Power battery pack SOH estimation method for optimizing NARX neural network based on raccoon optimization algorithm

By applying a prolonged raccoon optimization algorithm in the power battery pack to optimize the NARX neural network, the problem of slow convergence speed and easy to fall into local optimality in the power battery pack SOH estimation is solved, and the accuracy and efficiency of the estimation are improved. Achieve a more realistic battery health status reflection.

CN119025909APending Publication Date: 2024-11-26JIANGSU UNIV

Patent Information

Application Number
CN202410897762.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-05
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

Traditional neural networks have problems such as slow convergence speed, overfitting and easy to fall into local optimality in the SOH estimation of power battery packs, resulting in large errors in estimation values.

Method used

The NARX neural network is optimized based on the prolong-nosed raccoon optimization algorithm (COA). By collecting and processing the historical data of the power battery pack, the characteristic data set is constructed, and the COA algorithm is used for optimization processing, avoiding local optimal situations and improving convergence accuracy and speed .

Benefits of technology

It improves the accuracy and efficiency of SOH estimation of the power battery pack, can more realistically reflect the battery health status, reduce the estimation time, and avoid local optimal problems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119025909A_ABST
    Figure CN119025909A_ABST
Patent Text Reader

Abstract

The invention discloses a power battery pack SOH estimation method for optimizing an NARX neural network based on a raccoon optimization algorithm, and the method comprises the steps: carrying out the estimation of the SOH of a battery pack according to the peak-valley value of the current of the battery pack and the corresponding position, the peak-valley value of the temperature of the battery pack in summer / winter and the corresponding position, the voltage of the battery pack, the non-uniform position of the current of the battery pack and the corresponding value; constructing a first data set, and performing normalization processing on the first data set to obtain a second data set; the data of the battery pack in the second data set are regarded as a raccoon, a raccoon optimization algorithm is executed, an optimal solution is output, and a third data set is constructed based on the optimal solution; and training and testing the constructed NARX neural network by using the third data set, and during actual use, inputting BMS data of the power battery pack acquired in real time into the trained NARX neural network to obtain a prediction result of the SOH of the power battery pack. The SOH estimation accuracy of the power battery pack can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of power battery management, and in particular relates to a power battery pack SOH estimation method based on a coati optimization algorithm (COA) optimizing a NARX neural network. Background Art

[0002] With the popularity of new energy vehicles such as electric vehicles, the performance and safety testing and evaluation of power batteries as key core components have become crucial. As an indicator that describes the current health status of the battery, accurate estimation of SOH is of great significance for ensuring the power performance and safety performance of the entire electric vehicle. SOH is usually used to indicate the remaining life of the battery. As the number of battery cycles increases, the battery SOH will gradually decrease, the remaining life will decrease, and the available capacity will gradually decrease. Therefore, it is particularly important to accurately predict the battery capacity decay curve and effectively manage the battery health status.

[0003] At present, the main methods for evaluating the SOH of power batteries can be roughly divided into model-based methods and data-driven methods. Compared with model-based methods, the development of big data technology and artificial intelligence technology has made data-driven methods more promising, among which neural networks are widely used. However, when the neural network performs nonlinear transformation on the input characteristics, the correlation between the input characteristics and the SOH will be weakened, making the error of the later SOH estimation value larger. Secondly, traditional neural networks have problems in parameter selection, such as slow convergence, overfitting, and easy to fall into local optimality. Summary of the invention

[0004] In view of the shortcomings in the prior art, the present invention provides a power battery pack SOH estimation method based on the coati optimization algorithm (COA) to optimize the NARX neural network, in order to solve the problems of slow convergence speed, overfitting, and easy falling into local optimality in parameter selection of traditional neural networks, and improve the accuracy of SOH estimation of power battery packs.

[0005] The present invention achieves the above technical objectives through the following technical means.

[0006] Power battery pack SOH estimation method based on NARX neural network optimized by raccoon optimization algorithm:

[0007] Collect the BMS historical data of the power battery pack within two years, including the battery pack voltage, current and temperature;

[0008] A first data set is constructed from the peak and valley values ​​of the battery pack current and their corresponding positions, the peak and valley values ​​of the battery pack temperature in summer / winter and their corresponding positions, the battery pack voltage, and the uneven positions of the battery pack current and their corresponding values, and the first data set is normalized to obtain a second data set;

[0009] Treat each data of the battery group in the second data set as a raccoon, execute the raccoon optimization algorithm, output the best solution, and construct a third data set based on the best solution;

[0010] The constructed NARX neural network is trained and tested using the third data set. In actual use, the real-time collected BMS data of the power battery pack is input into the trained NARX neural network to obtain the prediction result of the SOH of the power battery pack.

[0011] A further technical solution is that the coati optimization algorithm includes the following process:

[0012] Initialize the parameters of the algorithm and characterize the position of the coati in the search and hunting space at the time of initialization;

[0013] The first stage: the coati is divided into the first half that climbs up the tree to scare the prey and the second half that waits under the tree for the prey to fall. The position of the coati in the first stage is characterized. If the position of the coati in the first stage improves the value of its fitness function, the position of the coati in the first stage is used to replace the initial position, otherwise the coati remains at the initial position.

[0014] The second stage: The coatis escapes from the predator. A random position is generated near the current position of each coatis. If the random position improves the value of its fitness function, the random position is used to replace the coatis's position in the first stage, that is, it reaches a safe position. Otherwise, the coatis remains at the position in the first stage.

[0015] A further technical solution, in order to avoid the local optimum, is to insert the nearest neighbor comparison algorithm before the first stage of the coati optimization, including:

[0016] (1) For each trial vector v, the nearest neighbor ζ n Use the distance shown in the following formula to explore:

[0017]

[0018] Where X is the sample vector to be tested, v is the test vector, d(X,v) is the distance between vectors X and v, Xmax and Xmin are the maximum and minimum values ​​of vector X, and D represents the problem dimension;

[0019] Nearest Neighbor n is the minimum distance between vectors X and v;

[0020] (2) Compare the nearest neighbors ζ n , the distance between the target solution and the test vector v, if the nearest neighbor ζ nIf the distance between the target solution and the test vector v is greater than the distance between the target solution and the test vector v, then the test vector v will skip the nearest neighbor ζ n , otherwise the test vector v uses the nearest neighbor ζ n .

[0021] According to a further technical solution, the peak and valley values ​​of the battery pack current and their corresponding positions represent the driving behavior characteristics of the vehicle.

[0022] According to a further technical solution, the peak and valley values ​​of the battery pack temperature in summer and winter and their corresponding positions represent weather characteristics.

[0023] According to a further technical solution, the battery pack voltage represents a conventional characteristic.

[0024] According to a further technical solution, the uneven positions of the battery pack current and their corresponding values ​​represent inconsistent characteristics within the battery pack.

[0025] The beneficial effects of the present invention are:

[0026] (1) The present invention takes the temperature of the power battery pack in summer and winter as weather characteristics, considers the impact of the electric vehicle battery pack on SOH in low and high temperature environments, and also takes the current unevenness as a feature input, considering the impact of the internal inconsistency of the battery pack on SOH; so that the estimated SOH can more realistically reflect the battery health status in the actual vehicle.

[0027] (2) The present invention uses the COA algorithm to optimize the characteristic data (second data set) of the power battery pack. The COA algorithm can select features with higher accuracy and improve convergence accuracy. The COA algorithm also has a faster convergence speed. The third data set is obtained by using the COA algorithm to train and test the NARX neural network. When used in practice, the SOH is estimated using the tested NARX neural network, which can reduce the estimation time.

[0028] (3) The present invention inserts the nearest neighbor comparison algorithm before the start of the first stage of the COA algorithm, which can avoid the algorithm from reaching a local optimum. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a flow chart of the method for estimating SOH of a power battery pack based on optimizing the NARX neural network using the coati optimization algorithm according to the present invention;

[0030] Figure 2 Schematic diagram of the NARX neural network described in the present invention. DETAILED DESCRIPTION

[0031] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments, but the protection scope of the present invention is not limited thereto.

[0032] like Figure 1 As shown, the present invention provides a power battery pack SOH estimation method based on the NARX neural network optimized by the coati optimization (COA) algorithm, which specifically includes the following steps:

[0033] S1. Collect the BMS historical data of the power battery pack within two years, including the battery pack voltage, current and temperature.

[0034] S2, the peak value of the battery current Valley value of battery pack current And the corresponding positions of the two and As input feature vector Represents the driving behavior characteristics of the vehicle; the peak temperature of the battery pack in summer / winter Valley And the corresponding positions of the two and As input feature vector Represents weather characteristics; battery pack voltage as a general characteristic The position of uneven current in the battery pack is recorded as The value corresponding to the uneven position of the battery pack current is recorded as As a feature vector Indicates the inconsistent characteristics within the battery pack; thereby obtaining the first data set

[0035] Normalize the first data set to obtain the second data set The second data set Divided into test set (20%) and training set (80%).

[0036] The first data set is normalized using min-max standardization to map the data to [0, 1], as shown in formula (1):

[0037]

[0038] Among them, x represents the data in the first data set, x min Indicates the minimum value of the data in the first data set, x max Indicates the maximum value of the data in the first data set.

[0039] S3, the second data set Run the COA algorithm, output the best solution, and construct a third data set based on the best solution

[0040] The second data set The battery pack data is viewed as a raccoon and the COA algorithm is executed:

[0041] (1) Initialize various parameters in the COA algorithm, including: initializing the population size N, the maximum number of iterations T, the problem dimension D, the upper bound lb and lower bound ub of the search hunting space boundary, define the current number of iterations as t, and initialize t = i = 1.

[0042] At initialization, the position of the raccoon in the search hunting space is expressed using formula (2):

[0043] X i :x i,j =lb j +r·(ub j -lb j ), i=1,2,…,N, j=1,2,…,m(2)

[0044] Among them, X i is the position of the ith coati in the search hunting space, x i,j is the value of the jth decision variable for the ith coatis, N is the total number of coatis, m is the total number of decision variables, r is a random real number in the interval [0, 1], and lb j ,ub j are the lower and upper bounds of the j-th decision variable, respectively.

[0045] The position of the coati population in the search hunting space is represented by the following matrix X, called the population matrix:

[0046]

[0047] (2) Entering the first stage: hunting and attacking strategies against iguanas (exploration stage).

[0048] During the exploration phase, coatis are divided into two groups: the first half climbs up the tree to scare the prey, and the second half waits under the tree for the frightened prey to fall; following this movement pattern, coatis can thoroughly search the hunting space. The coatis' prey is called iguana, and the best position for a member of the coati population is considered to be the position of the iguana. The position of the coatis in the tree is shown in equation (4). When the iguana falls to the ground, the position of the iguana is randomized according to equation (5). The coatis waiting under the tree move according to equation (6).

[0049]

[0050] in, represents the position of the i-th coati on the tree in the first stage, represents the jth dimension of the i-th coati in the first stage tree, x iu,jis the value of the jth decision variable for the i-th coati in the tree, Iguana j represents the j-th dimension of the iguana, and I represents an integer randomly selected from the set {1, 2};

[0051]

[0052] Among them, Iguana G Represents an iguana on the ground, Iguana G The jth dimension of

[0053]

[0054] in, Iguana G The fitness function value, F i Indicates the initial position X of the coati i The fitness function value of represents the position of the i-th coati under the tree in the first stage, represents the jth dimension of the i-th coati under the first stage tree, x id,j is the value of the jth decision variable for the i-th coati under the tree.

[0055] If the position of the coati in the first stage improves the value of the fitness function, that is, The position change in the first stage is considered to be valid, and the position of the coati in the first stage is used to replace the initialization position. Otherwise, the coati is still at the initialization position, which is specifically expressed by formula (7):

[0056]

[0057] in, yes The fitness function value of include and

[0058] (3) Entering the second stage: the process of escaping from predators (development stage).

[0059] The second stage of development is to describe the natural behavior of coatis when encountering and escaping from predators. When a predator attacks a coati, it will flee from its position. The coati’s actions in this strategy lead it to a safe position close to its current position (the position in the first stage). To simulate this behavior, random positions are generated near the current position of each coati according to equations (8) and (9).

[0060]

[0061]

[0062] in, and are the local upper and lower bounds of the j-th decision variable, respectively. represents the position of the i-th coati in the second stage, represents the jth dimension of the i-th coati in the second stage;

[0063] If the random position improves the value of the fitness function, that is, The position change in the second stage is considered to be valid, and the position of the coati in the first stage is replaced by a random position (i.e., it reaches a safe position). Otherwise, the coati is still at the position in the first stage, as shown in formula (10).

[0064]

[0065] in, yes The fitness function value, F′ i is the first stage position X′ i The fitness function of .

[0066] The fitness functions of the first and second phases are obtained as follows:

[0067]

[0068] Among them, F i Indicates F i1 or

[0069] The value of the fitness function is obtained by the following formula:

[0070]

[0071] Where, constant λ∈[0,1], |ξ| represents The number of features in (the fitness function value The |ξ| in the formula represents Iguana G The number of features in the second dataset), |ξ1| represents the total number of features in the second dataset, and γ s (D) represents the error rate of the classification process.

[0072] In order to avoid the local optimum, the nearest neighbor comparison algorithm (NNC) is inserted before the first stage of the COA algorithm. NNC is divided into two steps:

[0073] In the first step, for each trial vector v, the nearest neighbor ζn Use the distance shown in equation (13) to explore:

[0074]

[0075] Among them, X is the sample vector to be tested, v is the test vector, d(X,v) is the distance between vectors X and v, Xmax and Xmin are the maximum and minimum values ​​of vector X; the nearest neighbor ζ n is the minimum distance between vectors X and v;

[0076] The second step is to compare the nearest neighbors ζ n , the distance between the target solution and the test vector v, if the nearest neighbor ζ n If the distance from the trial vector v is greater than the target solution, then the trial vector v will skip the nearest neighbor ζ n , otherwise the test vector v uses the nearest neighbor ζ n .

[0077] After the second stage, after T iterations, the position of the coati obtained is the best solution, and the best solution is used as the third data set.

[0078] S4, dividing the third data set into a training set (80%) and a test set (20%) for training and testing the constructed NARX neural network;

[0079] In actual use, the BMS data of the power battery pack collected in real time is input into the trained NARX neural network, and the SOH of the power battery pack is used as the output of the NRAX neural network to obtain the prediction result of the SOH of the power battery pack.

[0080] The output SOH predicted by the NARX neural network is expressed by formula (14):

[0081] SOH=F(y(n),y(n-1),…,y(nL M ),x(n),x(n-1),…,x(nL N )) (14)

[0082] The predicted output SOH is regressed based on the current and previous values ​​of the output delay layer data y(n) and the input delay layer data x(n) (i.e., BMS real-time data), L N and L M denote the order of input delay and output delay respectively, and F(·) is a nonlinear function.

[0083] For a NARX neural network with N hidden layers, the output y(n) can be expressed by formula (15):

[0084]

[0085] Among them, G(·) and S(·) are the activation functions of the output layer and the hidden layer respectively, b o and W o is the bias and weight vector of the output layer, b h and W h are the bias and weight of the hidden layer, W uh Represents the weight vector from the input layer to the hidden layer, W vh Represents the weight vector from the output layer to the hidden layer.

[0086] The embodiments are preferred implementations of the present invention, but the present invention is not limited to the above-mentioned implementations. Any obvious improvements, substitutions or modifications that can be made by those skilled in the art without departing from the essential content of the present invention belong to the protection scope of the present invention.

Claims

1. A power battery pack SOH estimation method based on the NARX neural network optimized by the coati optimization algorithm, characterized in that: Collect the BMS historical data of the power battery pack within two years, including the battery pack voltage, current and temperature; A first data set is constructed from the peak and valley values ​​of the battery pack current and their corresponding positions, the peak and valley values ​​of the battery pack temperature in summer / winter and their corresponding positions, the battery pack voltage, and the uneven positions of the battery pack current and their corresponding values, and the first data set is normalized to obtain a second data set; Treat each data of the battery group in the second data set as a raccoon, execute the raccoon optimization algorithm, output the best solution, and construct a third data set based on the best solution; The constructed NARX neural network is trained and tested using the third data set. In actual use, the real-time collected BMS data of the power battery pack is input into the trained NARX neural network to obtain the prediction result of the SOH of the power battery pack.

2. The power battery pack SOH estimation method according to claim 1, characterized in that: The coati optimization algorithm includes the following steps: Initialize the parameters of the algorithm and characterize the position of the coati in the search and hunting space at the time of initialization; The first stage: the coati is divided into the first half that climbs up the tree to scare the prey and the second half that waits under the tree for the prey to fall. The position of the coati in the first stage is characterized. If the position of the coati in the first stage improves the value of its fitness function, the position of the coati in the first stage is used to replace the initial position, otherwise the coati remains at the initial position. The second stage: The coatis escapes from the predator. A random position is generated near the current position of each coatis. If the random position improves the value of its fitness function, the random position is used to replace the coatis's position in the first stage, that is, it reaches a safe position. Otherwise, the coatis remains at the position in the first stage.

3. The power battery pack SOH estimation method according to claim 2, characterized in that: In order to avoid the local optimum, the nearest neighbor comparison algorithm is inserted before the first stage of the coati optimization algorithm, including: (1) For each trial vector v, the nearest neighbor ζ n Use the distance shown in the following formula to explore: Where X is the sample vector to be tested, v is the test vector, d(X,v) is the distance between vectors X and v, Xmax and Xmin are the maximum and minimum values ​​of vector X, and D represents the problem dimension; Nearest Neighbor n is the minimum distance between vectors X and v; (2) Compare the nearest neighbors ζ n , the distance between the target solution and the test vector v, if the nearest neighbor ζ n If the distance between the target solution and the test vector v is greater than the distance between the target solution and the test vector v, then the test vector v will skip the nearest neighbor ζ n , otherwise the test vector v uses the nearest neighbor ζ n .

4. The power battery pack SOH estimation method according to claim 1, characterized in that: The peak and valley values ​​of the battery pack current and their corresponding positions represent vehicle driving behavior characteristics.

5. The power battery pack SOH estimation method according to claim 1, characterized in that: The peak and valley values ​​of the battery pack temperature in summer / winter and their corresponding positions represent weather characteristics.

6. The power battery pack SOH estimation method according to claim 1, characterized in that: The battery pack voltages described represent conventional characteristics.

7. The power battery pack SOH estimation method according to claim 1, characterized in that: The battery pack current uneven positions and their corresponding values ​​represent inconsistent features within the battery pack.

Citation Information

Patent Citations

  • SOH time sequence prediction method for vehicle lithium ion power battery

    CN113901707A

Cited By

  • Distributed heterogeneous factory cell-PACK cooperative scheduling method and system

    CN120430564A