Characteristic dimensionality reduction lithium ion battery state estimation method and device and storage medium

By extracting the constant current and constant voltage charging curve of lithium-ion batteries and building a KPCA-GRU model, the problems of complex calculations and inaccurate predictions in the health status estimation of lithium-ion batteries are solved, and more efficient estimation of the health status of lithium-ion batteries are achieved.

CN119064785APending Publication Date: 2024-12-03CHINA UNIV OF GEOSCIENCES (WUHAN)
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202411074967.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-07
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

The prior art has complex calculations and inaccurate prediction results in the health status estimation of lithium-ion batteries, especially the nonlinear aging characteristics cannot be effectively extracted, resulting in difficulty in dimensionality reduction optimization.

Method used

By obtaining the constant current and constant voltage charging curve of lithium-ion batteries, aging characteristics are extracted and Pearson correlation coefficient is calculated, and the core principal component analysis (KPCA) and gated cycle unit (GRU) models are constructed to perform feature matrix dimensionality reduction and health status estimation.

Benefits of technology

It improves the efficiency and accuracy of estimation of health status of lithium-ion batteries, can effectively extract nonlinear features, reduce the computational burden, and improve the model training efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119064785A_ABST
    Figure CN119064785A_ABST
Patent Text Reader

Abstract

The invention relates to the field of battery state monitoring, and discloses a lithium ion battery state estimation method and device based on feature dimensionality reduction and a storage medium, and the method comprises the steps: obtaining a charging curve of a constant-current and constant-voltage stage in each charging and discharging cycle of a lithium ion battery; obtaining multiple lithium ion battery aging characteristics according to the charging curve; calculating a Pearson's correlation coefficient between each aging characteristic and the maximum capacity of the lithium ion battery to measure the correlation between the aging characteristics and the maximum capacity of the lithium ion battery, selecting the aging characteristics with high correlation coefficients to form a characteristic matrix, and normalizing the characteristic matrix; performing dimension reduction processing on the feature matrix by using kernel principal component analysis; and constructing a GRU network model, training the GRU model, and estimating the SOH of the lithium ion battery through the trained model. The lithium ion battery health state estimation method can effectively realize accurate lithium ion battery health state estimation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of battery status monitoring, and in particular to a method, device and storage medium for estimating the status of a lithium-ion battery with feature dimension reduction. Background Art

[0002] Lithium-ion batteries have the advantages of small size, long life, high efficiency, high power density, fast dynamic response and environmental friendliness, and have become an important component of electric vehicles, grid energy storage systems and other portable facilities. Battery capacity will decrease with the increase of aging processes such as normal charging and discharging, abnormal overcharging and over-discharging. Battery degradation will affect the normal operation of the battery system. If severely aged lithium batteries are not replaced in time, they may even cause safety accidents such as explosions.

[0003] The aging and performance degradation of lithium-ion batteries are one of the causes of lithium-ion battery safety incidents. Battery degradation can be reflected by changes in capacity, resistance, and voltage, among which the state of health (SOH) is used in many cases to characterize the capacity degradation of lithium-ion batteries. However, the capacity degradation of lithium-ion batteries is a long and slow process. Online and accurate evaluation of the SOH value of lithium-ion batteries is of great significance to the operation and maintenance of battery systems.

[0004] Lithium-ion battery SOH estimation methods are mainly divided into two categories: model-based methods and data-driven methods. Model-based methods have high accuracy, but due to the complex internal reactions and many side reactions of lithium-ion batteries, it is very difficult to model and identify model parameters. In addition, the model is based on lithium-ion batteries of specific models and structures, and has poor mobility. Data-driven methods are faster and involve fewer complex mechanical factors. The data-driven model is like a "black box" that extracts the mapping relationship between features and SOH from the aging data of lithium-ion batteries (such as voltage, current, and time). In feature extraction, most researchers study the aging process based on the discharge process of lithium-ion batteries. However, the discharge process of lithium-ion batteries in electric vehicles mainly depends on the application environment and working conditions, which has great uncertainty and is not conducive to extracting corresponding aging features. After feature extraction, it is usually necessary to degrade the dimension of the feature matrix to improve the training efficiency of the model. Considering that some aging features are correlated, in order to reduce the dimension of the aging features and reduce the computational burden, the extracted aging features need to be optimized by dimensionality reduction. Among them, principal component analysis (PCA) is an effective method. A small number of principal components can be used to reflect the original sample with complex information as much as possible. Principal component analysis can effectively reduce the dimension of the sample space, which is conducive to studying the correlation between variables. However, PCA can only extract the linear features of the data and cannot better reduce the dimension of the aging features. Therefore, it is urgent to design a hybrid algorithm that can extract aging features from the charging curve and extract nonlinear features from the aging features to achieve better dimensionality reduction optimization. Summary of the invention

[0005] The purpose of the present invention is to propose a lithium-ion battery state estimation method with feature dimensionality reduction to solve the technical problems of complex calculation and inaccurate prediction results in the prior art during lithium-ion battery health estimation.

[0006] A lithium-ion battery state estimation method based on feature dimension reduction comprises the following steps:

[0007] S1, obtaining a charging curve of a lithium-ion battery in a constant current and constant voltage stage in each charge and discharge cycle;

[0008] S2, obtaining a plurality of lithium-ion battery aging characteristics F1-F10 according to the charging curve;

[0009] S3, calculating the Pearson correlation coefficient between each aging feature and the maximum capacity of the lithium-ion battery to measure the correlation between the two, selecting aging features with high correlation coefficients to form a feature matrix, and normalizing the feature matrix;

[0010] S4, performing dimensionality reduction processing on the feature matrix using kernel principal component analysis;

[0011] S5. Construct a gated recurrent unit (GRU) network model, use the feature matrix obtained by dimensionality reduction as input, train the GRU model, complete the training process, and obtain a trained GRU model; finally, estimate the SOH of the lithium-ion battery through the trained GRU model.

[0012] A storage medium stores instructions and data for implementing a lithium-ion battery state estimation method using feature dimension reduction.

[0013] A lithium-ion battery state estimation device with feature dimensionality reduction comprises: a processor and the storage medium; the processor loads and executes instructions and data in the storage medium to implement a lithium-ion battery state estimation method with feature dimensionality reduction.

[0014] The beneficial effects provided by the present invention are: on the one hand, the method can be more conducive to extracting the aging information of the lithium-ion battery by extracting the aging characteristics of the lithium-ion battery in the constant-current and constant-voltage charging stage, and thus is conducive to realizing the health state estimation of the lithium-ion battery; on the other hand, the method realizes the dimensionality reduction optimization of the lithium-ion battery feature matrix through the KPCA algorithm, solves the problem that the nonlinear characteristics cannot be extracted from the aging characteristics of the lithium-ion battery, which leads to the inability to realize better dimensionality reduction optimization, and can greatly improve the efficiency of the health state estimation of the lithium-ion battery. Through the above steps, the health state estimation of the lithium-ion battery is effectively realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic flow chart of the method of the present invention;

[0016] Figure 2 It is a schematic diagram of constant current and constant voltage charging current curve;

[0017] Figure 3 It is a schematic diagram of constant current and constant voltage charging voltage curve;

[0018] Figure 4 It is a schematic diagram of the capacity increment IC curve;

[0019] Figure 5 It is a schematic diagram of the calculation results of Pearson correlation coefficient;

[0020] Figure 6 This is a schematic diagram of the results of the GRU model on the test set;

[0021] Figure 7 It is a working schematic diagram of the hardware device of the present invention. DETAILED DESCRIPTION

[0022] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0023] Before formally describing the present invention, the scheme of the present invention is first generally described for easy understanding.

[0024] Please refer to Figure 1 , Figure 1 It is a schematic flow diagram of the method of the present invention;

[0025] The present invention provides a lithium-ion battery state estimation method based on feature dimension reduction, comprising the following steps:

[0026] S1, obtaining a charging curve of a lithium-ion battery in a constant current and constant voltage stage in each charge and discharge cycle;

[0027] It should be noted that step S1 is specifically as follows:

[0028] Step 1.1: First, obtain the lithium-ion battery degradation data D B ={(I t ,V t ,C t ,Q) m} n , where t is the sampling time, and sampling is performed at equal time intervals, t∈(0,T C ) is the charging process, t∈(T C ,T D ) is the discharge process; m is the cycle number, m = 1, 2, ..., M; n is the lithium-ion battery serial number, n = 0, 1, 2, ..., N; I, V, C represent the measured current, voltage and battery cumulative charge and discharge capacity of the n-th lithium-ion battery in the m-th cycle respectively; Q is the maximum cumulative charge or cumulative discharge capacity of the n-th lithium-ion battery in the m-th cycle, and the calculation formula is or

[0029] Most aging features are extracted based on the discharge process, but the discharge process of lithium-ion batteries in electric vehicles mainly depends on the application environment and working conditions, which has great uncertainty and is not conducive to extracting corresponding aging features.

[0030] The present invention mines the aging information of the lithium-ion battery by extracting the constant-current and constant-voltage charging data of the lithium-ion battery.

[0031] Step 1.2: Degrade the data D B Obtain the charging current curve, charging voltage curve, and charging time curve in the constant current and constant voltage stage. Further obtain the IC curve through the constant current charging current curve, constant current charging time curve, and constant current charging voltage curve. The calculation formula of the IC curve is as follows:

[0032] Q=I·t

[0033]

[0034] Where Q is the battery capacity, t is the constant current charging time, I is the constant current charging current, and V is the constant current charging voltage. The original IC curve obtained has large noise, and the Savitzky-Golay filter is selected to filter out the noise interference. The calculation formula of the Savitzky-Golay filter is:

[0035]

[0036] Among them, b k are the polynomial coefficients, i is the obtained IC curve, and y(n) is the filtered IC curve.

[0037] S2, obtaining a plurality of lithium-ion battery aging characteristics F1-F10 according to the charging curve;

[0038] It should be noted that step S2 is specifically as follows:

[0039] Step 2.1, extracting time-related features: constant current stage charging time F1, constant voltage stage charging time F2, constant current and constant voltage stage total charging time F3, constant current stage charging time to total charging time ratio F4;

[0040] Step 2.2, extract current related features: the area of ​​the current curve and the time axis in the constant current charging stage F5, the area of ​​the current curve and the time axis in the constant voltage charging stage F6, and the area of ​​the current curve and the time axis in the constant current and constant voltage charging stage F7;

[0041] Step 2.3, extract voltage-related features: the difference between the maximum and minimum charging voltage in the constant current stage F8;

[0042] Step 2.4, extract the relevant features of the IC curve: the maximum peak value F9 of the IC curve, and the voltage value F10 corresponding to the maximum peak point of the IC curve.

[0043] S3, calculating the Pearson correlation coefficient between each aging feature and the maximum capacity of the lithium-ion battery to measure the correlation between the two, selecting aging features with high correlation coefficients to form a feature matrix, and normalizing the feature matrix;

[0044] It should be noted that step S3 is specifically as follows:

[0045] Step 3.1: Calculate the Pearson correlation coefficient (PCC) between the aging characteristics (F1-F10) obtained in S2 and the maximum capacity to measure the correlation between the two. The Pearson correlation coefficient calculation formula is:

[0046]

[0047] Among them, zi is the characteristic quantity, is the average value of the characteristic quantity, q i is the maximum capacity, is the average value of the maximum capacity.

[0048] Step 3.2: Select features with an absolute value of the Pearson correlation coefficient greater than 0.9 to form a feature matrix, and use the maximum and minimum normalization method for the feature matrix, as shown in the following formula:

[0049]

[0050] Among them, F Sim is the degradation data of lithium-ion batteries, F Sim The maximum and minimum values ​​in are denoted as and is the normalized data.

[0051] S4, performing dimensionality reduction processing on the feature matrix using kernel principal component analysis;

[0052] Step 4.1. Compare the correlation coefficient and partial correlation coefficient between features through the KMO test statistic, which is used to test whether the sample can be optimized for dimensionality reduction. The value of KMO is between 0 and 1. The larger the value, the higher the correlation between the features, and the more suitable the feature matrix is ​​for dimensionality reduction optimization. Generally, this value is required to be greater than 0.5 before the feature matrix can be optimized for dimensionality reduction. The principal component analysis method is usually used to reduce the dimensionality of the feature matrix, but the principal component analysis method can only extract linear features in the data and cannot achieve better optimization of the feature matrix. Therefore, based on the principal component analysis, a kernel principal component analysis (KPCA) method is proposed to extract nonlinear features in the data. The KPCA algorithm process is as follows:

[0053] 1. The present invention selects Gaussian kernel function as the kernel function of KPCA, and the formula of Gaussian kernel function is:

[0054] k*x i ,x j )=exp(-γ∥∥x i -x j ∥∥ 2 )

[0055] Among them, x i and x j is the aging feature, and γ is the hyperparameter of the Gaussian kernel function.

[0056] 2. Get the Gaussian kernel matrix K of the feature matrix through the Gaussian kernel function:

[0057]

[0058] 3. Centralize the Gaussian kernel matrix. The formula for centralization is:

[0059] K ′ =K-1 n K-K1 n +1 n K1 n

[0060] Among them, K ′ is the decentralized Gaussian kernel matrix, K is the Gaussian kernel matrix, 1 n is an n-dimensional matrix, in which each element is equal to 1 / n;

[0061] 4. Calculate K ′ The eigenvalue λ k With the eigenvector χ k ;

[0062] 5. Calculate the contribution and cumulative contribution of the characteristic value. The contribution rate calculation formula is as follows:

[0063]

[0064] Among them, λ k is the kth eigenvalue, is the contribution rate of the kth eigenvalue.

[0065] The cumulative contribution rate is calculated as follows:

[0066]

[0067] Among them, m Represents the cumulative contribution rate corresponding to the mth eigenvalue.

[0068] 6. Take the eigenvectors corresponding to the first k eigenvalues ​​with a cumulative contribution rate of 80% as the principal components to construct a new feature matrix.

[0069] S5. Construct a GRU network model, use the feature matrix obtained by dimensionality reduction as input, train the GRU model, complete the training process, and obtain a trained GRU model; finally, estimate the SOH of the lithium-ion battery through the trained GRU model.

[0070] S5. Construct a gated recurrent unit (GRU) network model, use the feature matrix obtained by dimensionality reduction as input, train the GRU model, complete the training process, and obtain a trained GRU model; finally, estimate the SOH of the lithium-ion battery through the trained GRU model.

[0071] It should be noted that step S5 is specifically as follows:

[0072] Step 5.1, select three lithium-ion batteries with the same working conditions, of which A and B are used for model training, and the other lithium-ion battery C is used for model testing; battery A is used as the training set, battery B is used as the validation set, and battery C is used as the test set;

[0073] Step 5.2: Take the feature matrix obtained by dimensionality reduction of battery A as input to train the GRU model. The input layer of the model is set to 4, the number of GRU layers is set to 2, and the hidden layer is set to 60. To improve the accuracy of the model, the Adam optimizer is used to optimize the model parameters, and the learning rate is set to 0.001 to implement model training. The trained GRU model is verified by battery B to obtain the trained GRU model; finally, the trained GRU model is tested by battery C to complete the SOH estimation of lithium-ion batteries.

[0074] As an embodiment, the present invention first selects three lithium-ion batteries with the same working conditions for training and testing of the GRU model, wherein batteries A and B are used for model training, and battery C is used for model testing; wherein battery A is used as a training set, battery B is used as a validation set, and battery C is used as a test set. For the feature extraction link of lithium-ion batteries, firstly, the curve of the constant current and constant voltage charging stage is obtained through the aging data of the lithium-ion battery, and then the aging characteristics of the lithium-ion battery are extracted through the obtained curve of the constant current and constant voltage charging stage. Subsequently, the Pearson correlation coefficient between the aging characteristics and the maximum capacity is calculated to measure the correlation between the two, and the aging characteristics with high correlation coefficients are selected to form a feature matrix. The KPCA algorithm is used to reduce the dimension of the extracted feature matrix to improve the model training efficiency. Finally, a GRU model is constructed, and the feature matrix obtained by dimensionality reduction is used as input to train the GRU model, complete the training process, and obtain a trained GRU model; finally, the SOH of the lithium-ion battery is estimated by the trained GRU model. The data for this case comes from the MIT public dataset, which contains 124 Panasonic APR18650M1A lithium ion phosphate (LFP) / graphite batteries with a nominal capacity of 1.1Ah and a nominal voltage of 3.3V.

[0075] The specific steps are as follows:

[0076] (1) Extracting constant current and constant voltage charging stage curves and aging characteristics from MIT data

[0077] Based on the above MIT public data set, three lithium-ion batteries, b1c8, b1c10 and b1c22, are selected. b1c8 and b1c10 are used for the training process, and b1c22 is used for the testing process. Among them, b1c10 is used as the training set, b1c8 is used as the validation set, and b1c22 is used as the test set.

[0078] Select the constant current and constant voltage charging data of 80% SOC ~ 100% SOC, and the constant current charging rate is 1C. The charging current curve of b1c No. 10 lithium-ion battery is as follows Figure 2 As shown in the figure, with the increase of the number of charge and discharge cycles, the constant current charging time and the battery full charge time will be shortened, and the area between the current curve and the horizontal axis will gradually decrease, which indicates that with the increase of the number of cycles, the performance of the battery begins to decline. Therefore, the charging time and curve area that reflect the maximum capacity change are selected as features, namely, the constant current stage charging time (F1), the constant voltage stage charging time (F2), the total charging time in the constant current and constant voltage stage (F3), the ratio of the constant current stage charging time to the total charging time (F4), the area of ​​the current curve and the time axis in the constant current charging stage (F5), the area of ​​the current curve and the time axis in the constant voltage charging stage (F6), and the area of ​​the current curve and the time axis in the constant current and constant voltage charging stage (F7); the charging voltage curve of the b1c10 lithium-ion battery is shown in the figure Figure 3 As shown in the figure, with the increase of the number of charge and discharge cycles, the charging voltage in the constant current stage will gradually increase, and the battery performance will gradually decrease. Therefore, the difference between the starting voltage during constant current charging and the voltage at the end of the constant current stage can be selected to reflect the change in the maximum capacity of the battery, that is, the difference between the maximum and minimum charging voltage in the constant current stage (F8); b1c10 lithium-ion battery IC curve is as follows Figure 4 As shown in the figure, during the battery charging process, the capacity and voltage can reflect the battery attenuation. The IC curve can be used to describe the change in battery performance. As the number of cycles increases, the IC curve gradually becomes flatter, the maximum peak of the IC curve gradually decreases, and the position of the maximum peak gradually moves to the right. Therefore, these two values ​​can be used as characteristics to describe battery aging, namely the maximum peak of the IC curve (F9) and the voltage value corresponding to the maximum peak point of the IC curve (F10).

[0079] (2) Correlation between maximum capacity and aging characteristics

[0080] The Pearson correlation coefficient between the aging feature and the maximum capacity is calculated to measure the correlation between the two, and the aging features with high correlation coefficients are selected to form a feature matrix. The calculation formula is as follows:

[0081]

[0082] Among them, z i is the characteristic quantity, is the average value of the characteristic quantity, q i is the maximum capacity, is the average value of the maximum capacity. The value range of this coefficient is [-1,1]. When PCC is 1 or -1, it means that there is a complete positive or negative correlation between the two variables. 0 means no correlation; the PCC calculation results of lithium-ion batteries are as follows Figure 5As shown, it can be shown that most of the features can well reflect the aging mechanism of lithium-ion batteries. The present invention selects 8 aging features (F1, F3, F4, F5, F7, F8, F9, F10, respectively represented by S1-S8) of the b1c10 battery with an absolute value of the Pearson correlation coefficient exceeding 0.9 to form a feature matrix, and then normalizes the feature matrix using the maximum and minimum normalization method.

[0083] (3) Reduce the dimension of the feature matrix based on the KPCA algorithm

[0084] After obtaining the feature matrix, the correlation coefficient and partial correlation coefficient between the features are first compared through the KMO test statistic, and the value is generally required to be greater than 0.5. The results of the KMO test statistic for the training set b1c No. 10 battery are shown in Table 1. This shows that the extracted feature matrix is ​​suitable for dimensionality reduction processing.

[0085] Table 1b1c10 KMO test statistics of each feature

[0086]

[0087] The feature matrix is ​​reduced in dimension by the KCPA algorithm, and its Gaussian kernel function is:

[0088] k(x i ,x j )=exp(-γ∥∥x i -x j ∥∥ 2 )

[0089] The Gaussian kernel matrix of the feature matrix is ​​obtained through the Gaussian kernel function:

[0090]

[0091] The Gaussian kernel matrix is ​​centered, and the eigenvalues ​​(λ1~λ8) and eigenvectors of the Gaussian kernel matrix after the centering process are calculated. The cumulative contribution rate is calculated as follows:

[0092]

[0093] The cumulative contribution rate of the characteristic values ​​of the b1c10 battery in the training set is calculated by the above method, and the results are shown in Table 2:

[0094] Table 2b1c10 Cumulative contribution rate of each eigenvalue

[0095] Battery number\Cumulative contribution rate <![CDATA[λ1]]> <![CDATA[λ2]]> <![CDATA[λ3]]> <![CDATA[λ4]]> <![CDATA[λ5]]> <![CDATA[λ6]]> <![CDATA[λ7]]> <![CDATA[λ8 <!-- 6 -->]]> b1c10 0.33 0.51 0.67 0.83 0.97 0.99 1.0 1.0

[0096] The eigenvectors corresponding to the first four eigenvalues ​​with cumulative contribution rates of 80% are taken as principal components to construct a new feature matrix.

[0097] (4) Offline training and online testing of GRU model

[0098] Train the GRU model, set the model's input layer to 4, the number of GRU layers to 2, and the hidden layer to 60. To improve the accuracy of the model, use the Adam optimizer to optimize the model parameters, set the learning rate to 0.001, implement the model training, and save the model after training. b1c8 and b1c10 are used for training, b1c10 is used as the training set, and b1c8 is used as the validation set.

[0099] In addition, the capacity degradation data also needs to be normalized. The maximum and minimum normalization method is used based on the training set lithium-ion battery data F Sim The maximum and minimum values ​​are denoted as and Then the maximum and minimum normalization formula is used for normalization. In the model testing phase, in order to ensure the performance of the model on the test set data, the maximum and minimum values ​​in the normalization of the test set capacity degradation data are still selected. and

[0100] After obtaining the trained model, the trained model was tested by using a b1c22 lithium-ion battery. To further analyze the results numerically, the model training time (Time), root mean square error (RMSE), mean absolute error (MAE), and SOH estimation coefficient of determination (R 2 ) is used to measure the performance of the algorithm. The calculation formula is as follows:

[0101]

[0102]

[0103] Where n is the sample size of all data used in the evaluation; y i and represent the actual data value and the estimated data value respectively; Represents the average of the actual values ​​of the data.

[0104] In order to verify that the efficiency of the proposed method in model training can be greatly improved, the above indicators are calculated, and the estimation results of the KPCA-GRU algorithm and the other five algorithms are compared as shown in Table 3. The loss value of the KPCA-GRU model training is used as the threshold, and the training of other models is stopped when the threshold is reached. It can be seen from Table 3 that the time based on the KPCA-GRU algorithm is 0.23s. Compared with the LSTM and GRU models, although the prediction accuracy of the proposed method has slight fluctuations, the training time is greatly reduced, and the training efficiency is improved by 98.68% and 98.34% respectively. Compared with other feature dimensionality reduction models, such as PCA-LSTM and PCA-GRU, the proposed method has improved both training efficiency and accuracy, with training efficiency increased by 87.90% and 87.63% respectively, and accuracy increased by 0.16% and 0.28%. In addition, the present invention also compares the proposed method with KPCA-LSTM, and the training time is shortened by 61.67%, while the accuracy is improved by 1.01%, which demonstrates the rationality of selecting KPCA-GRU as the health status estimation model for lithium-ion batteries.

[0105] Table 3b1c22 Comparison of estimation results based on KPCA-GRU algorithm and other five algorithms

[0106] algorithm Time(s) RMSE MAE <![CDATA[R 2 (%)]]> LSTM 17.41 0.0014 0.0018 99.51 PCA-LSTM 1.90 0.0024 0.0020 99.29 KPCA-LSTM 0.60 0.0034 0.0030 98.46 GRU 13.87 0.0019 0.0015 99.50 PCA-GRU 1.86 0.0026 0.0021 99.17 KPCA-GRU 0.23 0.0021 0.0016 99.45

[0107] To verify that the proposed method can greatly improve its accuracy under the same training conditions (epoch=100), the results are as follows Figure 6 As shown. The above indicators are calculated and the results are shown in Table 4. It can be seen from the data in the table that the RMSE when the KPCA-GRU algorithm is used to estimate SOH is 0.0021. Compared with the GRU model, its RMSE has decreased by 94.17%; compared with the LSTM model, its RMSE has decreased by 97.27%; compared with the PCA-GRU model, its RMSE has decreased by 43.24%; compared with PCA-LSTM, its RMSE has decreased by 93.18%; compared with KPCA-LSTM, its RMSE has decreased by 93.29%. It proves that the KPCA-GRU algorithm performs well in the problem of estimating the health status of lithium-ion batteries and can be applied to actual lithium-ion battery management systems.

[0108] Table 4 Comparison of the estimation results of b1c22 based on KPCA-GRU algorithm and other five algorithms under the same training conditions

[0109] algorithm RMSE MAE <![CDATA[R 2 (%)]]> LSTM 0.0769 0.0533 40.66 PCA-LSTM 0.0308 0.0305 / KPCA-LSTM 0.0313 0.0312 / GRU 0.0360 0.0283 66.85 PCA-GRU 0.0037 0.0031 98.29 KPCA-GRU 0.0021 0.0016 99.45

[0110] See also Figure 7 , Figure 74 is a schematic diagram of the working of the hardware device of an embodiment of the present invention, wherein the hardware device specifically comprises: a lithium-ion battery state estimation device 401 with feature dimension reduction, a processor 402 and a storage medium 403.

[0111] A device 401 for estimating a lithium-ion battery state with feature dimensionality reduction: The device 401 for estimating a lithium-ion battery state with feature dimensionality reduction implements the method for estimating a lithium-ion battery state with feature dimensionality reduction.

[0112] Processor 402: The processor 402 loads and executes the instructions and data in the storage medium 403 to implement the lithium-ion battery state estimation method with feature dimension reduction.

[0113] Storage medium 403: the storage medium 403 stores instructions and data; the storage medium 403 is used to implement the lithium-ion battery state estimation method with feature dimension reduction.

[0114] The beneficial effects of the present invention are as follows: on the one hand, the method can be more conducive to extracting the aging information of the lithium-ion battery by extracting the aging characteristics of the lithium-ion battery in the constant-current and constant-voltage charging stage, and thus is conducive to realizing the health state estimation of the lithium-ion battery; on the other hand, the method realizes the dimensionality reduction optimization of the lithium-ion battery feature matrix through the KPCA algorithm, solves the problem that the nonlinear characteristics cannot be extracted from the aging characteristics of the lithium-ion battery, which leads to the inability to realize better dimensionality reduction optimization, and can greatly improve the efficiency of the health state estimation of the lithium-ion battery. Through the above steps, the health state estimation of the lithium-ion battery is effectively realized.

[0115] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A lithium-ion battery state estimation method based on feature dimension reduction, characterized in that: The following steps are involved: S1, obtaining a charging curve of a lithium-ion battery in a constant current and constant voltage stage in each charge and discharge cycle; S2, obtaining a plurality of lithium-ion battery aging characteristics F1-F10 according to the charging curve; S3, calculating the Pearson correlation coefficient between each aging feature and the maximum capacity of the lithium-ion battery to measure the correlation between the two, selecting aging features with high correlation coefficients to form a feature matrix, and normalizing the feature matrix; S4, performing dimensionality reduction processing on the feature matrix using kernel principal component analysis; S5. Construct a gated recurrent unit (GRU) network model, use the feature matrix obtained by dimensionality reduction as input, train the GRU model, complete the training process, and obtain a trained GRU model; finally, estimate the SOH of the lithium-ion battery through the trained GRU model.

2. The method for estimating a lithium-ion battery state by feature dimension reduction according to claim 1, characterized in that: The charging curve in step S1 includes: a charging current curve, a charging voltage curve, a charging time curve and a capacity increment IC curve.

3. A lithium-ion battery state estimation method with feature dimension reduction as claimed in claim 2, characterized in that: Step S1 is specifically as follows: S11. Obtain lithium-ion battery degradation data D B ; S12. According to the degradation data D B Obtain charging data in the constant current and constant voltage stage; S13, obtaining a charging current curve, a charging voltage curve, and a charging time curve according to the charging data; S14, obtaining a capacity increment IC curve through a constant current charging current curve, a constant current charging time curve, and a constant current charging voltage curve.

4. A lithium-ion battery state estimation method with feature dimension reduction as claimed in claim 3, characterized in that: In step S14, the IC curve is obtained: the calculation formula of the IC curve is: Q=I·t Among them, Q is the battery capacity, t is the constant current charging time, I is the constant current charging current, and V is the constant current charging voltage.

5. A lithium-ion battery state estimation method with feature dimension reduction as claimed in claim 4, characterized in that: For the capacity increment IC curve, a Savitzky-Golay filter is selected to filter out noise interference. The calculation formula of the Savitzky-Golay filter is: Among them, b k are the polynomial coefficients, i is the obtained IC curve, and y(n) is the filtered IC curve.

6. The method for estimating a lithium-ion battery state by feature dimension reduction according to claim 1, characterized in that: Step S2 is specifically as follows: S21, according to the obtained charging time curve, obtain the constant current charging time F1, the constant voltage charging time F2, the total constant current and constant voltage charging time F3, and the ratio of the constant current stage charging time to the total charging time F4; S22, according to the obtained charging current curve, obtaining a constant current charging current curve area F5, a constant voltage charging current curve area F6, and a total area F7 of the constant current and constant voltage charging current curve with time as the horizontal axis; S23, according to the obtained charging voltage curve, obtaining the difference F8 between the maximum voltage and the minimum voltage in the constant current stage; S24, obtaining the maximum peak value F9 of the IC curve and the voltage F10 corresponding to the maximum peak value.

7. A lithium-ion battery state estimation method using feature dimension reduction as claimed in claim 6, characterized in that: The specific steps of step S3 are as follows: S31. The correlation between the aging characteristics F1-F10 obtained in S2 and the maximum capacity is measured by calculating the Pearson correlation coefficient PCC. The Pearson correlation coefficient calculation formula is: Among them, z i For aging characteristics, is the average value of the aging characteristics, q i is the maximum capacity, is the average value of the maximum capacity; S32, selecting aging features whose absolute values ​​of Pearson correlation coefficients are above a preset value to form a feature matrix, and normalizing the feature matrix using a maximum and minimum normalization method, as shown in the following formula: Among them, F Sim is the degradation data of lithium-ion batteries, F Sim The maximum and minimum values ​​in are denoted as and is the normalized data.

8. The method for estimating a lithium-ion battery state by feature dimension reduction according to claim 1, characterized in that: The specific steps of step S4 are as follows: S41, comparing the correlation coefficient and partial correlation coefficient between the features through the KMO test statistic, when the KMO value exceeds the preset value, proceed to step S42, otherwise no processing is performed; S42. Calculate the Gaussian kernel matrix of the feature matrix. The Gaussian kernel function is: k(x i ,x j )=exp(-γ||x i -x j || 2 ) Among them, x i and x j is the aging feature, and γ is the hyperparameter of the Gaussian kernel function. The Gaussian kernel matrix K of the feature matrix is ​​obtained through the Gaussian kernel function: S43. Perform the following centralization processing on the Gaussian kernel matrix: K′=K-1 n K-K1 n +1 n K1 n Among them, K′ is the centralized Gaussian kernel matrix, K is the Gaussian kernel matrix, 1 n is an n-dimensional matrix, in which each element is equal to 1 / n; S44, obtaining the eigenvalues ​​and eigenvectors of the centralized Gaussian kernel matrix; S45, calculating the contribution rate and cumulative contribution rate of the eigenvalue according to the eigenvalue obtained in S44; S46. Take the eigenvectors corresponding to the first k eigenvalues ​​whose cumulative contribution rates reach 80% as principal components, and use them to construct a new feature matrix to obtain a matrix after dimensionality reduction.

9. A storage medium, characterized in that: The storage medium stores instructions and data for implementing a lithium-ion battery state estimation method with feature dimension reduction as described in any one of claims 1 to 8.

10. A lithium-ion battery state estimation device with feature dimension reduction, characterized in that: include: Processor and storage medium; the processor loads and executes instructions and data in the storage medium to implement a lithium-ion battery state estimation method with feature dimension reduction as described in any one of claims 1 to 8.

Citation Information

Patent Citations

  • Identifying method and system for battery capacity fading mechanism

    CN103698714A

  • Dailyload curve dimensionality reduction clustering method based on kernel principal component analysis

    CN109871860A

  • Online estimation method for vehicle power battery capacity based on constant current charging process

    CN110146822A

  • Battery health state estimation method based on charging data and LSTM neural network

    CN113917337A