An indirect prediction method for the remaining service life of a lithium-ion battery

Abstract

An indirect prediction method for the remaining service life of a lithium-ion battery comprises the following steps: First, extract appropriate indirect health factors that represent the aging of the lithium battery and verify the correlation with the capacity; then, construct an indirect health factor prediction model based on the support vector regression algorithm; next, construct a relationship model between the indirect health factor and the capacity based on the Gaussian process regression; finally, input the predicted value of the indirect health factor into the Gaussian process regression model to achieve the prediction of the lithium battery capacity. When the battery capacity reaches the set failure threshold, output the prediction result of the remaining service life. This method solves the problems that traditional support vector regression can only obtain single-point predicted values and the long-term prediction performance of Gaussian process regression is poor. The prediction performance is stable. While obtaining accurate prediction results, the prediction confidence interval is obtained, which is convenient for the lithium battery system to make decision-making for maintenance and management in advance when the battery fails.

Description

Technical Field

[0001] The present invention relates to the field of monitoring the electrical performance of lithium batteries, and particularly to an indirect prediction method for the remaining service life of lithium-ion batteries. Technical Background

[0002] Compared with lead-acid batteries, nickel-metal hydride batteries, nickel-cadmium batteries, etc., lithium-ion batteries have the advantages of small size, light weight, greatly reducing the volume and weight of the vehicle battery pack, high working voltage, high energy density, no memory effect, small self-discharge, and long cycle life. They are widely used in electronic devices, energy storage systems, aerospace and other fields, becoming the mainstream energy source for electric vehicles and playing an important role in modern society. However, during repeated use, the performance of lithium batteries will gradually degrade and fail, resulting in the battery being prone to leakage and short circuit, affecting the normal operation of the equipment system, and even causing economic losses and explosive disasters. Therefore, obtaining accurate and stable prediction results of the RUL of lithium batteries is beneficial to making a correct judgment on the aging degree of lithium batteries, thereby formulating an optimal secondary utilization plan and equipment system maintenance strategy to ensure the long-term safe and reliable operation of the battery system. It is of great significance for the maintenance, management and operation of the battery system to prevent catastrophic accidents caused by battery aging.

[0003] In recent years, the prediction technology for the remaining service life of lithium batteries has received extensive attention from researchers. Among them, data-driven methods have received more and more extensive attention. By using the collected battery voltage, current, temperature and other data, through various data processing and analysis methods, the relationship between the hidden information and the future degradation trend is found to predict the remaining service life of the battery. Compared with model-based methods, data-driven methods have received more and more extensive attention. They do not need to accurately simulate the complex aging mechanism and evolution rules inside the lithium battery, are no longer limited to a specific lithium battery degradation model, and have good universality.

[0004] Currently, in data-driven methods, one type of method is to predict the remaining useful life of lithium batteries based on lithium battery capacity and internal resistance data, and signal processing methods such as wavelet analysis and empirical mode decomposition are used to reduce the influence of capacity regeneration and local fluctuations. Although the above methods can obtain satisfactory prediction results, due to the difficulty of online measurement of lithium battery capacity and internal resistance in practical applications, methods based on battery capacity or internal resistance cannot be used for online RUL prediction. Another type of method is to extract indirectly measurable health factors from the curves of voltage, current, and temperature to characterize the aging process of lithium-ion batteries and achieve online prediction of the RUL of lithium batteries. In the existing literature methods of this type, the support vector regression method is often used for the RUL prediction of lithium batteries. SVR improves the generalization ability by seeking the minimum structural risk and can well model even when the sample size is small. However, this method cannot obtain the uncertainty expression of the prediction results. The GPR model is a flexible non-parametric model based on Bayesian theory. While obtaining the point estimate of the predicted value, it gives the prediction confidence interval to represent the prediction uncertainty. However, when using the GPR method for the RUL prediction of lithium batteries, the long-term prediction performance is poor. Moreover, these methods do not consider the entire charge-discharge process of lithium batteries and the influence of temperature on the RUL prediction of lithium batteries. Summary of the Invention

[0005] In view of the problems pointed out in the above background technology, in order to obtain accurate and stable prediction information on the remaining useful life of lithium batteries, the present invention provides an indirect prediction method for the remaining useful life of lithium-ion batteries that improves the defects of SVR and GPR. The technical solutions adopted by the present invention include the following steps:

[0006] Step 1, data collection: Use sensor devices to monitor the charging and discharging operation process of lithium batteries, and extract the voltage, current, and temperature data information during the entire charging and discharging process of lithium batteries.

[0007] Step 2, feature extraction: Respectively extract the time intervals of equal charging voltage differences, the time intervals of equal charging current differences, the time intervals of equal discharging voltage differences, and the average temperature of equal charging current differences from the constant current charging stage, constant voltage charging stage, constant current discharging stage, and temperature signal as four indirect health factors characterizing the degradation of lithium batteries, and verify the correlation with the lithium battery capacity.

[0008] Step 3, construct an indirect health factor prediction model, and use the indirect health factors of lithium batteries and the number of charge-discharge cycles to train a support vector regression model to achieve the prediction of the indirect health factors of lithium batteries.

[0009] Step 4: Construct a lithium battery capacity prediction model. Use the predicted value of the indirect health factor as the input of the lithium battery capacity prediction model to achieve lithium battery capacity prediction. When the battery capacity reaches the set failure threshold, obtain the predicted result and confidence interval of the remaining service life of the lithium battery, and conduct prediction performance analysis.

[0010] As an optimization: The specific method of Step 1 is to extract the constant current charging voltage signal, constant voltage charging current signal, constant current discharging voltage signal, and temperature signal.

[0011] As an optimization: The correlation between the indirect health factor and the lithium battery capacity in Step 2 is obtained by using the grey parallel analysis method to obtain the correlation grade between the indirect health factor and the lithium battery capacity. The specific steps are as follows:

[0012] Step 2.1: Designate the comparison sequence X m ={x m (i)|i = 1, 2,..., n}, where X m represents the indirect health factor vector. Designate the reference sequence X 0 ={x 0 (i)|i = 1, 2,..., n}, where X 0 represents the lithium battery capacity;

[0013] Step 2.2: Perform data normalization processing;

[0014] Step 2.3: Calculate the correlation coefficient between the comparison sequence X m and the reference sequence X 0 ;

[0015] Step 2.4: Calculate the correlation degree between the comparison sequence and the reference sequence.

[0016] As an optimization: The specific method of Step 3 is to set the starting prediction point T and use the training set to train the SVR model, where i represents the i-th charge and discharge cycle, and HIs i represents the indirect health factor corresponding to the i-th charge and discharge cycle. Construct a lithium battery indirect health factor prediction model based on the support vector regression algorithm, and obtain the predicted value HIs_pre of the indirect health factor after the starting prediction point.

[0017] As an optimization: Step 4 specifically includes:

[0018] Step 4.1: Set the starting prediction point T and the battery capacity failure threshold Cap EOL ;

[0019] Step 4.2: Use the training set to train the GPFR model, where HIs idenote the i-th feature vector, Cap i denote the i-th battery capacity value, and construct a lithium battery capacity prediction model based on Gaussian process regression;

[0020] Step 4.3: Use the predicted value HIs_pre of the indirect health factor in Step 3 as the test input of the battery capacity prediction model to obtain the predicted value Cap_pre of the lithium battery capacity after the starting prediction point;

[0021] Step 4.4: When the predicted value Cap_pre of the lithium battery capacity in Step 4.3 reaches the battery capacity failure threshold Cap set in Step 4.1 EOL output the prediction result and confidence interval of the remaining useful life of the lithium battery, and conduct analysis of prediction performance indicators.

[0022] The beneficial effects of the present invention are as follows:

[0023] 1. By integrating the SVR method and the GPFR method, the present invention effectively solves the problem that the uncertainty expression of the prediction result cannot be obtained by using the support vector regression alone, and at the same time improves the problems of low long-term prediction accuracy and unstable prediction when using the Gaussian process regression alone.

[0024] 2. The present invention considers the entire charging and discharging process of the lithium battery and the influence of temperature, extracts appropriate indirect health factors characterizing the performance aging of the lithium battery from the constant current charging stage, constant voltage charging stage, constant current discharging stage and temperature data, and the model has a wider application range.

[0025] 3. It effectively improves the stability of the lithium battery RUL prediction performance and the accuracy of the prediction result, and at the same time gives the prediction confidence interval, which is conducive to the lithium battery management system to make decision-making maintenance and management before the battery fails, and makes a more reasonable judgment on the aging degree of the battery. Description of the Drawings

[0026] Figure 1 is a schematic diagram of the overall process of the method of the present invention.

[0027] Figure 2 is a graph of the voltage, current and temperature of the B6 lithium ion battery

[0028] Figure 3 is a graph of the extracted indirect health factor changing with the number of charge and discharge cycles

[0029] Figure 4 is a graph of the prediction result of the indirect health factor of the B6 lithium ion battery

[0030] Figure 5 is a graph of the battery capacity prediction results of different algorithms for the B6 lithium battery when the starting prediction period is 70 Detailed Embodiments

[0031] A method for predicting the remaining service life of a lithium-ion battery is as follows Figure 1 shown: In the data preprocessing stage, a lithium battery dataset is obtained through repeated charge and discharge experiments of the lithium battery, and then signal extraction is performed to obtain voltage, current, and temperature curves, from which the HIs extraction required for HIs analysis is performed, that is, indirect health factors are extracted and correlation coefficient analysis is carried out; in the SVR prediction model stage, an SVR prediction model is constructed based on the extracted HIs to obtain the HIs prediction value; in the capacity prediction model stage, the dataset is divided into a training set and a test set by setting the starting prediction point, the model parameters are initialized, and the GPFR model is trained based on the training set. The hyperparameters of the model are optimized to obtain the GPFR prediction model of the lithium battery capacity. The HIs prediction result is input into the GPFR prediction model to obtain the capacity prediction result of the battery; in the result analysis stage, when the battery capacity reaches the failure threshold, the RUL prediction result and the prediction interval are output, and the prediction result is analyzed.

[0032] Specifically, it includes the following steps

[0033] Step 1: Use a sensor device to monitor the charging and discharging operation process of the lithium battery, and extract the voltage, current, and temperature data information during the entire charging and discharging process of the lithium battery.

[0034] Step 2: Feature extraction: Considering the entire charging and discharging process of the lithium battery and the influence of temperature, extract appropriate indirect health factors that represent the aging of the lithium battery, and verify the correlation with the lithium battery capacity.

[0035] Step 2.1: Considering the influence of the entire charging and discharging process of the lithium battery and temperature, extract the time interval of equal charging voltage difference, the time interval of equal charging current difference, the time interval of equal discharging voltage difference, and the average temperature of equal charging current difference from the constant current charging stage, constant voltage charging stage, constant current discharging stage, and temperature signal respectively as four indirect health factors representing the degradation of the lithium battery.

[0036] Step 2.2: Adopt the grey parallel analysis method to verify the association degree between the indirect health factors extracted in Step 2.1 and the lithium battery capacity. The specific steps are as follows

[0037] Step (1), specify the comparison sequence X m ={x m (i)|i = 1, 2,..., n}, where X m represents the indirect health factor vector, and specify the reference sequence X 0 ={x 0 (i)|i = 1, 2,..., n}, where X 0 represents the lithium battery capacity.

[0038] Step (2), perform data normalization processing

[0039] Step (3), calculate the comparison sequence X m and the reference sequence X 0 and calculate the correlation coefficient between them;

[0040] Step (4), calculate the association degree between the comparison sequence and the reference sequence.

[0041] Step 3: Obtain the indirect health factor dataset representing the performance aging of the lithium battery through Step 2.2. Construct a prediction model for the indirect health factor of the lithium battery based on the support vector regression algorithm to achieve the prediction of the indirect health factor of the lithium battery. The specific steps are as follows:

[0042] Step 3.1: Set the starting prediction point T, divide the dataset into a training set and a test set, and use the training set to train the SVR model, where i represents the i-th charge-discharge cycle, and HIs i represents the indirect health factor corresponding to the i-th charge-discharge cycle, and construct a prediction model for the indirect health factor of the lithium battery based on the support vector regression algorithm;

[0043] Step 3.2: Through the trained prediction model for the indirect health factor of the lithium battery in Step 3.1, obtain the predicted value HIs_pre of the indirect health factor after the starting prediction point;

[0044] Step 4: Construct a lithium battery capacity prediction model, use the predicted value of the indirect health factor as the input of the lithium battery capacity prediction model to achieve the prediction of the lithium battery capacity. The specific steps are as follows:

[0045] Step 4.1: Set the starting prediction point T, divide the dataset into a training set and a test set, and set the battery capacity failure threshold Cap EOL ;

[0046] Step 4.2: Use the training set to train the GPFR model, where HIs i represents the i-th feature vector, and Cap i represents the i-th battery capacity value, and construct a lithium battery capacity prediction model based on Gaussian process regression. The specific steps are as follows:

[0047] Step (1), select the linear mean function as the mean function of Gaussian process regression, and select the squared exponential covariance function as the covariance function of Gaussian process regression to improve the long-term prediction performance of traditional GPR,

[0048] Step (2), initialize the hyperparameters of the GPFR model;

[0049] Step (3), use the training set Train the GPFR model, optimize the model parameters using the conjugate gradient algorithm, and establish a GPFR lithium battery capacity prediction model;

[0050] Step 4.3: Use the predicted value of the indirect health factor HIs_pre in Step 3 as the test input of the battery capacity prediction model to obtain the predicted value of the lithium battery capacity Cap_pre after the starting prediction point;

[0051] Step 4.4: When the predicted value of the lithium battery capacity Cap_pre in Step 4.3 reaches the battery capacity failure threshold Cap EOL set in Step 4.1, output the predicted result and confidence interval of the remaining service life of the lithium battery, and conduct an analysis of the prediction performance index.

[0052] The following combines the drawings and embodiments to describe the specific implementation manners of the present invention in detail. Select three batteries B5, B6, and B7 as the data sets in the specific embodiments.

[0053] The indirect prediction method for the remaining service life of lithium-ion batteries based on SVR-GPFR is as Figure 1 shown. The specific steps of this data-driven fusion algorithm are as follows:

[0054] Step 1: Use sensor devices to monitor the charging and discharging operation process of the lithium battery, and extract the voltage, current, and temperature data information during the entire charging and discharging process of the lithium battery;

[0055] Step 2: Feature extraction: Considering the entire charging and discharging process of the lithium battery and the influence of temperature, extract appropriate indirect health factors representing the aging of the lithium battery, and verify the correlation with the lithium battery capacity;

[0056] Step 2.1: Considering the influence of the entire charging and discharging process of the lithium battery and temperature, respectively extract four indirect health factors representing the degradation of the lithium battery, namely the time interval of the charging voltage difference such as HI1, the time interval of the charging current difference such as HI2, the time interval of the discharging voltage difference such as HI3, and the average temperature of the charging current difference, from the constant current charging stage, constant voltage charging stage, constant current discharging stage, and temperature signal;

[0057] Step 2.2: Adopt the grey parallel analysis method to verify the association degree between the indirect health factors extracted in Step 2.1 and the lithium battery capacity. Taking the B6 battery as an example, calculate the correlation degree between each HIs and the lithium battery capacity respectively. The relationship levels between each HIs and the battery capacity are shown in Table 1. The specific steps are as follows:

[0058] Step (1): Designate the comparison sequence X m ={x m (i)|i = 1, 2,..., n}, where X mDenote the indirect health factor vector and specify the reference sequence X 0 ={x 0 (i)|i = 1, 2, …, n}, where X 0 represents the lithium battery capacity;

[0059] Step (2), data normalization processing;

[0060] Step (3), calculate the correlation coefficient between the comparison sequence X m and the reference sequence X 0 The specific formula is shown in Equation 1:

[0061]

[0062] Step (4), calculate the correlation degree between the comparison sequence and the reference sequence. The specific formula is shown in Equation 2:

[0063]

[0064] Table 1 Correlation degree analysis between HIs and battery capacity

[0065] Health factor HI1 HI2 HI3 HI4 Degree of association 0.7161 0.6869 0.8853 0.7776

[0066] Step 3. Obtain the dataset of indirect health factors representing the performance aging of lithium batteries through Step 2.2. Construct a prediction model for the indirect health factors of lithium batteries based on the support vector regression algorithm to achieve the prediction of the indirect health factors of lithium batteries. The specific steps are as follows:

[0067] Step 3.1. Set the starting prediction point T and divide the dataset into a training set and a test set.

[0068] Step 3.2. Initialize the penalty coefficient C and kernel width σ of the support vector regression algorithm. In the present invention, the radial basis kernel function (RBF) is selected as the kernel function of the support vector regression algorithm. The mathematical expression of the kernel function is shown in Equation 3:

[0069]

[0070] Step 3.3. Use the training set to train the SVR model, where i represents the i-th charge-discharge cycle, and HIs i represents the indirect health factor corresponding to the i-th charge-discharge cycle, and construct a prediction model for the indirect health factors of lithium batteries based on the support vector regression algorithm;

[0071] Step 3.4. Through the prediction model for the indirect health factors of lithium batteries trained in Step 3.3, obtain the predicted value HIs_pre of the indirect health factor after the starting prediction point;

[0072] Step 4: Build a lithium battery capacity prediction model, use the predicted value of the indirect health factor as the input of the lithium battery capacity prediction model to achieve lithium battery capacity prediction. The specific steps are as follows:

[0073] Step 4.1: Set the starting prediction point T, divide the data set into a training set and a test set, and set the battery capacity failure threshold Cap EOL ;

[0074] Step 4.2: Use the training set to train the GPFR model, where HIs i represents the i-th feature vector, and Cap i represents the i-th battery capacity value. Build a lithium battery capacity prediction model based on Gaussian process regression. The specific steps are as follows:

[0075] Step (1), select the linear mean function as the mean function of Gaussian process regression, and select the squared exponential covariance function as the covariance function of Gaussian process regression to improve the long-term prediction performance of traditional GPR. It is defined as the GPFR model in the method of the present invention. The specific function form is as follows:

[0076] m(x) = a * x + b Equation 4

[0077]

[0078] Step (2), initialize the hyperparameters q = [a, b, s f , l];

[0079] Step (3), use the training set to train the GPFR model, and use the conjugate gradient algorithm to optimize the model parameters to establish a GPFR lithium battery capacity prediction model;

[0080] Step 4.3: Use the predicted value of the indirect health factor HIs_pre in Step 3 as the test input of the battery capacity prediction model to obtain the predicted value of the lithium battery capacity Cap_pre after the starting prediction point;

[0081] Step 4.4: When the predicted value of the lithium battery capacity Cap_pre in Step 4.3 reaches the battery capacity failure threshold Cap set in Step 4.1 EOL , output the predicted result and confidence interval of the remaining service life of the lithium battery, and conduct prediction performance index analysis.

[0082] To evaluate the indirect prediction model of the lithium battery RUL of the present invention, the root mean square error RMSE, the mean absolute percentage error MAPE, the R 2 coefficient, the RUL absolute error E RUL , and the RUL relative error RA are used to evaluate the performance of the prediction method. Among them, Qk Represents the actual value of the battery capacity Represents the predicted value of the battery capacity Represents the average value of the actual battery capacity, where n is the total number of capacity predicted values. If RMSE and MAPE are closer to 0, and the R 2 coefficient is closer to 1, the better the prediction performance of the proposed method. RUL true is the actual value of RUL, and RUL predicted is the predicted value of RUL. If E RUL is closer to 0 and RA is closer to 1, the more accurate the RUL prediction result of the proposed method.

[0083]

[0084] E RUL = |RUL true - RUL predicted | Equation 9

[0085]

[0086] Next, combined with specific examples, three groups of lithium - battery RUL prediction experiments are carried out using lithium - battery datasets B5, B6, and B7 batteries. Each group of experiments sets three different starting prediction points (T = 70, T = 80, T = 90) to verify the stability and accuracy of the SVR - GPFR lithium - battery RUL indirect prediction model. The rated capacity of the three lithium - batteries is 2 Ah, and the battery failure threshold is set to 1.4 Ah. Since the capacity degradation curve of the B7 battery cannot reach this threshold, the failure threshold of the B7 battery is set to 1.5 Ah.

[0087] Taking the B6 battery as an example, analyze the change trend of different indirect health factors with battery aging. The curves of battery voltage, current, and temperature under different charge - discharge cycle numbers are as Figure 2 shown Figure 2 (a) is the change curve of the charging voltage during the constant - current charging stage under different cycle numbers. It can be seen that the charging time during the constant - current charging stage decreases with the increase of the charge - discharge cycle number. This is due to the deepening of battery polarization, which causes the battery to age gradually. Therefore, the charging time from 3.5 V to 4.2 V is used as HI1 to represent the health state of the battery, and the extracted time intervals with equal charging - voltage differences are as Figure 3 (a) shown; Figure 2(b) shows the charging current curves during constant-voltage charging under different charge-discharge cycles. It can be clearly seen that as the degree of battery degradation increases, the rate of change of the current gradually slows down, resulting in an increasing trend in the charging time at this stage. This indicates that the lithium intercalation ability weakens and the battery ages more severely. Therefore, the time interval from the start of the CV charging stage to when the current decreases to 100 mA is selected as an indirect health indicator to characterize battery aging, as shown in Figure 3 (b); Figure 2 (c) shows the discharge voltage change curves of lithium-ion batteries at different charge-discharge cycle numbers. As the battery is repeatedly charged and discharged, the performance of the lithium battery gradually degrades, and the duration of use after the battery is fully charged becomes shorter and shorter. As shown in Figure 3 (c), the time interval of equal discharge voltage drop is consistent with the above description; Figure 2 (d) plots the temperature change curves during the charging stage of the battery at different charge-discharge cycle numbers. The temperature drop during the constant-voltage charging stage is the result of the combined effects of battery heat dissipation, irreversible heat release, and electrochemical reaction heat absorption, which can better reflect the aging characteristics of the battery. Therefore, in the present invention, the average temperature from the start of the constant-voltage charging stage until the current reaches 100 mA is used as an indirect health factor to describe battery aging, and the average temperature of the equal charging current difference extracted is as shown in Figure 3 (d). Figure 4 Figure shows the prediction results of four indirect health factors when T = 70. The prediction trend of the indirect health factors reflects the future aging trend of the lithium battery. It can be seen that the overall change trend of the indirect health factors can be well captured by the SVR prediction method, and the predicted value of HIs can characterize the future aging characteristics of the lithium battery.

[0088] The experimental prediction effects of the method SVR-GPFR of the present invention are compared with the SVR-GPR and SVR-RVM fusion methods. From Figure 5 the lithium battery capacity prediction results, it can be seen that compared with other methods, the battery capacity prediction results of the proposed method are closer to the actual battery capacity curve, and the degradation trend of the battery capacity can be well captured. Tables 2 to 4 respectively give the RUL prediction performance analysis of the three batteries under different initial prediction cycles. The RMSE of the three groups of batteries of the method of the present invention is lower than 0.04, and the MAPE is lower than 0.026, and is closer to 0. R 2The coefficients are all higher than 0.85 and closer to 1. The RUL prediction errors of the proposed method at different starting prediction periods are all lower than 6, the relative error RA is all higher than 0.84 and closer to 1, and the prediction confidence interval intervals of batteries B5, B6, and B7 at different starting prediction points are less than 9, 11, and 20 respectively, with a narrower confidence interval. Compared with other methods, the proposed SVR-GPFR method has a more accurate RUL prediction result, is less affected by different starting prediction points, has stable prediction performance, and at the same time gives an expression of the uncertainty of the prediction result, which is convenient for the lithium battery management system to make decision-making maintenance and management before the battery fails.

[0089] Table 2 Analysis of RUL prediction performance of B5 lithium battery

[0090]

[0091]

[0092] Table 3 Analysis of RUL prediction performance of B6 lithium battery

[0093]

[0094] Table 4 Analysis of RUL prediction performance of B7 lithium battery

[0095]

Claims

1. An indirect prediction method for the remaining service life of a lithium-ion battery, characterized in that, the method specifically includes the following steps: Step 1, data collection: Use sensor devices to monitor the charging and discharging operation process of the lithium battery, and extract the voltage, current, and temperature data information during the entire charging and discharging process of the lithium battery; specifically, extract the constant-current charging voltage signal, constant-voltage charging current signal, constant-current discharging voltage signal, and temperature signal; Step 2, feature extraction: Respectively extract the time interval of equal charging voltage difference, the time interval of equal charging current difference, the time interval of equal discharging voltage difference, and the average temperature of equal charging current difference from the constant-current charging stage, constant-voltage charging stage, constant-current discharging stage, and temperature signal, which are four indirect health factors characterizing the degradation of the lithium battery, and verify the correlation with the lithium battery capacity; The correlation between the indirect health factor and the lithium battery capacity in Step 2 is obtained by using the grey parallel analysis method to obtain the association degree between the indirect health factor and the lithium battery capacity. The specific steps are as follows: Step 2.1, specify the comparison sequence X m ={x m (i) i = 1, 2, ..., n}, where X m represents the indirect health factor vector, specify the reference sequence X 0 ={x 0 (i) i = 1, 2, ..., n}, where X 0 represents the lithium battery capacity; Step 2.2, data normalization processing; Step 2.3: Calculate the correlation coefficient between the comparison sequence X m and the reference sequence X 0 ; Step 2.4, calculate the association degree between the comparison sequence and the reference sequence; Step 3: Construct an indirect health factor prediction model. Use the indirect health factor of the lithium battery and the number of charge-discharge cycles to train a support vector regression model to achieve the prediction of the indirect health factor of the lithium battery. The specific method of the said Step 3 is to set the starting prediction point T and use the training set to train the SVR model, where i represents the i-th charge-discharge cycle, and HIs i represents the indirect health factor corresponding to the i-th charge-discharge cycle, construct an indirect health factor prediction model of the lithium battery based on the support vector regression algorithm, and obtain the predicted value HIs_pre of the indirect health factor after the starting prediction point; Step 4, construct a lithium battery capacity prediction model, use the predicted value of the indirect health factor as the input of the lithium battery capacity prediction model to realize the prediction of the lithium battery capacity. When the battery capacity reaches the set failure threshold, obtain the prediction result and confidence interval of the remaining service life of the lithium battery, and conduct prediction performance analysis. Step 4 specifically includes: Step 4.1: Set the starting prediction point T and the battery capacity failure threshold Cap EOL ; Step 4.2: Use the training set to train the GPFR model, where HIs i represents the i-th feature vector, and Cap i represents the i-th battery capacity value, and construct a lithium battery capacity prediction model based on Gaussian process regression; Step 4.3, use the predicted value HIs_pre of the indirect health factor in Step 3 as the test input of the battery capacity prediction model to obtain the predicted value Cap_pre of the lithium battery capacity after the starting prediction point; Step 4.

4. When the predicted value of the lithium battery capacity Cap_pre in Step 4.3 reaches the battery capacity failure threshold Cap set in Step 4.1 EOL output the prediction result of the remaining service life of the lithium battery and the confidence interval, and conduct an analysis of the prediction performance index.

Images