Lithium battery health state prediction method and system based on fusion health indexes

Abstract

The invention provides a lithium battery health state prediction method and system based on fusion health indicators, and the method comprises the steps: obtaining lithium battery charging and discharging parameters and a data set, extracting indirect health indicators for reflecting the degradation characteristics of a battery, and verifying the reasonability of the indirect health indicators through a correlation analysis method; performing dimension reduction processing on the indirect health index by using a KPCA method to obtain first principal component data, constructing a lithium battery health state comprehensive estimation model, and further obtaining a battery health state; and decomposing an original battery capacity sequence into a local fluctuation sequence and a global degradation sequence, performing prediction by using a radial basis function neural network and Gaussian process regression, and performing superposition to obtain a battery residual life prediction value. Aiming at the problems of low accuracy, unstable model and the like in the prior art, the battery health state is measured and calculated by adopting a mode based on the fusion health indexes, the method has the advantages of high estimation precision, good application prospect and the like, and the prediction accuracy is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium batteries, and in particular to a method and system for predicting the health status of a lithium battery based on fusion health indicators. Background Art

[0002] With the development of science and technology, the automobile industry has ushered in a booming development, and people's travel has become more convenient. Traditional fuel vehicles consume a large amount of non-renewable energy, and their market share has been impacted by electric vehicles in the past two years. Pure electric vehicles have no tail gas emissions and are close to zero pollution, so they can effectively improve environmental pollution problems.

[0003] Power batteries are a key area of ​​research in electric vehicles. Based on the differences in electrode materials, the mainstream power batteries at this stage are mainly divided into lead-acid batteries, nickel-metal hydride batteries, nickel-cadmium batteries and lithium batteries. Compared with the other three power batteries, lithium batteries have the advantages of high rated voltage, high energy density, long life and no memory effect, making them the best choice for vehicle power batteries.

[0004] However, the safety issues in the application of lithium batteries cannot be ignored. As the number of charge and discharge cycles increases, the performance of lithium batteries gradually declines, often manifested as a decrease in maximum available capacity and an increase in internal resistance. This is due to the presence of various degradation modes inside lithium batteries, such as the absence of active materials (LAM) and lithium ion inventory deficiency (LLI). When the maximum available capacity of the battery decreases or the internal resistance increases to a certain threshold, the battery must be replaced in a timely manner, otherwise it will cause safety issues such as abnormal equipment operation and battery thermal runaway.

[0005] Currently, there are three main methods for state estimation of lithium-ion batteries:

[0006] The first is the direct measurement method based on experiments. It is a relatively direct method that uses experimental equipment to measure the capacity and internal resistance of the battery, and directly calculates the battery's state of health (SOH) through definition. However, most experiments need to be conducted under laboratory conditions, which has high requirements for the test environment and cannot achieve online estimation. In addition, the direct measurement method has a complex experimental process and a long experimental cycle. It is easily interfered by external factors in real life and is difficult to apply in practice.

[0007] The second is the model-based method. By establishing an appropriate model to simulate the charging and discharging behavior of the battery, it mainly includes the aging mechanism model, the equivalent circuit model and the empirical model. Although a relatively accurate SOH estimation result is obtained, it may be affected by the working conditions during use. At the same time, the model-based method needs to establish a suitable model when using it, and the estimation result depends on the accuracy of the model, which is not universally applicable.

[0008] The third is based on data-driven methods. Extracting health indicators that are highly correlated with battery aging is a key factor in the data-driven method to accurately estimate SOH, but the data-driven method is highly dependent on high-quality, representative training data, requiring sufficient data to cover various working conditions. In practical applications, it may be difficult to collect comprehensive and timely data, which may limit the generalization ability of the model. When the battery operating state exceeds the range of the training data, the accuracy of the prediction may drop significantly.

[0009] In summary, although the current existing technologies have promoted important progress in the prediction of lithium battery health status, they still face problems such as insufficient application prospects, poor estimation accuracy, and high data quality requirements in practical applications that need to be urgently addressed. Summary of the invention

[0010] In view of this, the present invention proposes a lithium battery health status prediction method and system based on fusion health indicators. In view of the problems of low accuracy and model instability of the existing estimation method, the battery health status is calculated based on fusion health indicators. It has the advantages of high estimation accuracy and good application prospects, and effectively improves the prediction accuracy.

[0011] An embodiment of the present invention provides a method for predicting the health status of a lithium battery based on a fusion health index, comprising:

[0012] Obtaining lithium battery charge and discharge parameters and data sets, and extracting indirect health indicators used to reflect battery degradation characteristics, and using correlation analysis to verify the rationality of the indirect health indicators;

[0013] The indirect health indicators are processed by KPCA method to obtain the first principal component data, and a comprehensive estimation model of the health status of the lithium battery is constructed to obtain the health status of the battery;

[0014] The CEEMDAN algorithm is used to decompose the original battery capacity sequence into a local fluctuation sequence and a global degradation sequence. The radial basis function neural network and Gaussian process regression are used to predict them respectively, and the prediction results are superimposed to obtain the predicted value of the remaining battery life.

[0015] Exemplarily, the charge and discharge parameters include charge current, charge cut-off voltage, charge cut-off current, discharge current, discharge cut-off voltage, temperature and nominal capacity, and the data set includes complete charge and discharge cycle data after multiple charge and discharge cycles with abnormal points removed.

[0016] Exemplarily, the indirect health indicators include constant voltage rise charging time VHF1, constant voltage drop discharge time VHF2, constant current charging time proportion CHF1, constant current charging current curve coverage area CHF2, IC curve peak ICHF1 and IC curve peak corresponding voltage ICHF2. The extraction of indirect health indicators for reflecting battery degradation characteristics and the use of correlation analysis to verify the rationality of the indirect health indicators include:

[0017] The VHF1, VHF2, CHF1, CHF2, ICHF1 and ICHF2 are calculated using the charge and discharge parameter data set, wherein the VHF1 is calculated by the moments corresponding to the charging voltages of 4.1V and 3.9V, respectively; the VHF2 is calculated by the moments corresponding to the charging voltages of 3.6V and 3.9V, respectively; the CHF1 is calculated by the ratio of the time consumed by constant current charging to the total charging time; the CHF2 is calculated by the integration of the constant current charging current on the time axis; the ICHF1 and ICHF2 are calculated by plotting the capacity increment per unit voltage change and extracting the current and voltage peaks corresponding to the curves;

[0018] The indirect health indicators are preprocessed using maximum and minimum normalization, and the correlation coefficients between all indirect health indicators and the true value of the battery health state SOH are calculated using the Pearson correlation analysis method, and indirect health indicators with correlation coefficients greater than 0.8 are extracted as pre-input variables for the comprehensive estimation model of the lithium battery health state.

[0019] Exemplarily, the method of using the KPCA method to perform dimensionality reduction processing on the indirect health indicator to obtain the first principal component data, construct a comprehensive estimation model for the health status of the lithium battery, and then obtain the battery health status includes:

[0020] The pre-input variables are mapped to a high-dimensional feature space by applying a kernel function using a KPCA method for dimensionality reduction, and the first principal component in the kernel principal component analysis is extracted as a fusion health indicator;

[0021] Introduce autoencoders into extreme learning machines, stack multiple extreme learning machine units with autoencoders to build a deep extreme learning machine model;

[0022] The fusion health index is used as the model input variable, and the improved sparrow search algorithm based on the hybrid strategy is used to optimize the model parameters, and the root mean square error, mean absolute error and determination coefficient are introduced to verify the accuracy of the model estimation;

[0023] The data to be detected is input into the lithium battery health state comprehensive estimation model to obtain the battery health state.

[0024] Exemplarily, the method of optimizing model parameters using the improved sparrow search algorithm based on a hybrid strategy includes:

[0025] Initialize the population size, maximum number of iterations, producer ratio, alert ratio and warning value parameters, and use the reverse learning strategy to initialize the population;

[0026] Calculate the fitness value of each sparrow, determine the individual positions of the sparrows with the best and worst fitness values, extract the producers, set the sparrows other than the producers as followers, randomly select the guards, and update the positions of the producers, followers and guards;

[0027] The initial weights of the comprehensive estimation model of the health status of lithium batteries are optimized, and the optimal individual position of the sparrow is iteratively updated until the iteration end condition is met. The optimal individual position of the sparrow and the corresponding fitness value are used as output to obtain the optimal initial weight.

[0028] Exemplarily, the population size is 30, the maximum number of iterations is 1000, and the ratio of alerters is 0.2.

[0029] Exemplarily, it also includes: using the root mean square error between the SOH estimated value and the true value as the fitness function, and the smaller the fitness function is, the higher the coincidence degree between the estimation result and the true value is.

[0030] Exemplarily, the CEEMDAN algorithm is used to decompose the original battery capacity sequence into a local fluctuation sequence and a global degradation sequence, and radial basis function neural network and Gaussian process regression are used to perform predictions respectively, and the prediction results are superimposed to obtain the battery remaining life prediction value, which includes:

[0031] The battery capacity sequence data is decomposed and reconstructed using the CEEMDAN algorithm to obtain a local fluctuation sequence of data jitter and a global degradation sequence of time series linearity, and the local fluctuation sequence and the global degradation sequence are divided into a training set and a test set, wherein the local fluctuation sequence is an intrinsic mode component and the global degradation sequence is a residual component;

[0032] Establishing a radial basis function neural network and a Gaussian process regression model, initializing the model, training the Gaussian process regression model using a training set divided by a local fluctuation sequence, and training the radial basis function neural network using a training set divided by a global degradation sequence;

[0033] The divided test sets are input into the trained radial basis function neural network and Gaussian process regression model respectively, and the output results of each model are superimposed to obtain the predicted value of the remaining battery life and the confidence interval of the response.

[0034] Another embodiment of the present invention provides a lithium battery health status prediction system based on fusion health indicators, including:

[0035] A data acquisition and analysis unit, used to acquire the charging and discharging parameters and data sets of the lithium battery, and to extract indirect health indicators for reflecting the battery degradation characteristics, and to verify the rationality of the indirect health indicators by using a correlation analysis method;

[0036] A model building unit, used to perform dimensionality reduction processing on the indirect health indicators using a KPCA method to obtain first principal component data, build a comprehensive estimation model for the health status of the lithium battery, and then obtain the health status of the battery;

[0037] The calculation prediction unit is used to decompose the original battery capacity sequence into a local fluctuation sequence and a global degradation sequence using the CEEMDAN algorithm, and to make predictions using a radial basis function neural network and a Gaussian process regression respectively, and to superimpose the prediction results to obtain a predicted value of the remaining battery life.

[0038] Another embodiment of the present invention provides a terminal, comprising: a processor and a memory, wherein the memory stores a computer program, and the processor is used to execute the computer program to implement the above-mentioned lithium battery health status prediction method based on fusion health indicators.

[0039] Yet another embodiment of the present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed, the above-mentioned lithium battery health status prediction method based on fusion health indicators is implemented.

[0040] The present invention proposes a lithium battery health status prediction method and system based on fusion health indicators, including obtaining lithium battery charge and discharge parameters and data sets, and extracting indirect health indicators used to reflect battery degradation characteristics, using correlation analysis to verify the rationality of indirect health indicators; using the KPCA method to reduce the dimension of the indirect health indicators to obtain the first principal component data, constructing a comprehensive estimation model for the health status of lithium batteries, and then obtaining the battery health status; decomposing the original battery capacity sequence into a local fluctuation sequence and a global degradation sequence, using radial basis function neural network and Gaussian process regression to predict respectively, and superimposing to obtain the battery remaining life prediction value. In view of the problems of low accuracy and unstable model in the prior art, the present invention adopts a method based on fusion health indicators to measure the battery health status, which has the advantages of high estimation accuracy and good application prospects, and effectively improves the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solution of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope of protection of the present invention. In each of the drawings, similar components are numbered similarly.

[0042] Figure 1 A schematic flow chart of a lithium battery health status prediction method based on fusion health indicators provided in an embodiment of the present invention;

[0043] Figure 2 A schematic diagram of a battery charging voltage curve provided by an embodiment of the present invention;

[0044] Figure 3 A schematic diagram of a battery discharge voltage curve provided by an embodiment of the present invention;

[0045] Figure 4 A schematic diagram of a battery charging current curve provided by an embodiment of the present invention;

[0046] Figure 5 A schematic diagram of a battery discharge current curve provided by an embodiment of the present invention;

[0047] Figure 6 The charging IC curves of the battery provided in the embodiment of the present invention under different cycles. DETAILED DESCRIPTION

[0048] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0049] As an important component of the battery management system, the state of health (SOH) cannot be directly measured by sensors, but can only be estimated by measuring other characterization parameters. It is usually defined as the ratio of a certain parameter to the nominal parameter after the battery has been used for a period of time. As the battery continues to age, many parameters will change, so parameters from different angles will be selected to define SOH, such as defining SOH by capacity, internal resistance or number of cycles. Although it is relatively easy to measure the number of charge and discharge cycles, the battery's charge and discharge lack regularity when used by users, and it is almost impossible to go through a complete charge and discharge cycle. The change in the internal resistance of the battery during the initial charge and discharge is not obvious, which will cause a large overall measurement error, and the measurement results of the internal resistance using internal resistance meters from different manufacturers vary greatly, so the application value of defining SOH by internal resistance is not high. The method of defining SOH by capacity is more intuitive and widely used, so the embodiment of the present invention uses the following formula to measure the size of SOH.

[0050]

[0051] Among them, C i is the maximum available capacity of the current cycle; C new The nominal capacity. When the maximum available capacity of the battery decays to 70% to 80% of the nominal capacity, it indicates that the battery life is over.

[0052] The reasonable selection of health indicators often determines the accuracy of the data-driven method, which is mainly divided into direct health indicators and indirect health indicators. Direct health indicators can directly describe the aging state of the battery, such as a decrease in capacity or an increase in internal resistance. At present, it is difficult to directly measure the real-time capacity, and the process is relatively cumbersome. Although the internal resistance can be measured by electrochemical impedance spectroscopy (EIS) frequency scanning, strict laboratory conditions are required. Indirect health indicators are parameters that can indirectly characterize the aging state of the battery, such as voltage, current and temperature. Compared with direct health indicators, indirect health indicators are easier to obtain and more convenient for correlation analysis. Reference Figure 1 , an embodiment of the present invention provides a lithium battery health status prediction method based on fusion health indicators, mainly comprising: step S101, obtaining lithium battery charge and discharge parameters and data sets, and extracting indirect health indicators for reflecting battery degradation characteristics, and using correlation analysis method to verify the rationality of the indirect health indicators;

[0053] Specifically, the charge and discharge parameters include charge current, charge cut-off voltage, charge cut-off current, discharge current, discharge cut-off voltage, temperature and nominal capacity. The data set includes complete charge and discharge cycle data after multiple charge and discharge cycles with abnormal points removed.

[0054] Indirect health indicators can be simply divided into three categories: voltage-related, current-related, and IC curve-related, including constant voltage rise charging time VHF1, constant voltage drop discharge time VHF2, constant current charging time ratio CHF1, constant current charging current curve coverage area CHF2, IC curve peak ICHF1, and IC curve peak voltage ICHF2. Due to the differences in the charging and discharging conditions of lithium batteries in different data sets, individual health indicators need to be adjusted in a targeted manner. For example, VHF1 and VHF2 need to select a reasonable charging interval or discharging interval when the data set changes. The specific relevant intervals are adjusted as follows: In the NASA data set, VHF1 is the time consumed by the charging voltage rising from 3.9V to 4.1V, and VHF2 is the time consumed by the discharge voltage falling from 3.9V to 3.6V; in the Oxford data set, VHF1 is the time consumed by the charging voltage rising from 3.8V to 4.1V, and VHF2 is the time consumed by the discharge voltage falling from 3.9V to 3.7V.

[0055] Figure 2 and Figure 3The charge and discharge voltage curves of the battery at the 20th, 50th, 80th, 110th, 140th, and 160th cycles were selected. The horizontal axis represents the time required for a complete charge and discharge process, and the vertical axis represents the voltage during each charge and discharge process. Figure 2 The curves between 0 and 3000 s are getting darker from left to right, corresponding to the 160th to 20th cycle charge and discharge from left to right, and the 20th cycle curve has a sudden change at around 9000 s. Figure 3 The color of the curves between 0 and 2000 s becomes darker from left to right, corresponding to the 160th to 20th cycle charge and discharge from left to right, and the structures of the curves are similar. Figure 4 and Figure 5 The charging and discharging current curves of the battery at the 10th, 40th, 70th, 100th, 130th, and 160th cycles were selected. The horizontal axis represents the time required for a complete charging or discharging process, and the vertical axis represents the current size during each charging and discharging process. Figure 4 The curves between 2000 and 8000 s are getting darker from left to right, corresponding to the 160th to 10th cycle charge and discharge from left to right. Figure 5 The color of the curve between 2000 and 4000 s becomes darker from left to right, corresponding to the 160th to 10th cycle charge and discharge from left to right. Figure 6 The IC curves of the battery under different cycles are shown in the embodiment of the present invention. The sampling time interval is set to 0.015 seconds, and the capacity increments of the 1st, 30th, 60th, 90th, 120th, and 150th cycles are extracted from the constant current charging voltage data, and the data are smoothed and denoised by Kalman filtering. Figure 6 The middle curve becomes lighter in color from top to bottom, corresponding to the 1st to 150th cycle charge and discharge. Figures 2 to 6 The schematic diagram can give a general understanding of the changing trend of the relevant curves, which is helpful for understanding the selection of indicators and will not be elaborated here.

[0056] The embodiment of the present invention adopts the Pearson method to calculate the correlation coefficient r. When -1<r<0, it means that the health indicator is negatively correlated with SOH; when 0<r<1, it means that the health indicator is positively correlated with SOH. If |r| is closer to 1, the stronger the correlation is; if |r| is closer to 0, the weaker the correlation is. The maximum-minimum normalization is used to preprocess the data of indirect health indicators. The purpose is to scale the values ​​of different health indicators to a similar range to eliminate the deviation caused by the characteristics of different health indicators, so that the model converges better and improves the stability of training.

[0057] Step S102, using the KPCA method to perform dimensionality reduction processing on the indirect health indicators to obtain the first principal component data, construct a comprehensive estimation model for the health status of the lithium battery, and then obtain the battery health status;

[0058] Specifically, if all health indicators are directly input into the model, not only will the training efficiency of the model be reduced, but it will also be impossible to determine whether there is redundancy between the health indicators. Therefore, a good health indicator should be able to map the aging trend of the battery and have the significant advantages of low dimension and low redundancy to avoid overfitting of the model. The embodiment of the present invention uses the kernel principal component analysis (KPCA) method to reduce the dimension of the extracted 6 groups of health indicators, screen out the best health indicators for the training and estimation of the SOH estimation model, and select the first principal component as the fusion health indicator. Through experimental measurement, the embodiment of the present invention selects the deep extreme learning machine (DELM) as the battery SOH estimation model. In order to make the result estimation accurate, the embodiment of the present invention adopts a hybrid strategy to improve the SSA algorithm in three aspects: initial population, producer position update and follower position update, and then optimize the initial weight of the deep extreme learning machine. The specific steps are as follows: (1) Initialize parameters, such as population size, maximum number of iterations, producer ratio, vigilant ratio, warning value, etc., and use the reverse learning strategy to initialize the sparrow population. (2) Calculate the fitness value of each sparrow and determine the individual positions of the sparrows with the best and worst current fitness values. (3) Select some sparrows from the sparrows with better fitness values ​​as producers and update their positions. (4) The remaining sparrows update their positions as followers; finally, randomly select some sparrows as sentinels and update their positions. (5) Determine whether the end condition is met. If so, proceed to (6), otherwise return to (3). (6) End the iteration and output the optimal individual sparrow position and fitness value.

[0059] Step S103, decomposing the original battery capacity sequence into a local fluctuation sequence of data jitter and a global degradation sequence of time series linearity, using radial basis function neural network and Gaussian process regression to perform predictions respectively, and superimposing the prediction results to obtain a predicted value of the remaining battery life.

[0060] Specifically, the remaining service life prediction of lithium batteries can be regarded as a prediction of a time series, but lithium batteries will have capacity regeneration and noise interference during the degradation process, which greatly reduces the accuracy of the prediction model. In this regard, an embodiment of the present invention proposes a remaining service life RUL prediction method that combines CEEMDAN, radial basis function neural network RBF and Gaussian process regression GPR. First, the CEEMDAN algorithm is used to decompose and reconstruct the battery capacity sequence data, and then the GPR model and RBF neural network are used to predict the local fluctuation sequence and the global degradation sequence respectively. The RBF neural network has significant advantages in predicting time series and global approximation. The GPR model can simultaneously output the predicted value and uncertainty interval of RUL to achieve the purpose of improving the prediction accuracy of the model.

[0061] First, the capacity degradation sequence data is obtained from the battery data set, and the battery capacity sequence is decomposed using the decomposition step of the CEEMDAN algorithm to obtain several intrinsic mode components IMF and a residual component RES. Each IMF component is used to capture the capacity regeneration phenomenon of the battery, and the RES sequence reflects the global degradation trend of the capacity. The decomposed local fluctuation sequence and global degradation sequence are used to construct the data set, and the prediction starting point is set according to the number of cycles of the battery, and then the training set and test set are divided. Secondly, the GPR model and RBF neural network are established, and their respective structures are determined and the parameters are initialized. The GPR model is trained using the training set divided by the local fluctuation sequence. Similarly, the RBF neural network is trained using the training set divided by the global degradation sequence. Then, the test set is substituted into the trained GPR and RBF models respectively, and the output results of each model are superimposed to obtain the final prediction result and 95% confidence interval of the battery capacity. Finally, the actual capacity corresponding to each battery when it reaches the life threshold is obtained from the battery data set, and then the actual value and predicted value of the RUL of the battery at the current prediction starting point are calculated respectively, and the prediction accuracy of the model is analyzed by calculating the error evaluation index. Based on the estimated health status, the embodiment of the present invention can achieve short-term prediction and trend extrapolation through capacity degradation sequence decomposition and trend modeling, and gradually perform verification and optimization to obtain a more accurate battery remaining life value.

[0062] The present invention proposes a lithium battery health status prediction method and system based on fusion health indicators, including obtaining lithium battery charge and discharge parameters and data sets, and extracting indirect health indicators used to reflect battery degradation characteristics, using correlation analysis to verify the rationality of indirect health indicators; using the KPCA method to reduce the dimension of the indirect health indicators to obtain the first principal component data, constructing a comprehensive estimation model for the health status of lithium batteries, and then obtaining the battery health status; decomposing the original battery capacity sequence into a local fluctuation sequence and a global degradation sequence, using radial basis function neural network and Gaussian process regression to predict respectively, and superimposing to obtain the battery remaining life prediction value. In view of the problems of low accuracy and unstable model in the prior art, the present invention adopts a method based on fusion health indicators to measure the battery health status, which has the advantages of high estimation accuracy and good application prospects, and effectively improves the prediction accuracy.

[0063] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A lithium battery health status prediction method based on fusion health indicators, characterized in that: include: Obtaining lithium battery charge and discharge parameters and data sets, and extracting indirect health indicators used to reflect battery degradation characteristics, and using correlation analysis to verify the rationality of the indirect health indicators; The indirect health indicators are processed by KPCA method to obtain the first principal component data, and a comprehensive estimation model of the health status of the lithium battery is constructed to obtain the health status of the battery; The CEEMDAN algorithm is used to decompose the original battery capacity sequence into a local fluctuation sequence and a global degradation sequence. The radial basis function neural network and Gaussian process regression are used to predict them respectively, and the prediction results are superimposed to obtain the predicted value of the remaining battery life.

2. A lithium battery health status prediction method based on fusion health indicators according to claim 1, characterized in that: The charge and discharge parameters include charge current, charge cut-off voltage, charge cut-off current, discharge current, discharge cut-off voltage, temperature and nominal capacity. The data set is complete charge and discharge cycle data after multiple charge and discharge cycles with abnormal points removed.

3. A lithium battery health status prediction method based on fusion health indicators according to claim 2, characterized in that: The indirect health indicators include constant voltage rise charging time VHF1, constant voltage drop discharge time VHF2, constant current charging time ratio CHF1, constant current charging current curve coverage area CHF2, IC curve peak ICHF1 and IC curve peak corresponding voltage ICHF2. The extraction of indirect health indicators for reflecting battery degradation characteristics and the use of correlation analysis to verify the rationality of the indirect health indicators include: The VHF1, VHF2, CHF1, CHF2, ICHF1 and ICHF2 are calculated using the charge and discharge parameter data set, wherein the VHF1 is calculated by the moments corresponding to the charging voltages of 4.1V and 3.9V, respectively; the VHF2 is calculated by the moments corresponding to the charging voltages of 3.6V and 3.9V, respectively; the CHF1 is calculated by the ratio of the time consumed by constant current charging to the total charging time; the CHF2 is calculated by the integration of the constant current charging current on the time axis; the ICHF1 and ICHF2 are calculated by plotting the capacity increment per unit voltage change, and extracting the peak value of the curve and the corresponding voltage; The indirect health indicators are preprocessed using maximum and minimum normalization, and the correlation coefficients between all indirect health indicators and the true value of the battery health state SOH are calculated using the Pearson correlation analysis method, and indirect health indicators with correlation coefficients greater than 0.8 are extracted as pre-input variables for the comprehensive estimation model of the lithium battery health state.

4. The method for predicting the health status of a lithium battery based on fusion health indicators according to claim 3, characterized in that: The KPCA method is used to perform dimensionality reduction processing on the indirect health indicators to obtain the first principal component data, and a comprehensive estimation model for the health status of the lithium battery is constructed to obtain the battery health status, which includes: Using the KPCA method, the pre-input variables are mapped to the high-dimensional feature space by applying the kernel function to perform dimensionality reduction processing, and the first principal component in the kernel principal component analysis is extracted as the fusion health indicator; Introduce autoencoders into extreme learning machines, stack multiple extreme learning machine units with autoencoders to build a deep extreme learning machine model; The fusion health index is used as the model input variable, and the improved sparrow search algorithm based on the hybrid strategy is used to optimize the parameters, and the root mean square error, mean absolute error and determination coefficient are introduced to verify the accuracy of the model estimation; The data to be detected is input into the lithium battery health state comprehensive estimation model to obtain the battery health state.

5. The method for predicting the health status of a lithium battery based on fusion health indicators according to claim 4, characterized in that: The method of optimizing model parameters by using the improved sparrow search algorithm based on the hybrid strategy includes: Initialize the population size, maximum number of iterations, producer ratio, alert ratio and warning value parameters, and use the reverse learning strategy to initialize the population; Calculate the fitness value of each sparrow, determine the individual positions of the sparrows with the best and worst fitness values, extract the producers, set the sparrows other than the producers as followers, randomly select the guards, and update the positions of the producers, followers and guards; The initial weights of the comprehensive estimation model of the health status of lithium batteries are optimized, and the optimal individual position of the sparrow is iteratively updated until the iteration end condition is met. The optimal individual position of the sparrow and the corresponding fitness value are used as output to obtain the optimal initial weight.

6. A lithium battery health status prediction method based on fusion health indicators according to claim 5, characterized in that: The population size is 30, the maximum number of iterations is 1000, and the alerter ratio is 0.

2.

7. The method for predicting the health status of a lithium battery based on fusion health indicators according to claim 5, characterized in that: Also includes: The root mean square error between the SOH estimated value and the true value is used as the fitness function. The smaller the fitness function is, the higher the coincidence degree between the estimated result and the true value is.

8. The method for predicting the health status of a lithium battery based on fusion health indicators according to claim 1, characterized in that: The CEEMDAN algorithm is used to decompose the original battery capacity sequence into a local fluctuation sequence and a global degradation sequence, and radial basis function neural network and Gaussian process regression are used to perform predictions respectively, and the prediction results are superimposed to obtain the battery remaining life prediction value including: The battery capacity sequence data is decomposed and reconstructed using the CEEMDAN algorithm to obtain a local fluctuation sequence of data jitter and a global degradation sequence of time series linearity, and the local fluctuation sequence and the global degradation sequence are divided into a training set and a test set, wherein the local fluctuation sequence is an intrinsic mode component and the global degradation sequence is a residual component; Establishing a radial basis function neural network and a Gaussian process regression model, initializing the model, training the Gaussian process regression model using a training set divided by a local fluctuation sequence, and training the radial basis function neural network using a training set divided by a global degradation sequence; The divided test sets are input into the trained radial basis function neural network and Gaussian process regression model respectively, and the output results of each model are superimposed to obtain the predicted value of the remaining battery life and the confidence interval of the response.

9. A lithium battery health status prediction system based on fusion health indicators, characterized in that: include: A data acquisition and analysis unit, used to acquire the charging and discharging parameters and data sets of the lithium battery, and to extract indirect health indicators for reflecting the battery degradation characteristics, and to verify the rationality of the indirect health indicators by using a correlation analysis method; A model building unit, used to perform dimensionality reduction processing on the indirect health indicators using a KPCA method to obtain first principal component data, build a comprehensive estimation model for the health status of the lithium battery, and then obtain the health status of the battery; The calculation prediction unit is used to decompose the original battery capacity sequence into a local fluctuation sequence and a global degradation sequence using the CEEMDAN algorithm, and to make predictions using a radial basis function neural network and a Gaussian process regression respectively, and to superimpose the prediction results to obtain a predicted value of the remaining battery life.

Images