Lithium battery SOH prediction method based on AO-AVOA-BP neural network model

Abstract

The invention discloses a lithium battery SOH prediction method based on an AO-AVOA-BP neural network model. A prediction model of a BP neural network is optimized based on an African vulture optimization algorithm AVOA and an eagle optimization algorithm AO. Through analysis of voltage, current and temperature data in a battery charging process, correlation between health factors and battery SOH is verified based on grey correlation analysis, four health factors are determined as input of a model, and more accurate SOH prediction is realized in combination with a BP neural network model based on AO-AVOA optimization. The SOH of the lithium battery is predicted through the provided model and other optimization models, all indexes are compared and analyzed, and the result shows that the mean absolute error of the provided prediction model is smaller than 0.0089, the root-mean-square error is smaller than 0.0112, and the method has the advantages of being high in precision, high in generalization and the like and can be effectively used for SOH prediction of the lithium battery.

Description

Technical Field

[0001] The present invention belongs to the field of lithium battery health prediction, and in particular relates to a lithium battery SOH prediction method based on an AO-AVOA-BP neural network model. Background Art

[0002] Lithium-ion batteries are favored by people due to their high energy density, long service life, high stability and moderate price. They are widely used in electric vehicles, aerospace and smart grid. However, as the number of battery charge and discharge cycles accumulates, lithium-ion batteries will inevitably age, resulting in a decrease in available battery capacity and safety issues. Therefore, accurate and effective SOH prediction is of great significance to the efficient operation of electric vehicles and is a key technology to prevent battery failure and ensure safety.

[0003] Currently, there are two main methods for battery SOH prediction: model-based methods and data-driven methods. Model-based methods use precise modeling to analyze the intrinsic relationship between battery capacity and the number of charge and discharge cycles, including electrochemical models and equivalent circuit models. For model-based methods, it is difficult to solve the complexity of the internal principles of the battery and the high uncertainty of external operating conditions. Data-driven methods rely on the acquired battery aging data, but the performance of such methods, such as BP neural networks, is highly dependent on the initialization and subsequent optimization of model parameters. Traditional parameter optimization methods may not be able to fully explore the parameter space, causing the model to fall into a local optimal solution, affecting the prediction accuracy. A single metaheuristic algorithm often has an imbalance in exploration and development capabilities. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a lithium battery SOH prediction method based on the AO-AVOA-BP neural network model. The SOH prediction model proposed in the present invention has the characteristics of high accuracy, strong robustness, strong generalization, etc., and can be effectively used for lithium battery SOH prediction.

[0005] Technical solution: A lithium battery SOH prediction method based on an AO-AVOA-BP neural network model of the present invention comprises the following steps:

[0006] Step 1: Analyze the data during the battery charging process, select health factors with high correlation with battery SOH using grey correlation analysis, and construct an input feature set for lithium battery SOH prediction;

[0007] Step 2: construct an improved AO-AVOA algorithm;

[0008] Step 3: Optimize the weights and thresholds of the BP neural network model through the improved AO-AVOA algorithm, build a battery SOH prediction model of the AO-AVOA-BP neural network, and train the model based on the input feature set;

[0009] Step 4: Experimentally verify the model using the NASA battery dataset, and predict the SOH of lithium batteries based on the model.

[0010] Furthermore, step 1 specifically includes the following steps:

[0011] Step 1.1. Battery SOH is defined from the capacity perspective, as shown in formula (1):

[0012]

[0013] In the formula, C now Indicates the current maximum capacity of the battery when fully charged, C rated Indicates the rated capacity of the battery when it leaves the factory;

[0014] Step 1.2, extracting health factors from relevant data of constant current charging stage CC and constant voltage charging stage CV to predict the SOH of the battery, and obtaining the charging time changes of CC charging stage and CV charging stage under different cycle states;

[0015] Step 1.3: To test the correlation between the selected health factors and the battery SOH, grey relational analysis (GRA) is used to measure the quality of feature points, and the feature vector is composed of health factors.

[0016] Further, in step 1.2, the different cycle states include 10, 40, 70, 100, 130 and 160 cycle states, and the health factor includes CC charging time H 1 、CC charging capacity H 2 、CC charging temperature rise change H 3 、CV charging time H 4 、CV charging capacity H 5 、CV charging average temperature H 6 , the ratio of CC charging time to CV charging time H 7 、The ratio of CC charging capacity to CV charging capacity H 8 .

[0017] Furthermore, step 1.3 is as follows: first, determine the label vector and the feature vector, the label vector is SOH, and the feature vector is composed of health factors. Then, use the initial value method to perform dimensionless processing on the original data, as shown in formula (2):

[0018]

[0019] In the formula, is the dimensionless comparable eigenvector, x i (1) is the first data of the i-th sequence in the original data sequence, k is the dimension of the vector, x i (k) is the kth data point in the original data sequence. The relationship between the comparable feature data and the label data is clarified by calculating the grey correlation coefficient ξ between the comparable feature vector and the label vector, as shown in formula (3):

[0020]

[0021] In the formula, y i (k) is the label vector, ρ is the resolution coefficient, which is usually 0.5. Finally, the above grey correlation coefficients are weighted averaged to obtain the grey correlation coefficient r between the i-th feature vector and the label vector. i , as shown in formula (4):

[0022]

[0023] The purpose of calculating the grey correlation coefficient is to quantitatively evaluate the strength of the correlation between the feature vector and the label vector; if the correlation between the feature vector and the label vector is stronger, the grey correlation coefficient is closer to 1, and vice versa.

[0024] Furthermore, step 2 specifically includes the following steps:

[0025] Step 2.1, construct the eagle optimization algorithm AO, which is based on the population meta-heuristic optimization method and optimizes the problem by simulating the hunting behavior of the eagle;

[0026] Step 2.2, construct the African vulture optimization algorithm AVOA, which is a meta-heuristic swarm intelligence optimization algorithm, and solves the optimization problem by simulating the food hunting and flight trajectory behavior of African vultures;

[0027] Step 2.3: Construct the improved AO-AVOA algorithm.

[0028] Furthermore, step 2.1 is specifically as follows:

[0029] In the first stage, the search space is extensively explored to search for prey. The mathematical model of this behavior is as follows:

[0030]

[0031] Where, X m (t) represents the average position of all individuals in the current population, X best(t) represents the best position achieved so far, N is the population size, rand is a random number between 0 and 1, t and T are the current iteration and the maximum number of iterations respectively;

[0032] In the second stage, the exploration is narrowed down, and the prey is attacked by short gliding. The formula for updating the position of the Sky Eagle is as follows:

[0033] X 2 (t+1)=X best (t)×LF(D)+X R (t)+(yx)×rand)(7)

[0034] In the formula, D is the dimensional space, X R (t) represents the random position of the Skyhawk, LF(D) represents the Levy flight function, and x and y represent the spiral shape of the algorithm in the search space;

[0035] The third stage is to expand the development and vertically land to initially attack the prey. This stage can be expressed by the mathematical formula:

[0036] X 3 (t+1)=(X best (t)-X m (t))×α-rand+((UB-LB)×rand+LB)×δ (8)

[0037] Where UB and LB are the upper and lower bounds of the expected problem, respectively, and α and δ are development adjustment parameters equal to 0.1;

[0038] The fourth stage, shrinking, exploiting, moving and catching prey, is mathematically modeled as follows:

[0039] X 4 (t+1)=QF×X best (t)-(G 1 ×X(t)×rand)-G 2 ×LF(D)+rand×G 1 (9)

[0040] Where QF represents the quality function used to balance the search strategy, G 1 represents the motion parameters when tracking prey, G 2 Indicates the flight slope when chasing prey.

[0041] Furthermore, step 2.2 is specifically as follows:

[0042] In the first stage, the best vulture individual is determined, and its mathematical expression is:

[0043]

[0044] Where R(i) represents the position of the best vulture, BestVuture 1,2 are the best vulture and the second best vulture, L 1 and L 2 It is a parameter whose value is between 0 and 1 initialized before the search operation. The sum of the two parameters is equal to 1. i is the probability of the optimal vulture being selected;

[0045] In the second stage, the vulture's hunger index is calculated, and its mathematical expression is:

[0046]

[0047] Where F is the vulture's hunger index, t is the current iteration number, T is the total number of iterations, and rand 1 , z and h are all random numbers, and rand 1 ∈(0,1), z∈(-1,1), h∈(-2,2), ω is a constant parameter set before the optimization operation, and its value is 2.5;

[0048] The third stage is the exploration stage. When |F|≥1, the vulture searches for food in different areas, and the algorithm enters the exploration stage. Its mathematical expression is:

[0049]

[0050] Where P(i) is the current position of the vulture, P(i+1) represents the position of the vulture in the next iteration, X=2×rand is the coefficient vector for increasing the random motion of the vulture, and rand 2 , rand 3 ∈(0,1), ub and lb are the upper and lower bounds of the expected problem respectively;

[0051] The fourth stage is the development stage. When |F|<1, AVOA enters the development stage. When |F|∈(0.5,1), it enters the first stage of the development stage. When , the slow siege strategy is implemented, and its mathematical expression is:

[0052]

[0053] When , the rotating flight strategy is executed, and its mathematical expression is:

[0054]

[0055] When |F|∈(0,0.5), we enter the second stage of the development phase. Many vulture species will gather around food sources, and the mathematical expression for this behavior is:

[0056]

[0057] if Then implement the siege strategy, and the mathematical expression of the position of this behavior is:

[0058] P(i+1)=R(i)-|d(t)×F×LF(D)(16).

[0060] Furthermore, step 2.3 is specifically as follows: an improved African vulture fusion eagle optimization algorithm AO-AVOA is proposed, and the original position update rules are replaced by formulas (5) and (7) in the exploration phase of AVOA, as shown below:

[0061]

[0062] The exploration phase of AO is integrated with the development phase of AVOA, retaining their high efficiency in global and local search and improving the convergence speed.

[0063] Furthermore, step 3 is specifically as follows: using the African vulture fusion eagle optimization algorithm to optimize the weights and thresholds of the BP neural network model, and establishing an AO-AVOA-BP neural network battery SOH prediction model, specifically: first, constructing a BP neural network model, determining the network topology, and initializing the BP neural network weights and thresholds; secondly, initializing the AO-AVOA algorithm parameters and population positions, etc., according to the fitness function, finding the optimal and suboptimal vulture positions, and updating the entire population position; then, using the results as the optimal solution for the weights and thresholds, and inputting them into the BP neural network for continued training, and completing the algorithm optimization process after the conditions are met; finally, using the trained AO-AVOA-BP neural network model to predict the SOH of the battery data.

[0064] Further, step 4 is specifically as follows: using MATLAB as a platform, building models of BP neural network algorithm, PSO-BP neural network algorithm, AO-BP neural network algorithm, AVOA-BP neural network algorithm and AO-AVOA-BP neural network algorithm, taking the screened 4 health factors as input, battery SOH as output, selecting the first 70% of the battery cycle data as a training set, and the remaining data as a test set, to predict the battery SOH. In order to measure the prediction performance of the model, the mean absolute error MAE, root mean square error RMSE and mean absolute percentage error MAPE are selected as evaluation functions, and the calculation formula is as follows:

[0065]

[0066] Where N is the total number of battery cycles; y iis the actual value of the battery SOH, is the predicted value.

[0067] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: The present invention proposes a SOH prediction model based on the AO-AVOA-BP neural network combination model. First, by deeply analyzing the data during the battery charging process, the gray correlation analysis method is used to carefully screen out four health factors that are highly correlated with the battery SOH: CC charging time, CC charging capacity, CV charging average temperature and the ratio of CC charging capacity to CV charging capacity as the input feature set, thereby constructing an accurate input feature set. On this basis, this paper innovatively combines the advantages of the Sky Eagle optimization algorithm and the African Vulture optimization algorithm, and proposes an improved AO-AVOA algorithm. The fusion of the exploration phase of AO and the development phase of AVOA not only retains the high efficiency of both in global and local search, but also improves the convergence speed. Using this improved algorithm, we have deeply optimized the weights and thresholds of the BP neural network and successfully constructed the SOH prediction model of the AO-AVOA-BP neural network. The accuracy and generalization of the designed model were verified by using B0005, B0006, B0007, and B00018 batteries. The results showed that the MAE of the battery SOH predicted by this method was less than 0.0089, the RMSE was less than 0.0112, and the MAPE was less than 1.4512%. The SOH prediction accuracy on multiple battery samples was significantly better than other comparison models. Therefore, the SOH prediction model proposed in the present invention has the characteristics of high accuracy, strong robustness, and strong generalization. It not only provides new ideas and solutions for the SOH prediction of lithium batteries, but also further expands the application scope of metaheuristic algorithms in the field of battery health management. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 This is the SOH change curve of lithium battery;

[0069] Figure 2 The charging curves of different cycle numbers, including (a) charging cycle current change curve; (b) charging cycle voltage change curve;

[0070] Figure 3 The capacity and time variation curves of different charging stages, including (a) the capacity and time variation curve of CC charging stage; (b) the capacity and time variation curve of CV charging stage;

[0071] Figure 4 The temperature change curves for different charging cycles;

[0072] Figure 5 It is the battery temperature rise and average temperature change curve;

[0073] Figure 6 This is the flow chart of the AO-AVOA-BP model;

[0074] Figure 7 The following is the overall flow chart;

[0075] Figure 8 The SOH prediction results and errors of different models of B0005 battery; (a) prediction curve; (b) error curve; DETAILED DESCRIPTION

[0076] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0077] The battery data used in the present invention comes from the battery data set of the National Aeronautics and Space Administration (NASA) Center of Excellence Forecasting Data Set. The battery is a 18650 lithium-ion power battery. The present invention selects B0005, B0006, B0007, and B0018 as data support. Its rated current is 2.2Ah and its rated voltage is 3.7V. At a room temperature of 24°C, all batteries are tested in three modes: constant current-constant voltage charging mode, constant current discharge mode, and electrochemical impedance spectroscopy impedance test mode. In the constant current (constant current, CC) charging stage, a constant current of 1.5A is used to charge the battery until the voltage climbs to 4.2V. Subsequently, the constant voltage (constant voltage, CV) charging stage is entered, and this voltage is maintained to continue charging until the charging current gradually decreases to 20mA. In the discharge phase, the battery is discharged at a constant current of 2A until the voltage of B0005, B0006, B0007, and B0018 drops to 2.7V, 2.5V, 2.2V, and 2.5V, respectively. When the rated capacity of the battery is reduced by 30%, the battery life ends and the experiment is stopped. Figure 1 shown.

[0078] In view of the fact that the discharge conditions of lithium-ion batteries are usually highly unstable and difficult to accurately control during actual operation, the environment in which lithium-ion batteries are located is relatively stable during charging. Therefore, the present invention selects to extract health factors from the relevant data of the CC charging stage and the CV charging stage to predict the SOH of the battery. Taking the B0005 battery as an example, the charging time changes of the CC charging stage and the CV charging stage under the conditions of 10, 40, 70, 100, 130 and 160 cycles are obtained, as shown in FIG. Figure 2 It can be seen that with the increase of the number of cycles, the charging time of the battery in the CC charging stage gradually decreases from 3221s to 1579s, while the charging time of the battery in the CV charging stage gradually increases from 6900s to 8700s with the increase of the number of cycles.

[0079] Figure 3 The capacity and time variation curves of different charging stages. Figure 3It can be seen that in the CC charging stage, as the number of cycles increases, the time required for charging and the charging capacity both show a trend of gradually decreasing. On the contrary, in the CV charging stage, the time required for charging and the charging capacity increase with the increase in the number of cycles. Therefore, mathematical methods can be used to process the original parameters of lithium-ion power batteries such as voltage and current to construct new features as the health factors of lithium-ion batteries, namely CC charging time, CC charging capacity, CV charging time and CV charging capacity. From the perspective of battery charging integrity, the entire charging process can be considered an important factor in the health of lithium-ion batteries. Therefore, the ratio of CC charging time to CV charging time and the ratio of CC charging capacity to CV charging capacity can be selected as battery health factors.

[0080] In addition, the temperature change curves under the 10th, 40th, 70th, 100th, 130th, and 160th cycles are as follows: Figure 4 As shown. Figure 4 It can be seen that for each cycle, the temperature of the lithium-ion battery first drops, then rises, and finally drops to a stable state. From the different cycles of the battery's life cycle, as the number of cycles gradually increases, the temperature change of the battery also shows a specific trend. This trend is particularly significant in the temperature rise change in the CC charging stage and the average temperature change in the CV charging stage. Figure 5 The figure shows the changes of both with the number of battery cycles. Although the temperature change of the battery has no direct correlation with the capacity, a series of complex chemical reactions will occur during the charging and discharging process of the battery, which will inevitably lead to the generation of heat and temperature fluctuations. Therefore, the temperature change of the battery can be selected as the health factor for battery SOH estimation.

[0081] In summary, a total of 8 health factors are selected as input to estimate the SOH of lithium-ion batteries, namely: CC charging time H 1 、CC charging capacity H 2 、CC charging temperature rise change H 3 、CV charging time H 4 、CV charging capacity H 5 、CV charging average temperature H 6 , the ratio of CC charging time to CV charging time H 7 、The ratio of CC charging capacity to CV charging capacity H 8 .

[0082] The purpose of calculating the grey correlation coefficient is to quantitatively evaluate the strength of the correlation between the feature vector and the label vector. If the correlation between the feature vector and the label vector is stronger, the grey correlation coefficient is closer to 1, and vice versa. The grey correlation coefficients of the four batteries are shown in Table 1. Taking the B0005 battery as an example, it can be seen from Table 1 that the effects of these parameters on the battery SOH are as follows: H 1 >H 2 >H 6 >H 8 >H 7 >H 5 >H 4 >H 3 For the determination of health factors, the four health factors with the highest grey correlation coefficients were selected as the input feature set, namely CC charging time, CC charging capacity, CV charging average temperature, and the ratio of CC charging capacity to CV charging capacity.

[0083] Table 1 Grey correlation coefficients between different health factors and SOH

[0084]

[0085] BP neural network is a multi-layer feedforward network with the characteristics of signal forward transmission and error back propagation

[25] . BP neural network has strong generalization ability. It can learn the mapping relationship of data through training process and better reflect the nonlinear attenuation process of battery capacity in the whole life cycle. However, BP neural network also has its limitations, especially the problem of parameter initialization. In order to solve this problem, the present invention is based on the established theoretical model, uses the African vulture fusion eagle optimization algorithm to optimize the weights and thresholds of the BP neural network model, and establishes the AO-AVOA-BP neural network battery SOH prediction model. The process is as follows: Figure 6 shown.

[0086] The overall process of the lithium battery SOH prediction model based on the AO-AVOA-BP neural network model is as follows: Figure 7 As shown. First, data preprocessing, input NASA original data set, extract charging data from the original data for analysis, divide the data set into training set and test set, calculate battery voltage, current and temperature data; second, health factor extraction, extract health factors from the battery charging process voltage, current and temperature data, and perform correlation analysis on the extracted health factors through grey correlation analysis to construct input feature set; then, establish a prediction model, take the weight and threshold of BP neural network as optimization objects, use improved AO-AVOA algorithm to optimize BP neural network, and construct AO-AVOA-BP prediction model; finally, SOH prediction, take the constructed input feature set as input, realize SOH prediction, and use evaluation index function to verify the accuracy of the proposed model.

[0087] In order to verify the prediction ability of the model proposed in this invention, it is compared and analyzed with BP neural network, PSO-BP neural network, AO-BP neural network and AVOA-BP neural network models. Taking B0005 battery as an example, the battery SOH prediction results and errors are shown in Figure 2. Figure 8 The B0006, B0007, and B0018 batteries were input into the model for prediction, and the comparison results of the evaluation indicators are shown in Table 2.

[0088] pass Figure 8 It can be found from Table 2 that compared with the other four models, the AO-AVOA-BP neural network model proposed in the present invention has better prediction effect, and the MAE, RMSE and MAPE indicators on all sample sets are better. The MAE error value corresponding to the proposed model is 0.0079, which is 51.34% lower than the MAE value of PSO-BP of 0.0162, 27.24% lower than the MAE value of AO-BP of 0.0108, and 56.71% lower than the MAE value of AVOA-BP of 0.0182; the RMSE error value is 0.0089, which is 49.81% lower than the RMSE value of PSO-BP of 0.0177, and 1.34% lower than the RMSE value of AO-BP of 0.0189. The SE value is 0.0119, which reduces the error by 25.47%. Compared with the RMSE value of AVOA-BP of 0.0214, the error is reduced by 58.60%. The MAPE error value is 1.0943%, which reduces the error by 51.59% compared with the MAPE value of PSO-BP of 2.2607%. Compared with the MAPE value of AO-BP of 1.5088%, the error is reduced by 27.47%. Compared with the MAPE value of AVOA-BP of 2.5507%, the error is reduced by 57.10%. In summary, the proposed AO-AVOA-BP model has higher accuracy in estimating the SOH of lithium-ion batteries than the traditional simple fusion model.

[0089] Table 2 Comparison of evaluation indicators of different models

[0090]

[0091] In order to further demonstrate the excellent performance of the AO-AVOA-BP neural network model proposed in the present invention, taking B0005 battery as an example, the prediction results of the model proposed in the present invention are compared with those of the models proposed in recent related literature. The comparison results are shown in Table 3.

[0092] Table 3 Comparison of evaluation indicators of the present invention and other literature models

[0093]

[0094] As can be seen from Table 3, the SOH prediction model based on the spatiotemporal convolutional network and pyramid segmentation attention fusion Transformer takes the time, temperature, voltage integral and internal resistance of the entire charging stage of constant current charging as input features, and the prediction errors MAE, RMSE and MAPE are only 0.0198, 0.0221 and 2.21%; the long short-term memory network based on Bayesian optimization takes the extracted partial incremental capacity curve as the feature of SOH prediction, and the prediction errors MAE, RMSE and MAPE are 0.0127, 0.0092 and 1.35% respectively; the temporal convolutional network is combined with the improved flower pollination algorithm, and the external sensor raw data and the ohmic resistance trajectory identified in the equivalent circuit model are input into the model to predict SOH, and the prediction errors MAE and RMSE are 0.0093 and 0.0097 respectively. In summary, the AO-AVOA-BP neural network model proposed in the present invention still shows relatively better performance in prediction performance, thereby verifying the effectiveness and superiority of the proposed method.

Claims

1. A lithium battery SOH prediction method based on AO-AVOA-BP neural network model, characterized in that: The steps include: Step 1: Analyze the data during the battery charging process, select health factors with high correlation with battery SOH using grey correlation analysis, and construct an input feature set for lithium battery SOH prediction; Step 2: construct an improved AO-AVOA algorithm; Step 3: Optimize the weights and thresholds of the BP neural network model through the improved AO-AVOA algorithm, build a battery SOH prediction model of the AO-AVOA-BP neural network, and train the model based on the input feature set; Step 4: Experimentally verify the model using the NASA battery dataset, and predict the SOH of lithium batteries based on the model.

2. A lithium battery SOH prediction method based on AO-AVOA-BP neural network model according to claim 1, characterized in that: Step 1 specifically includes the following steps: Step 1.

1. Battery SOH is defined from the capacity perspective, as shown in formula (1): In the formula, C now Indicates the current maximum capacity of the battery when fully charged, C rated Indicates the rated capacity of the battery when it leaves the factory; Step 1.2, extracting health factors from relevant data of constant current charging stage CC and constant voltage charging stage CV to predict the SOH of the battery, and obtaining the charging time changes of CC charging stage and CV charging stage under different cycle states; Step 1.3: To test the correlation between the selected health factors and the battery SOH, grey relational analysis (GRA) is used to measure the quality of feature points, and the feature vector is composed of health factors.

3. A lithium battery SOH prediction method based on AO-AVOA-BP neural network model according to claim 2, characterized in that: In step 1.2, the different cycle states include 10, 40, 70, 100, 130 and 160 cycle states, and the health factors include CC charging time H1, CC charging capacity H2, CC charging temperature rise change H3, CV charging time H4, CV charging capacity H5, CV charging average temperature H6, ratio of CC charging time to CV charging time H7, and ratio of CC charging capacity to CV charging capacity H8.

4. The lithium battery SOH prediction method based on the AO-AVOA-BP neural network model according to claim 2 is characterized in that: Step 1.3 is as follows: First, determine the label vector and feature vector, the label vector is SOH, and the feature vector is composed of health factors. Then, use the initial value method to perform dimensionless processing on the original data, as shown in formula (2): In the formula, is the dimensionless comparable eigenvector, x i (1) is the first data of the i-th sequence in the original data sequence, k is the dimension of the vector, x i (k) is the kth data point in the original data sequence. The relationship between the comparable feature data and the label data is clarified by calculating the grey correlation coefficient ξ between the comparable feature vector and the label vector, as shown in formula (3): In the formula, y i (k) is the label vector, ρ is the discrimination coefficient, and finally the above grey correlation coefficients are weighted averaged to obtain the grey correlation coefficient r between the i-th feature vector and the label vector i , as shown in formula (4): The purpose of calculating the grey correlation coefficient is to quantitatively evaluate the strength of the correlation between the feature vector and the label vector; If the correlation between the feature vector and the label vector is stronger, the grey correlation coefficient is closer to 1, otherwise it is closer to 0.

5. The lithium battery SOH prediction method based on the AO-AVOA-BP neural network model according to claim 1 is characterized in that: Step 2 specifically includes the following steps: Step 2.1, construct the eagle optimization algorithm AO, which is based on the population meta-heuristic optimization method and optimizes the problem by simulating the hunting behavior of the eagle; Step 2.2, construct the African vulture optimization algorithm AVOA, which is a meta-heuristic swarm intelligence optimization algorithm, and solves the optimization problem by simulating the food hunting and flight trajectory behavior of African vultures; Step 2.3: Construct the improved AO-AVOA algorithm.

6. A lithium battery SOH prediction method based on AO-AVOA-BP neural network model according to claim 5, characterized in that: Step 2.1 is as follows: In the first stage, the search space is extensively explored to search for prey. The mathematical model of this behavior is as follows: Where, X m (t) represents the average position of all individuals in the current population, X best (t) represents the best position achieved so far, N is the population size, rand is a random number between 0 and 1, t and T are the current iteration and the maximum number of iterations respectively; In the second stage, the exploration is narrowed down, and the prey is attacked by short gliding. The formula for updating the position of the Sky Eagle is as follows: X2(t+1)=X best (t)×LF(D)+X R (t)+(y-x)×rand)(7) In the formula, D is the dimensional space, X R (t) represents the random position of the Skyhawk, LF(D) represents the Levy flight function, and x and y represent the spiral shape of the algorithm in the search space; The third stage is to expand the development and vertically land to initially attack the prey. This stage can be expressed by the mathematical formula: X3(t+1)=(X best (t)-X m (t))×α-rand+((UB-LB)×rand+LB)×δ (8) Where UB and LB are the upper and lower bounds of the expected problem, respectively, and α and δ are development adjustment parameters equal to 0.1; The fourth stage, shrinking, exploiting, moving and catching prey, is mathematically modeled as follows: X4(t+1)=QF×X best (t)-(G1×X(t)×rand)-G2×LF(D)+rand×G1 (9) Where QF represents the quality function used to balance the search strategy, G1 represents the motion parameter when tracking prey, and G2 represents the flight slope when chasing prey.

7. The lithium battery SOH prediction method based on the AO-AVOA-BP neural network model according to claim 5, characterized in that: Step 2.2 is as follows: In the first stage, the best vulture individual is determined, and its mathematical expression is: Where R(i) represents the position of the best vulture, BestVuture 1,2 are the optimal vulture and the suboptimal vulture, L1 and L2 are parameters whose values ​​are initialized between 0 and 1 before the search operation, and the sum of the two parameters is equal to 1, P i is the probability of the optimal vulture being selected; In the second stage, the vulture's hunger index is calculated, and its mathematical expression is: Where F is the hunger index of the vulture, t represents the current number of iterations, T is the total number of iterations, rand1, z and h are all random numbers, and rand1∈(0,1), z∈(-1,1), h∈(-2,2), ω is a constant parameter set before the optimization operation, and its value is 2.5; The third stage is the exploration stage. When |F|≥1, the vulture searches for food in different areas, and the algorithm enters the exploration stage. Its mathematical expression is: Where P(i) is the current position of the vulture, P(i+1) represents the position of the vulture in the next iteration, X=2×rand is the coefficient vector for increasing the random motion of the vulture, rand2, rand3∈(0,1) are random values, ub and lb are the upper and lower bounds of the expected problem respectively; The fourth stage is the development stage. When |F|<1, AVOA enters the development stage. When |F|∈(0.5,1), it enters the first stage of the development stage. When , the slow siege strategy is implemented, and its mathematical expression is: When , the rotating flight strategy is executed, and its mathematical expression is: When |F|∈(0,0.5), we enter the second stage of the development phase. Many vulture species will gather around food sources, and the mathematical expression for this behavior is: if Then implement the siege strategy, and the mathematical expression of the position of this behavior is: P(i+1)=R(i)-|d(t)×F×LF(D)(16) 8. The lithium battery SOH prediction method based on the AO-AVOA-BP neural network model according to claim 6 is characterized in that: Step 2.3 is as follows: an improved African vulture fusion eagle optimization algorithm AO-AVOA is proposed, and the original position update rules are replaced by formulas (5) and (7) in the exploration phase of AVOA, as shown below: The exploration phase of AO is integrated with the development phase of AVOA, retaining their high efficiency in global and local search and improving the convergence speed.

9. The lithium battery SOH prediction method based on the AO-AVOA-BP neural network model according to claim 1, characterized in that: Step 3 is as follows: Use the African vulture fusion Eagle optimization algorithm to optimize the weights and thresholds of the BP neural network model, and establish an AO-AVOA-BP neural network battery SOH prediction model. Specifically, first, build a BP neural network model, determine the network topology, and initialize the BP neural network weights and thresholds; second, initialize the AO-AVOA algorithm parameters and population position, find the optimal and suboptimal vulture positions according to the fitness function, and update the entire population position; then, use the results as the optimal solution for the weights and thresholds, and input them into the BP neural network for continued training. When the conditions are met, the algorithm optimization process is completed; finally, use the trained AO-AVOA-BP neural network model to predict the SOH of the battery data.

10. The lithium battery SOH prediction method based on the AO-AVOA-BP neural network model according to claim 1, characterized in that: Step 4 is as follows: Using MATLAB as a platform, build models of BP neural network algorithm, PSO-BP neural network algorithm, AO-BP neural network algorithm, AVOA-BP neural network algorithm and AO-AVOA-BP neural network algorithm, use the four screened health factors as input, battery SOH as output, select the first 70% of the battery cycle data as a training set, and the remaining data as a test set to predict the battery SOH. In order to measure the prediction performance of the model, the mean absolute error MAE, root mean square error RMSE and mean absolute percentage error MAPE are selected as evaluation functions, and the calculation formula is as follows: Where N is the total number of battery cycles; y i is the actual value of the battery SOH, is the predicted value.

Images