Lithium ion battery state-of-health estimation method based on multi-scale Gream matrix entropy

Abstract

The invention relates to a lithium ion battery state-of-health estimation method based on multi-scale Gream matrix entropy. The method comprises the following steps: acquiring a voltage signal and a current signal of a target lithium ion battery; converting the voltage signal and the current signal into a two-dimensional image by using a Grubrum matrix, and performing multi-scale decomposition on the two-dimensional image to obtain images of different scales; the Shannon entropy is adopted to quantify the images of different scales, health feature information is obtained, and the health feature information comprises a voltage entropy feature and a current entropy feature; the health feature information is input into an MLP model, the SOH estimated value of the target lithium ion battery is obtained, and the MLP model is obtained through training of an aging data set of the battery. According to the invention, in a battery aging cycle process, health features highly related to SOH can be extracted only through current and voltage signals, and the aging degree of the battery is objectively reflected.

Description

Technical Field

[0001] This invention relates to the field of lithium battery health state estimation technology, and in particular to a method for estimating the health state of lithium-ion batteries based on multi-scale Gram matrix entropy. Background Technology

[0002] Lithium-ion batteries, due to their high energy density, high charge-discharge efficiency, long cycle life, and environmental friendliness, have been widely used in various fields such as consumer electronics, electric vehicles, and aerospace. However, as a dynamic and time-varying electrochemical system, lithium-ion batteries undergo various irreversible electrochemical reactions during operation, such as lithium compound deposition, solid-liquid interface film growth, and increased charge transport impedance. These electrochemical reactions lead to gradual degradation of battery performance and a continuous decrease in usable capacity, and may even pose potential safety risks. Therefore, accurately assessing the state of harmonics (SOH) of lithium-ion batteries is crucial for ensuring the safe and stable operation of energy storage systems.

[0003] Methods for estimating the state of health (SOH) of lithium-ion batteries can be broadly categorized into two types: model-based methods and data-driven methods. Model-based methods primarily include equivalent circuit models and electrochemical models, which construct physical models to reflect the electrochemical characteristics during battery aging. While model-based methods offer strong physical interpretability and can reflect the internal electrochemical mechanisms of the battery, their modeling process is complex, parameter identification is difficult, they are susceptible to external factors, and computational costs are high, making them unsuitable for real-time performance. Data-driven methods, on the other hand, do not require modeling the complex electrochemical mechanisms within lithium-ion batteries. Instead, they achieve high-precision SOH estimation by organizing and analyzing historical operating data.

[0004] While data-driven methods have made significant progress in battery SOH estimation, they still have some significant limitations. Existing data-driven methods mainly fall into two categories for feature extraction: manual feature selection based on expert experience and automatic feature extraction based on deep learning models. Manual feature selection relies heavily on extensive professional experience, resulting in a complex and inefficient feature extraction process. While using deep learning methods for battery feature extraction avoids the tedious manual feature selection process, it demands high computational resources, and the model's interpretability and reproducibility are relatively poor.

[0005] Therefore, to address the above problems, this invention designs a method for estimating the health status of lithium-ion batteries based on multi-scale Gram matrix entropy. Summary of the Invention

[0006] The purpose of this invention is to provide a method for estimating the state of health (SOH) of lithium-ion batteries based on multi-scale Gram matrix entropy. This method automatically extracts features highly correlated with the SOH of batteries from different battery aging datasets and achieves accurate estimation of SOH.

[0007] To achieve the above objectives, the present invention provides the following solution:

[0008] A method for estimating the state of health of lithium-ion batteries based on multi-scale Gram matrix entropy includes:

[0009] Collect voltage and current signals from the target lithium-ion battery;

[0010] The voltage signal and the current signal are converted into two-dimensional images using a Gram matrix, and the two-dimensional images are then decomposed into multi-scale images to obtain images at different scales.

[0011] The images at different scales are quantized using Shannon entropy to obtain health feature information, which includes voltage entropy features and current entropy features.

[0012] The health feature information is input into the MLP model to obtain the SOH estimate of the target lithium-ion battery, wherein the MLP model is trained using the battery aging dataset.

[0013] Optionally, converting the voltage signal and the current signal into a two-dimensional image using a Gram matrix includes:

[0014] The voltage signal and the current signal are normalized to obtain a normalized sequence;

[0015] The angular components corresponding to the normalized sequence are calculated using the inverse cosine function, and the angular differences between each angular component are calculated using the cosine function to construct the Gram matrix GASF as the two-dimensional image.

[0016] Optionally, the constructed Gram matrix is:

[0017] ;

[0018] in, For Gram matrix, , , Let i, j, and n be the angular components corresponding to the i, j, and nth sequences, where i, j ∈ {1, 2, ..., n}.

[0019] Optionally, performing multi-scale decomposition on the two-dimensional image includes:

[0020] A two-dimensional Gaussian filter is used to perform a convolution operation on the two-dimensional image to obtain the filtered image;

[0021] The filtered image is downsampled to obtain images at different scales.

[0022] Optionally, obtaining the filtered image includes:

[0023] ;

[0024] in, The image is the filtered image, and G(u,v;σ) is the two-dimensional Gaussian kernel. Standard deviation This represents half the size of the Gaussian kernel, and u and v represent the horizontal and vertical coordinates in two-dimensional space relative to the center of the kernel, respectively.

[0025] Optionally, quantizing the images at different scales using Shannon entropy includes:

[0026] The pixel values ​​of the images at different scales are mapped to probability distributions, and the Shannon entropy is used to quantitatively analyze the probability distributions to obtain the health feature information.

[0027] Optionally, the health characteristic information is obtained as follows:

[0028] ;

[0029] in, Let H be a probability distribution with 256 gray levels. Arbitrary images at different scales, For health characteristics information.

[0030] The beneficial effects of the present invention are as follows: (1) The present invention innovatively introduces the Gram matrix into the signal entropy feature extraction process, and realizes data enhancement processing of battery voltage and current signals by converting one-dimensional signals into two-dimensional images.

[0031] (2) This invention proposes a multi-scale image decomposition method based on Gaussian filtering and downsampling, which can decompose the image layer by layer at multiple scales, thereby fully extracting its pixel distribution features.

[0032] (3) This invention proposes a battery SOH estimation framework based on MGME. This method can directly extract the battery's health characteristics from voltage and current signals and achieve accurate estimation of the battery's SOH. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a flowchart of a lithium-ion battery health state estimation method based on multi-scale Gram matrix entropy, according to an embodiment of the present invention.

[0035] Figure 2 This is a framework diagram of a lithium-ion battery health state estimation method based on multi-scale Gram matrix entropy according to an embodiment of the present invention.

[0036] Figure 3 The present invention provides a battery charging and discharging device and a battery SOH curve, wherein (a) is the battery charging and discharging device and (b) is the battery SOH curve;

[0037] Figure 4 The entropy heatmap of the lithium-ion battery health state estimation method MGME based on multi-scale Gram matrix entropy in an embodiment of the present invention is shown on each battery, where (a) is the current entropy and (b) is the voltage entropy.

[0038] Figure 5 The entropy characteristic curves of the lithium-ion battery health state estimation method MGME based on multi-scale Gram matrix entropy in the embodiments of the present invention are shown on each battery, where (a) is the current entropy and (b) is the voltage entropy.

[0039] Figure 6 The absolute correlation coefficients between the entropy features and the SOH of the battery in the embodiments of the present invention are given, wherein (a) is the absolute correlation coefficient between the entropy features and SOH of the same battery at different scales, and (b) is the absolute correlation coefficient between the entropy features and SOH of different batteries at the same scale.

[0040] Figure 7 The following are the SOH estimation results of the MLP for lithium batteries according to the embodiments of the present invention, wherein (a) is a comparison between the predicted and actual SOH values ​​of the battery, and (b) is a comparison of the error indices of the MLP model. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] like Figure 1-2 As shown in the figure, this embodiment proposes a method for estimating the health status of lithium-ion batteries based on multi-scale Gram matrix entropy, including:

[0044] Collect voltage and current signals from the target lithium-ion battery;

[0045] The voltage and current signals are converted into two-dimensional images using the Gram matrix, and the two-dimensional images are then decomposed into multiple scales to obtain images at different scales.

[0046] Shannon entropy is used to quantize images at different scales to obtain health feature information, which includes voltage entropy features and current entropy features.

[0047] Health characteristic information is input into the MLP model to obtain the SOH estimate of the target lithium-ion battery. The MLP model is trained using the battery aging dataset.

[0048] Specifically, the multi-scale Gram matrix entropy (MGME) algorithm is used to extract entropy domain features from current and voltage signals that can effectively characterize the aging state of the battery. This method first uses a Gram matrix to convert time-series current and voltage data into a two-dimensional image; then, two-dimensional Gaussian filtering and downsampling are used to decompose the resulting image at multiple scales to obtain feature information at different levels; finally, Shannon entropy is used to quantify the pixel distribution complexity of the image at each scale, thereby revealing the health status information contained in the signal.

[0049] The aging test data of the lithium-ion battery were collected using a CT-4008Tn-5V12A-S1 charge-discharge tester, with a sampling frequency set to 10 Hz. The obtained data were randomly divided into training, validation, and test sets in a 2:1:7 ratio. The lithium battery chemical composition was Li(NiCoMn)O2, the charge-discharge rate was 1C-1C, the charge-discharge cutoff voltages were 4.2V and 2.75V respectively, and the nominal capacity was 1.2 Ah. The battery data acquisition equipment and SOH curve are shown below. Figure 3 As shown in (a)-(b).

[0050] Furthermore, converting voltage and current signals into two-dimensional images using a Gram matrix includes:

[0051] The voltage and current signals are normalized to obtain the normalized sequence;

[0052] The angular components corresponding to the normalized sequence are calculated using the inverse cosine function, and the angular differences between each angular component are calculated using the cosine function to construct the Gram matrix GASF as a two-dimensional image.

[0053] Specifically, (1) for any current or voltage sequence Normalization is performed to obtain the normalized sequence. .

[0054] (1);

[0055] (2) Calculate the sequence using the arccosine function (arccos). Corresponding angular components Based on the Gramian Angular Summation Field (GASF) structure, a symmetric GASF matrix is ​​constructed by using the cosine function (cos) to measure the angular differences between the angular components. This matrix can be viewed as a two-dimensional image. .

[0056] (2);

[0057] (3);

[0058] in, For Gram matrix, , , Let i, j, and n be the angular components corresponding to the i, j, and nth sequences, where i, j ∈ {1, 2, ..., n}.

[0059] Furthermore, multi-scale decomposition of two-dimensional images includes:

[0060] A two-dimensional Gaussian filter is used to perform a convolution operation on a two-dimensional image to obtain the filtered image.

[0061] The filtered image is downsampled to obtain images at different scales.

[0062] Specifically, to fully extract two-dimensional images To extract battery health information from images, this paper employs a multi-scale image decomposition method combining two-dimensional Gaussian filtering and downsampling. Specifically, the two-dimensional Gaussian kernel is first defined as... And the image is checked using the Gaussian kernel. Perform convolution operations to obtain the filtered image. Subsequently, regarding Perform downsampling operations to obtain images at different scales. .

[0063] (4);

[0064] (5);

[0065] (6);

[0066] In the formula, The standard deviation is used to control the smoothness during the filtering process; The size of half of the Gaussian kernel determines the range of the filter in the spatial domain, and u and v represent the horizontal and vertical coordinates in two-dimensional space relative to the center of the kernel, respectively.

[0067] The above steps can generate image sets at multiple scales. In this embodiment, the scale factor s is set to 6. Wherein, when s=1, the corresponding image... That is, the original image Images at other different scales Then it is generated sequentially using the following method.

[0068] (7);

[0069] In the formula, This indicates that the Gaussian filtering operation defined by formulas (4) and (5) is used. This indicates that the downsampling process defined by formula (6) is used.

[0070] Furthermore, Shannon entropy is used to quantize images at different scales, including:

[0071] Pixel values ​​of images at different scales are mapped to probability distributions, and Shannon entropy is used to quantitatively analyze the probability distributions to obtain health characteristic information.

[0072] Specifically, for image sets any image in Based on formula (8), its pixel values ​​are mapped to a probability distribution. Subsequently, Shannon entropy theory was introduced to quantitatively analyze the complexity of this probability distribution in order to extract the health characteristics contained therein.

[0073] (8);

[0074] (9);

[0075] In the formula, H represents a grayscale level of 256. For health characteristics information.

[0076] The voltage and current signals of the battery are feature extracted using the MGME algorithm, and the Pearson correlation coefficient (Formula 10) is used as an indicator to measure the correlation between the extracted features and the battery's state of health (SOH). Figure 4 (a)-(b) Figure 5 (a)-(b) and Figure 6 (a)-(b) respectively present the heat map of entropy characteristics, the entropy characteristic curve, and the absolute correlation coefficient between entropy characteristics and the battery SOH. Figure 4 The color gradient clearly shows the trajectory of entropy characteristics as the battery cycles. Figure 5 It can be seen that, across all scales, the current entropy generally shows a gradual decreasing trend with increasing cycle number, while the voltage entropy shows the opposite trend, gradually increasing. From Figure 6 It can be seen that the absolute correlation coefficients between the current entropy and voltage entropy features extracted by the MGME algorithm and SOH both exceed 0.9, indicating a strong correlation.

[0077] (10);

[0078] Where r represents the correlation coefficient, ranging from 1 to -1. En and SOH represent the entropy characteristic and battery SOH, respectively. and This represents their respective average values.

[0079] The SOH of the extracted MGME features was estimated using the MLP method, and the results are as follows: Figure 7 As shown in (a)-(b). In Figure 7 In this study, current entropy features and voltage entropy features are used together as inputs to a deep learning model, and the coefficient of determination (R²) is used as the input. 2 The root mean square error (RMSE), mean absolute percentage error (MAPE), and mean absolute error (MAE) were used to quantitatively assess their impact on the battery's state of equilibrium (SOH) regression performance. The results are as follows: Figure 7 As shown. The coefficient of determination (R²) of the MGME+MLP estimation results. 2 The RMSE (0.9829) indicates that the MLP method has good fitting results for the MGME feature set, confirming the reliability of the proposed framework. Furthermore, the RMSE, MAPE, and MAE indices of MGME+MLP are 1.8501%, 1.8165%, and 1.417%, respectively. These results demonstrate that the predicted results of the proposed framework have small deviations from the actual SOH results, validating its effectiveness.

[0080] (11);

[0081] Where n is the number of samples, and y i These are the estimated value and the actual value, respectively. For y i The average value.

[0082] This embodiment can automatically extract health features that effectively characterize the aging state of a battery from voltage and current signals during the battery charging and discharging process. Furthermore, a mapping relationship between MGME features and battery SOH is established using a multilayer perceptron (MLP), constructing an MLP+MGME battery SOH estimation framework, thereby achieving high-precision SOH prediction. This method can extract health features highly correlated with SOH using only current and voltage signals during battery aging cycles, objectively reflecting the degree of battery aging, and has good potential for practical application.

[0083] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for estimating the health status of lithium-ion batteries based on multi-scale Gram matrix entropy, characterized in that, include: Collect voltage and current signals from the target lithium-ion battery; The voltage signal and the current signal are converted into two-dimensional images using a Gram matrix, and the two-dimensional images are then decomposed into multi-scale images to obtain images at different scales. The images at different scales are quantized using Shannon entropy to obtain health feature information, which includes voltage entropy features and current entropy features. The health feature information is input into the MLP model to obtain the SOH estimate of the target lithium-ion battery, wherein the MLP model is trained using the battery aging dataset.

2. The lithium-ion battery health state estimation method based on multi-scale Gram matrix entropy according to claim 1, characterized in that, Converting the voltage and current signals into a two-dimensional image using a Gram matrix includes: The voltage signal and the current signal are normalized to obtain a normalized sequence; The angular components corresponding to the normalized sequence are calculated using the inverse cosine function, and the angular differences between each angular component are calculated using the cosine function to construct the Gram matrix GASF as the two-dimensional image.

3. The lithium-ion battery health state estimation method based on multi-scale Gram matrix entropy according to claim 2, characterized in that, The constructed Gram matrix is ​​as follows: ； in, For Gram matrix, , , Let i, j, and n be the angular components corresponding to the i, j, and nth sequences, where i, j ∈ {1, 2, ..., n}.

4. The lithium-ion battery health state estimation method based on multi-scale Gram matrix entropy according to claim 1, characterized in that, Multi-scale decomposition of the two-dimensional image includes: A two-dimensional Gaussian filter is used to perform a convolution operation on the two-dimensional image to obtain the filtered image; The filtered image is downsampled to obtain images at different scales.

5. The lithium-ion battery health state estimation method based on multi-scale Gram matrix entropy according to claim 4, characterized in that, The filtered image is obtained by: ； in, The image is the filtered image, and G(u,v;σ) is the two-dimensional Gaussian kernel. Standard deviation This represents half the size of the Gaussian kernel, and u and v represent the horizontal and vertical coordinates in two-dimensional space relative to the center of the kernel, respectively.

6. The lithium-ion battery health state estimation method based on multi-scale Gram matrix entropy according to claim 1, characterized in that, The quantization of the images at different scales using Shannon entropy includes: The pixel values ​​of the images at different scales are mapped to probability distributions, and the Shannon entropy is used to quantitatively analyze the probability distributions to obtain the health feature information.

7. The lithium-ion battery health state estimation method based on multi-scale Gram matrix entropy according to claim 6, characterized in that, The health characteristic information is obtained as follows: ； in, Let H be a probability distribution with 256 gray levels. Arbitrary images at different scales, For health characteristics information.

Images