient to obtain strongly correlated health factors. The second calculation module calculates the variance inflation factor of the strongly correlated health factors and sets a variance inflation factor threshold. If the variance inflation factor is greater than the variance inflation factor threshold, then the strongly correlated health factors have multicollinearity. The module constructs a ridge regression health status estimation model using the ridge regression algorithm, solves for the optimal solution of the ridge regression health status estimation model, and obtains the optimal ridge regression health status estimation model. The estimation module inputs strongly correlated health factors with multicollinearity into the optimal ridge regression health status estimation model, removes the multicollinearity of the strongly correlated health factors, and obtains the estimated value of the lithium-ion battery health status. The prediction module uses an improved lithium-ion battery health status prediction model to predict the estimated value of the lithium-ion battery health status, thereby obtaining the predicted value of the lithium-ion battery health status.

[0015] The above-described one or more technical solutions in the embodiments of the present invention have at least one of the following technical effects: This invention uses Spearman correlation coefficients to screen strongly correlated health factors and eliminate weakly correlated interference terms, addressing the issues of traditional health factor selection relying on experience and being highly subjective. It resolves multicollinearity among health factors through ridge regression, and further captures long-term temporal dependencies of health status by combining a sliding window strategy, a temporal convolutional network (TCN), and a GRU model. This achieves full-cycle coverage of "current health status estimation + future decline trend prediction," filling the technical gap in "multicollinearity handling + long-term prediction accuracy" in traditional data-driven models. This invention significantly improves prediction accuracy, accurately predicting the health status of lithium-ion batteries. By using ridge regression to handle multicollinearity, Spearman coefficients to screen health factors, and combining a sliding window strategy and a temporal convolutional network (TCN), the improved lithium-ion battery health status prediction model exhibits high robustness. It can be deployed in energy storage power stations with different topologies, such as centralized and string systems, without requiring specific modifications.

[0016] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a multi-model collaborative estimation method for predicting the health status of lithium-ion batteries provided by the present invention.

[0019] Figure 2 This is a comparison chart of the predicted health status of lithium-ion batteries according to embodiments of the present invention.

[0020] Figure 3 This is a comparison chart of the prediction error indicators of the health status of lithium-ion batteries in embodiments of the present invention.

[0021] Figure 4 This is a schematic diagram of the structure of a multi-model collaborative estimation lithium-ion battery health status prediction system provided by the present invention.

[0022] Figure label: 101. Extraction module; 102. First calculation module; 103. Second calculation module; 104. Construction module; 105. Estimation module; 106. Prediction module. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention. The following embodiments are used to illustrate this invention but cannot be used to limit the scope of this invention.

[0024] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0025] The following is combined Figures 1 to 4 This invention describes a multi-model collaborative estimation method and system for predicting the health status of lithium-ion batteries.

[0026] like Figure 1 As shown, a multi-model collaborative estimation method for predicting the health status of lithium-ion batteries includes: S1: Extract the health factors of lithium-ion batteries and pre-treat the health factors; Health factors include direct and indirect health factors. Direct health factors include battery internal resistance, maximum temperature, average temperature, minimum temperature, charging time, and temperature range. Indirect health factors include the time difference between equal discharge voltage intervals, the time difference between equal charge voltage intervals, the time difference between equal voltage interval discharge capacity, the time difference between equal charge voltage interval capacity, the time to reach peak temperature, the time to reach the lower cutoff potential, and the capacity increment (IC) curve. The first intersection point ordinate Capacity increment curve and curve The second intersection point ordinate Peak height of capacity increment curve and the peak position of the capacity increment curve Capacity increment curve and curve Area of ​​enclosure Slope of the curve peak and the slope of the curve peak .

[0027] Direct health factors are obtained by measuring lithium-ion batteries in actual operation of energy storage power stations, while indirect health factors are extracted from voltage, current, and temperature.

[0028] In the actual operation of batteries in energy storage power stations, capacity measurement requires a full charge / discharge cycle, internal resistance measurement requires external equipment, and in the early stages of battery degradation, the performance degradation of the Direct Health Indicator (DHI) is not obvious. Therefore, it is necessary to extract various indirect health indicators (Refined Health Indicators, RHI) from data such as voltage, current, and temperature.

[0029] The battery's State of Charge (SOC) varies between 20% and 80%, exhibiting changes in voltage, current, and temperature. Voltage increases during charging and decreases during discharging. Due to the constant current charging / discharging mode, the current remains constant during both charging and discharging. Temperature increases during charging and decreases during discharging. The required High Intake (HI) needs to be extracted within a suitable charge / discharge data range. The effective charging and discharging segments are filtered as follows: in, For the first Battery No. 1 State of charge at time t, For the first Battery No. 1 Current at any given moment.

[0030] During continuous battery use, internal resistance and charging time gradually increase. Both high and low temperatures accelerate battery lifespan degradation and exacerbate battery inconsistencies. Therefore, the changing characteristics of these parameters can be used to identify battery degradation status. After standardization, the battery's direct health factors—internal resistance, maximum temperature, average temperature, minimum temperature, charging time, and temperature range—show the increasing trends as the number of charge-discharge cycles increases.

[0031] During charging, the battery temperature rises continuously due to Joule heating generated by the current passing through its internal resistance. During discharging, the battery temperature also rises due to Joule heating and exothermic reactions caused by the current passing through its resistance. However, the temperature gradually decreases after discharging. As the battery ages, the time to reach peak temperature gradually shortens; this time is denoted as the peak temperature time. During charging, lithium ions migrate from the positive electrode to the negative electrode through the electrolyte and embed themselves in the negative electrode material (usually graphite). During this process, the solute and solvent in the electrolyte are reduced and decomposed on the surface of the negative electrode, forming a solid electrolyte interphase (SEI) layer. Due to the presence of the SEI layer, the battery's internal resistance increases, and some lithium ions are consumed during long-term use, leading to battery capacity decay. The time it takes for the battery to reach its lower cutoff potential during charging and discharging gradually shortens; this time is denoted as the time it takes for the voltage to reach the lower cutoff potential. The calculation expression is: in, For the first Battery No. 1 The time it takes for the temperature to reach its peak in the next cycle. For the first Battery No. 1 The time it takes for the voltage to reach the lower cutoff potential in the next cycle. For the first Battery No. 1 The sampling time of the next cycle For the first Battery No. 1 Next cycle sampling time voltage, For the first Battery No. 1 Next cycle sampling time voltage, For the first Battery No. 1 Next cycle sampling time temperature, For the first Battery No. 1 Next cycle sampling time The temperature.

[0032] The lithium battery degradation state identification method based on the monitoring parameter change characteristics takes the equal discharge voltage interval time difference (TIEDVD), equal charge voltage interval time difference (TIECVD), equal voltage interval discharge capacity difference (EVIDCD), and equal charge voltage interval capacity difference (CDECVI) during the constant current charge and discharge operation as indirect health factors.

[0033] The time difference between equal discharge voltages gradually decreases with the increase of battery charge-discharge cycles. The calculation method is to take the time difference when the voltage changes between 2.8V and 3.2V when the current I < 0. The calculation expression is: in, For the first Battery No. 1 The time difference between equal discharge voltage intervals in each cycle. For the first Battery No. 1 At the moment when the voltage of the next cycle is 3.2V, For the first Battery No. 1 The moment when the voltage of the next cycle is 2.8V.

[0034] The time difference between charging and discharging voltages gradually decreases with the increase of battery charge-discharge cycles. The calculation method is to take the time difference between the voltage change from 3.2V to 3.5V when the current I > 0. The calculation expression is: in, For the first Battery No. 1 The time difference between equal charging voltage intervals in each cycle For the first Battery No. 1 The moment when the voltage of the next cycle is 3.5V.

[0035] The capacity difference during equal voltage interval discharge gradually decreases with increasing battery charge-discharge cycles. The calculation method involves taking the capacity difference when the voltage varies between 2.8V and 3.2V with current I < 0. The calculation expression is as follows: in, The difference in discharge capacity is due to the equal voltage interval. For the first Battery No. 1 The capacity has a voltage of 3.2V per cycle. For the first Battery No. 1 The capacity has a cycle voltage of 2.8V.

[0036] The capacity difference between equal charging voltage intervals gradually decreases with increasing battery charge-discharge cycles. The calculation method is to take the capacity difference when the voltage varies between 3.2V and 3.5V with current I > 0. The calculation expression is: in, The difference in capacity is due to the equal charging voltage interval. For the first Battery No. 1 The capacity has a cycle voltage of 3.5V.

[0037] This method identifies battery degradation states based on incremental capacity analysis (ICA). ICA transforms the flat voltage plateau during constant-current charging or discharging into peaks on the IC curve. Small voltage fluctuations within the plateau range correspond to larger capacities, essentially reflecting the battery's phase transition characteristics during lithium-ion intercalation and deintercalation in the positive and negative electrode materials. A higher peak indicates a greater rate of capacity change within that voltage range, meaning the battery stores or releases more lithium ions, indicating a more concentrated and pronounced reaction within the voltage plateau. The slope of the incremental capacity curve can be used to determine the rate of capacity change within a given voltage range; a steeper slope indicates faster capacity change at that voltage, reflecting higher reactivity of the battery materials within that range. The area enclosed by the curve and the horizontal axis represents the capacity added or released during the phase transition; a smaller area reflects a decrease in total usable capacity. Peak decreases, peak area reductions, and peak position shifts can reflect various changes within the lithium-ion battery, including SEI film growth, electrode material degradation, and phase transition behavior. As batteries age, the height, area, and position of the curve peaks all change. Therefore, the aging mechanism inside the battery can be inferred by analyzing the curve's change pattern.

[0038] The IC (capacity increment) curve is calculated as the gradient of battery charging or discharging capacity relative to voltage. During battery discharge, the voltage drops rapidly in the initial stage, then enters a plateau where the rate of decrease slows down. After the plateau, the voltage drops rapidly to reach the cutoff voltage. Converting the battery discharge data into a dQ / dV-V curve results in a single-peaked curve that gradually decreases on both sides, from which multiple health characteristics can be extracted. Assuming the IC curve and the curve... The ordinates of the two intersection points are , , IC curve and curve The first intersection point ordinate IC curve and curve The second intersection point ordinate The peak height of the IC curve is recorded as The position of the IC curve peak is denoted as IC curve and curve The area enclosed is denoted as Using IC curves and curves The slope of the IC curve peak is calculated from the two intersection points, the height of the IC curve peak, and the position of the IC curve peak. and the slope of the curve peak The calculation expression is: in, For capacity, For capacity increment, The differential change in capacity For voltage, For voltage increment, This is the differential change in voltage. The independent variable is The function, For the first Battery No. 1 Secondary cycle IC curve and curve The area enclosed, For the first Battery No. 1 Secondary cycle IC curve and curve The ordinate of the first intersection point For the first Battery No. 1 Secondary cycle IC curve and curve The ordinate of the second intersection point For the first Battery No. 1 The slope of the peak of the secondary cycle curve. For the first Battery No. 1 The slope of the lower peak of the secondary cycle curve.

[0039] Indirect health factors show a clear positive or negative correlation with the trend of declining health status.

[0040] Outliers exist in battery data due to human interference and equipment malfunction. The Interquartile Range (IQR) method is used to detect outliers, selecting data points below the lower bound or above the upper bound as outliers. After removing outliers, the average of the two points immediately before and after the outlier is used to fill the gap. The calculation expression is as follows: in, Interquartile range, For data points located in the 25th percentile, It is located at 75% of the data points. The lower boundary is... The upper boundary, After removing outliers Replacement data at any time, For the first Battery No. 1 Data that needs to be cleaned frequently For the first Battery No. 1 Data that needs to be cleaned frequently For the first Battery No. 1 Data that needs to be cleaned frequently For the first Battery No. 1 Data that needs to be cleaned at all times.

[0041] S2: Calculate the correlation coefficients of the preprocessed health factors, and select health factors based on the correlation coefficients to obtain strongly correlated health factors; The correlation coefficient between the two health factors was calculated using the Spearman correlation coefficient method. Because the trends of HI (hysteresis loop) differ, it is difficult to directly determine its association with health status. To quantify the correlation between HI and health status, the Spearman correlation coefficient is used to assess the monotonic relationship between the two variables. Assume two sets of HI sequences for a given battery. and There are in each Given several variables, sort their data values ​​in ascending order, and then... The first in the sequence Rank assignment of variables , The first in the sequence Rank assignment of variables For each pair of variables, the Spearman correlation coefficient is calculated as follows: in, for and The correlation coefficient; when hour, and Strong correlation.

[0042] In some specific embodiments of the present invention, the correlation coefficients corresponding to HI are shown in Table 1. Table 1. Correlation coefficients corresponding to HI In this embodiment, the HI (Highly Indicative Health Factor) with an absolute correlation coefficient greater than 0.95 was selected from battery indicators as a strongly correlated health factor. TIEDVD, EVIDCD As a strongly correlated health factor.

[0043] S3: Calculate the variance inflation factor of the strongly correlated health factors and set a variance inflation factor threshold. If the variance inflation factor is greater than the variance inflation factor threshold, then the strongly correlated health factors have multicollinearity. If the variance inflation factor is less than or equal to the variance inflation factor threshold, then the strongly correlated health factors do not have multicollinearity. Multicollinearity was determined in strongly associated health factors by using variance inflation factor and variance inflation factor threshold.

[0044] The variance inflation factor (VIF) is an indicator used to measure the degree of multicollinearity among independent variables. Its calculation formula is: in, for The coefficient of determination for least squares regression on other independent variables. For the sample size, For the first The model prediction value for each observation. For the first The true dependent variable value of each observation. The mean of the dependent variable for all observations.

[0045] In some specific embodiments of the present invention, the variance inflation factor threshold is set to 10. If the variance inflation factor is greater than 10, then the strongly associated health factors have multicollinearity. If the variance inflation factor is less than or equal to 10, then the strongly associated health factors do not have multicollinearity.

[0046] S4: Using the ridge regression algorithm, construct a ridge regression health status estimation model, solve for the optimal solution of the ridge regression health status estimation model, and obtain the optimal ridge regression health status estimation model. Ridge regression is a regularized linear regression method that introduces a penalty term to the regression coefficients. It is suitable for handling linear regression problems with multicollinearity, high-dimensional features, or noisy data, and can improve the stability and generalization ability of the model.

[0047] For a given sample , ,in For the first A multicollinearity health factor For the first The health status of a lithium-ion battery. For the battery charge / discharge data dimension, the ridge regression health status estimation model is as follows: in, The weighting coefficients of the ridge regression health status estimation model. For the bias term of the ridge regression health status estimation model, As a strongly associated health factor, This indicates the health status of the lithium-ion battery.

[0048] Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) are used as performance evaluation metrics. MAE is the average absolute value of the difference between predicted and true values, used to evaluate the overall prediction error of the model. RMSE is the root mean square of the difference between predicted and true values, which imposes a larger penalty on larger errors and is used to evaluate the performance of regression problems. The smaller the values ​​of both metrics, the higher the prediction accuracy.

[0049] In some specific embodiments of the present invention, representative test samples are selected from the dataset for demonstration. The remaining samples, excluding the test samples, are used as the training set. Four lithium-ion battery sample data points extracted within the daily charge / discharge state range of the energy storage power station's batteries are used as input, and the battery health status is used as the output. The health status estimation results (MAE and RMSE) on the test set are 1.21%, 0.73%, 0.60%, 0.69% and 1.60%, 0.99%, 0.76%, 0.84%, respectively. The coefficients of determination for least squares regression were 0.87, 0.94, 0.96, and 0.96, respectively, indicating that the model fits the data well.

[0050] Based on the state of charge (SOC) range of 20% to 80% during the daily operation and maintenance of energy storage power stations, the required charge and discharge data segments are extracted. The health status degradation information contained in the operation and maintenance data such as temperature, voltage, current, and charge / discharge amounts is fully utilized. Using the extracted HI (Health Status Index) data, a mapping relationship between HI and health status is established based on a ridge regression model. The estimation errors (MAE and RMSE) of lithium-ion battery health status are within 1.5% and 2%, respectively. Above 0.85, the model performs well, and the ridge regression method can generalize well to unknown data.

[0051] Solving for the optimal solution of the ridge regression health state estimation model includes: S41: Add the L2 norm of the weight coefficients to the loss function of the ridge regression health status estimation model and multiply it by the regularization term of the regularization coefficient to obtain the regularized loss function. This invention transforms the process of solving for the weight coefficients and bias terms into a least squares problem. To address the issue of large output variations when the input error is small, the ridge regression model adds a weight coefficient to the loss function. L2 normal form multiplied by regularization coefficient The regular expression for the expression is: in, For loss, For the input quantity, for The corresponding health status of lithium-ion batteries. For the first A strongly correlated health factor The weighting coefficients of the ridge regression health status estimation model. For the bias term of the ridge regression health status estimation model, The regularization coefficient is . It is an L2 paradigm. Find the independent variable that minimizes the function; S42: Set the regularization coefficient to 1. When the regularization loss function is minimized, the optimal solution for the weight coefficients and the optimal solution for the bias term of the ridge regression health state estimation model are obtained. The calculation expression is as follows: in, It is the identity matrix. This is the transpose of the matrix. For the first The true dependent variable value of each observation. To find the inverse matrix.

[0052] S5: Input the strongly correlated health factors with multicollinearity into the optimal ridge regression health status estimation model, remove the multicollinearity of the strongly correlated health factors, and obtain the estimated value of the health status of the lithium-ion battery. The optimal ridge regression health status estimation model is used to remove the multicollinearity of strongly correlated health factors, thereby obtaining the estimated value of the health status of lithium-ion batteries.

[0053] S6: The estimated value of the lithium-ion battery health status is predicted by using an improved lithium-ion battery health status prediction model to obtain the predicted value of the lithium-ion battery health status.

[0054] An improved lithium-ion battery health status prediction model was used to predict the estimated value of lithium-ion battery health status, thereby improving the data time series adaptability and the model's feature extraction capability.

[0055] The lithium-ion battery health status prediction model includes a sliding window layer, a temporal convolutional network layer, and two GRU layers.

[0056] The input data sequence is divided based on a sliding window layer. Key time series features of the data sequence processed by the sliding window layer are extracted through a temporal convolutional network layer. The extracted key time series features are input into two GRU layers. The GRU layers use a gating mechanism to effectively capture long-term dependencies in the time series data and obtain the predicted value of the health status of lithium-ion batteries.

[0057] A sliding window strategy is adopted to divide the data sequence of lithium-ion battery health status. The sliding window fits the time series characteristics of battery SOH and improves the model's adaptability to SOH fluctuations and degradation stages.

[0058] The autocorrelation of the health status of the energy storage power station is calculated by using the autocorrelation function, and the historical data step size and prediction step size are determined based on the autocorrelation of the health status of the energy storage power station. The autocorrelation function (ACF) measures the autocorrelation of a variable with its values ​​at different lag times to determine the degree of correlation between an observation at a given time point and previous observations. When calculating the autocorrelation of the health status of an energy storage power station, the larger the observation window, the lower the correlation between a given time point and previous observations.

[0059] In some specific embodiments of the present invention, when the time interval is 32, the autocorrelation can be maintained at 0.9, and when the time interval is 64, the autocorrelation is 0.8. Since a smaller observation window requires more computing resources and processing time, and a larger window cannot capture rapid changes and short-term fluctuations in the data in time, the observation window of each model is set to 64, and the predicted future data length is set to 36. That is, in the operation of the energy storage power station, if the sampling frequency is once every 10 days, the health status changes in the next 360 days can be predicted.

[0060] The first 60% of the total cycle time for each sample is used as the training and validation sets, with the training set comprising 50% and the validation set 10%, leaving the remaining 40% as the test set. The dataset is divided into a fixed-size sliding window, progressively sliding one unit across the data sequence to form the training set. During the prediction phase, the prediction step size is increased by one unit at a time, forming the test set. The data sequence is divided, and a neural network training and prediction process is used. Historical data sequences are input into the GRU model, which outputs future data sequences to predict lithium-ion health status.

[0061] Temporal convolutional networks (TCNs) are an architecture that uses convolutional neural networks (CNNs) to process sequential data. They include causal convolution, dilated convolution, and residual connections. They have powerful temporal feature extraction capabilities, model distant temporal relationships with fewer parameters, thereby achieving efficient parallel processing across all time steps and are less prone to gradient explosion or vanishing.

[0062] The input data is fed into two GRU layers for forward computation, with each layer containing 64 neurons. A ReLU activation function is used for non-linear mapping to obtain the output. A Dropout layer is used to prevent overfitting, and MSE is used as the loss function. An Adam optimizer is employed with a learning rate of 0.005. The prediction starting point for the lithium-ion battery health state is set at 60% of the total cycle time. The health state values ​​for the remaining 40% of the cycle time are used as the true values ​​and compared with the predicted results.

[0063] Three algorithms—GRU, sliding window-GRU (F1), and sliding window-TCN-2GRU (F2)—were compared to establish a health status prediction model. The prediction accuracy was statistically analyzed by calculating the MAE and RMSE error indices. The prediction results for the health status of lithium-ion batteries are as follows: Figure 2 As shown, Figure 2 Figure (a) shows the predicted health status of lithium-ion batteries for the first sample data. Figure 2Figure (b) shows the predicted health status of lithium-ion batteries for the second sample data. Figure 2 Figure (c) shows the predicted health status of lithium-ion batteries for the third sample data. Figure 2 Figure (d) shows the lithium-ion battery health status prediction results for the fourth sample data. "F2" represents the prediction result obtained using the improved lithium-ion battery health status prediction model of this invention, "F1" represents the prediction result obtained using the sliding window-GRU model, "GRU" represents the prediction result obtained using the GRU model, and "true" represents the actual value of the lithium-ion battery health status. Figure 2 It can be seen that the GRU model has the lowest accuracy in predicting the health status, followed by the F1 model, and the F2 model has the highest accuracy. Through experimental comparison, it can be seen that the model of this invention has better model feature extraction capability, higher prediction accuracy, and is more suitable for establishing a health status prediction model for energy storage power stations.

[0064] For example, the prediction error index of lithium-ion battery health status. Figure 3 As shown, Figure 3 Figure (a) shows the prediction error index of lithium-ion battery health status for the first sample data. Figure 3 Figure (b) shows the prediction error index of lithium-ion battery health status for the second sample data. Figure 3 Figure (c) shows the prediction error index of lithium-ion battery health status for the third sample data. Figure 3 Figure (d) shows the prediction error indices for the lithium-ion battery health status of the fourth sample data, namely RMSE and MAE. The smaller the values ​​of these two indices, the higher the model's prediction accuracy. Figure 3 As shown, across the four samples, the highest RMSE and MAE results were 2.68% and 2.07% respectively, predicted using the GRU model, while the lowest were 0.34% and 0.24% respectively, predicted using the F2 model. Both prediction errors for the three models were within 3%, demonstrating their ability to effectively capture the relationships between time series data and track the aging trends of lithium batteries. Furthermore, they exhibited good generalization ability across battery samples using different charge / discharge strategies. The F2 model showed the best error performance among the battery samples, with both its RMSE and MAE prediction errors within 1%.

[0065] like Figure 4 As shown, a multi-model collaborative estimation lithium-ion battery health state prediction system is used to execute a multi-model collaborative estimation lithium-ion battery health state prediction method, including: The extraction module 101 extracts the health factors of lithium-ion batteries and preprocesses the health factors; The first calculation module 102 calculates the correlation coefficient of the preprocessed health factors and selects health factors based on the correlation coefficient to obtain strongly correlated health factors. The second calculation module 103 calculates the variance inflation factor of the strongly correlated health factors and sets a variance inflation factor threshold. If the variance inflation factor is greater than the variance inflation factor threshold, then the strongly correlated health factors have multicollinearity. Module 104 uses the ridge regression algorithm to construct a ridge regression health state estimation model, solves for the optimal solution of the ridge regression health state estimation model, and obtains the optimal ridge regression health state estimation model. The estimation module 105 inputs strongly correlated health factors with multicollinearity into the optimal ridge regression health status estimation model, removes the multicollinearity of the strongly correlated health factors, and obtains the estimated value of the health status of the lithium-ion battery. The prediction module 106 predicts the estimated value of the lithium-ion battery health status using an improved lithium-ion battery health status prediction model, thereby obtaining the predicted value of the lithium-ion battery health status.

[0066] Through the collaborative work of the above modules, the Spearman correlation coefficient is used to screen strongly correlated health factors and eliminate weakly correlated interference terms, solving the problem of traditional health factor selection relying on experience and being highly subjective. Ridge regression is used to solve the multicollinearity problem among health factors. Furthermore, by combining a sliding window strategy, a temporal convolutional network (TCN), and a GRU model to capture the long-term temporal dependence of health status, the full-cycle coverage of "current health status estimation + future decline trend prediction" is achieved, filling the technical gap in "multicollinearity handling + long-cycle prediction accuracy" of traditional data-driven models. This invention significantly improves prediction accuracy and can accurately predict the health status of lithium-ion batteries. By using ridge regression to handle multicollinearity and Spearman coefficients to screen health factors, combined with a sliding window strategy and a temporal convolutional network (TCN), the improved lithium-ion battery health status prediction model has high robustness. It can be deployed in energy storage power stations with different topologies such as centralized and string storage without requiring specific modifications.

[0067] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the health status of lithium-ion batteries using multi-model collaborative estimation, characterized in that, include: S1: Extract the health factors of lithium-ion batteries and pre-treat the health factors; S2: Calculate the correlation coefficients of the preprocessed health factors, and select health factors based on the correlation coefficients to obtain strongly correlated health factors; S3: Calculate the variance inflation factor of the strongly correlated health factors and set a variance inflation factor threshold. If the variance inflation factor is greater than the variance inflation factor threshold, then the strongly correlated health factors have multicollinearity. S4: Using the ridge regression algorithm, construct a ridge regression health status estimation model, solve for the optimal solution of the ridge regression health status estimation model, and obtain the optimal ridge regression health status estimation model. S5: Input the strongly correlated health factors with multicollinearity into the optimal ridge regression health status estimation model, remove the multicollinearity of the strongly correlated health factors, and obtain the estimated value of the health status of the lithium-ion battery. S6: The estimated value of the lithium-ion battery health status is predicted by using an improved lithium-ion battery health status prediction model to obtain the predicted value of the lithium-ion battery health status.

2. The method for predicting the health status of lithium-ion batteries using multi-model collaborative estimation according to claim 1, characterized in that, The health factors include direct and indirect health factors. Direct health factors include battery internal resistance, maximum temperature, average temperature, minimum temperature, charging time, and temperature range. Indirect health factors include the time difference between equal discharge voltage intervals, the time difference between equal charge voltage intervals, the time difference between equal voltage interval discharge capacity, the time difference between equal charge voltage interval capacity, the time to reach peak temperature, the time to reach the lower limit cutoff potential, and the capacity increment curve. The first intersection point ordinate, capacity increment curve and curve The second intersection point ordinate, the peak height and position of the capacity increment curve, and the relationship between the capacity increment curve and the curve The area enclosed, the slope of the curve peak, and the slope of the curve drooping.

3. The method for predicting the health status of lithium-ion batteries using multi-model collaborative estimation according to claim 2, characterized in that, The direct health factors are obtained by measuring the lithium-ion batteries in actual operation of the energy storage power station, while the indirect health factors are extracted from voltage, current, and temperature.

4. The method for predicting the health status of lithium-ion batteries using multi-model collaborative estimation according to claim 1, characterized in that, The preprocessing method uses the interquartile range to detect outliers in health factors. Health factor data points that are below the lower bound or above the upper bound are selected as outliers. After removing outliers, the average of the two points before and after the outlier is used to fill the gap.

5. The method for predicting the health status of lithium-ion batteries using multi-model collaborative estimation according to claim 1, characterized in that, In step S2, the correlation coefficient between the two health factors is calculated using the Spearman correlation coefficient method.

6. The method for predicting the health status of lithium-ion batteries using multi-model collaborative estimation according to claim 1, characterized in that, The expression for calculating the variance inflation factor is: in, for The coefficient of determination for least squares regression on other independent variables. For the sample size, For the first The model prediction value for each observation. For the first The true dependent variable value of each observation. The dependent variable is the mean of all observations. For the first The variance inflation factor for each observation.

7. The method for predicting the health status of lithium-ion batteries using multi-model collaborative estimation according to claim 1, characterized in that, The ridge regression health status estimation model is as follows: in, The weighting coefficients of the ridge regression health status estimation model, For the ridge regression health status estimation model, As a strongly associated health factor, This indicates the health status of the lithium-ion battery.

8. The method for predicting the health status of lithium-ion batteries using multi-model collaborative estimation according to claim 1, characterized in that, In step S4, finding the optimal solution for the ridge regression health state estimation model includes: S41: Add the L2 norm of the weight coefficients to the loss function of the ridge regression health status estimation model and multiply it by the regularization term of the regularization coefficient to obtain the regularized loss function. S42: Set the regularization coefficient to 1. When the regularization loss function is minimized, the optimal solution of the weight coefficients and the optimal solution of the bias term of the ridge regression health state estimation model are obtained.

9. The method for predicting the health status of lithium-ion batteries using multi-model collaborative estimation according to claim 1, characterized in that, The improved lithium-ion battery health status prediction model includes a sliding window layer, a temporal convolutional network layer, and two GRU layers. The temporal convolutional network layer includes causal convolution, dilated convolution, and residual connections connected in sequence. The input data sequence is divided based on a sliding window layer. Key time series features of the data sequence processed by the sliding window layer are extracted through a temporal convolutional network layer. The extracted key time series features are input into two GRU layers. The GRU layers use a gating mechanism to effectively capture long-term dependencies in the time series data and obtain the predicted value of the health status of lithium-ion batteries.

10. A multi-model collaborative estimation system for predicting the health status of lithium-ion batteries, characterized in that, A method for predicting the health status of lithium-ion batteries using multi-model collaborative estimation as described in any one of claims 1 to 9, comprising: An extraction module extracts health factors from lithium-ion batteries and preprocesses these health factors. The first calculation module calculates the correlation coefficient of the preprocessed health factors and selects health factors based on the correlation coefficient to obtain strongly correlated health factors. The second calculation module calculates the variance inflation factor of the strongly correlated health factors and sets a variance inflation factor threshold. If the variance inflation factor is greater than the variance inflation factor threshold, then the strongly correlated health factors have multicollinearity. The module constructs a ridge regression health status estimation model using the ridge regression algorithm, solves for the optimal solution of the ridge regression health status estimation model, and obtains the optimal ridge regression health status estimation model. The estimation module inputs strongly correlated health factors with multicollinearity into the optimal ridge regression health status estimation model, removes the multicollinearity of the strongly correlated health factors, and obtains the estimated value of the lithium-ion battery health status. The prediction module uses an improved lithium-ion battery health status prediction model to predict the estimated value of the lithium-ion battery health status, thereby obtaining the predicted value of the lithium-ion battery health status.

Images