A method for estimating the state of health of lithium-ion batteries based on graph representation learning

Abstract

The present invention discloses a method for estimating the health status of lithium-ion batteries based on graph representation learning. The method comprises the following steps: first, collecting the complete curves of the constant current charging stage, constant voltage charging stage or constant current discharging stage in each charge and discharge cycle of the lithium-ion battery; intercepting the complete curves of the constant current charging stage, constant voltage charging stage or constant current discharging stage in each charge and discharge cycle to form new segments of the constant current charging stage, constant voltage charging stage or constant current discharging stage in each charge and discharge cycle; using an extreme learning machine for interpolation prediction of the new segments, and establishing a database required for graph representation learning; performing graph representation and graph learning. Since most of the research on the health status of lithium-ion batteries focuses on feature engineering, the effect of health status estimation depends on feature selection and processing, while the method of the present invention avoids the heavy burden of feature engineering and has good practical significance; in addition, the present invention can also better guide the sampling process.

Description

Technical Field

[0001] The present invention relates to the field of battery health state estimation, and in particular to a lithium ion battery health state estimation method based on graph representation learning. Background Art

[0002] In recent years, lithium-ion batteries have been recognized by the industry and supported by policies due to their high energy density, low self-discharge rate, high efficiency and stability, but frequent accidents have also made people realize the importance of battery health management. The current battery health status estimation mainly relies on feature engineering, which extracts various features to describe the health status of the battery. However, the feature extraction process is arduous and complicated, and the extracted features are not generalizable. For example, the document "Joint Estimation Method of SOH and RUL of Lithium-ion Batteries Based on Health Feature Parameters" initially selected 12 features, which were initially screened by Pearson and Spearman analysis, and then optimized by principal component analysis. It can be seen that the workload of feature engineering is huge, and since the operating data of lithium-ion batteries is difficult to obtain, the essence of lithium-ion battery health status assessment is a small sample problem, and the features extracted by feature engineering have poor generalization in small sample problems and are difficult to be used for the assessment of other battery health status. In addition, most studies carry out feature engineering based on complete voltage, current or temperature curves, but in reality, it is difficult to obtain complete data. The present invention adopts an end-to-end approach, relying on the powerful expressive ability of graph representation learning to automatically extract features from curves and complete the estimation of health status. This method does not require feature engineering, has strong generalization performance, and is only applied to local curves, which has the prospect of rapid health status assessment. Summary of the invention

[0003] To solve the above problems, the present invention proposes a method for estimating the health state of lithium-ion batteries based on graph representation learning. First, the complete curves of the constant current charging, constant voltage charging or constant current discharging stages in each cycle of the lithium-ion battery are collected, and then the time used in the constant current charging, constant voltage charging or constant current discharging stage in the last cycle is used as a benchmark; next, according to the above benchmark, the complete curves of the constant current charging, constant voltage charging or constant current discharging stages in each cycle are intercepted to form new segments of the constant current charging, constant voltage charging or constant current discharging stages in each cycle; then, for the new segments of the constant current charging, constant voltage charging or constant current discharging stages in each cycle, time is used as a feature and voltage, current or temperature is used as a marker. The trained extreme learning machine is used to perform prediction and interpolation on the set sampling points, so as to obtain the voltage, current or temperature of the constant current charging, constant voltage charging or constant current discharging stage in each cycle corresponding to the sampling time point, thereby forming a database for graph representation learning; finally, for the constant current charging, constant voltage charging or constant current discharging stage in each cycle, the sampling time point is used as the node of the graph in the graph representation learning, and the voltage, current or temperature corresponds to the feature of the node, the successive connections of the sampling moments constitute the edges, and the health status is used as the label corresponding to the graph. Different cycles are isomorphic graphs, so the graph representation learning method can be used for learning and prediction of new samples.

[0004] The purpose of the present invention is achieved by at least one of the following technical solutions.

[0005] A method for estimating the health state of a lithium-ion battery based on graph representation learning comprises the following steps:

[0006] S1. First, collect the complete curve of the constant current charging stage, constant voltage charging stage or constant current discharging stage in each charge and discharge cycle of the lithium-ion battery;

[0007] S2, intercepting the complete curve of the constant current charging stage, the constant voltage charging stage or the constant current discharging stage in each charge and discharge cycle to form a new segment of the constant current charging stage, the constant voltage charging stage or the constant current discharging stage in each charge and discharge cycle;

[0008] S3. For new segments, an extreme learning machine is used to perform interpolation prediction and establish a database required for graph representation learning;

[0009] S4. Perform graph representation and graph learning.

[0010] Furthermore, in step S1, the current-time curve, the voltage-time curve and the temperature-time curve in the constant current charging stage, the constant voltage charging stage or the constant current discharging stage are collected.

[0011] Furthermore, the charging capacity of the charging process or the discharging capacity of the discharging process in the charge and discharge cycle, as well as the rated capacity of the lithium-ion battery are collected.

[0012] Further, in step S2, as the lithium-ion battery ages, the time used in the constant current charging stage, the constant voltage charging stage or the constant current discharging stage tends to decrease, so the time used in the constant current charging stage, the constant voltage charging stage or the constant current discharging stage in the last charge and discharge cycle is used as the corresponding unified benchmark;

[0013] The complete curves of different charge and discharge cycles of the constant current charging stage, the constant voltage charging stage or the constant current discharging stage collected in step S1 are clipped according to the corresponding unified benchmark, so as to obtain the maximum curve segment, i.e., the new segment, of the constant current charging stage, the constant voltage charging stage or the constant current discharging stage of each cycle that is not greater than the unified benchmark time length.

[0014] Further, in step S3, for the current-time curve, voltage-time curve or temperature-time curve at different stages, different sampling moments correspond to different voltages, currents or temperatures;

[0015] Due to sampling reasons, the voltage, current or temperature at the required time may not be available. Therefore, an extreme learning machine is used for interpolation prediction, and the sampling time data set τ = {τ0, τ1, τ2, ...τ n} as input, the voltage data set U at the corresponding moment = {v0, v1, v2, ...v n}、Current data set I={i0,i1,i2,...i n}、Temperature data set T={t0,t1,t2,...t n} are used as outputs, respectively, and the gradient descent method is used for training, and the trained extreme learning machine is obtained, where n is the nth sampling moment; in addition, τ n is the n+1th interpolation moment, τ is the data set consisting of n+1 interpolation moments, v n is the interpolation voltage corresponding to the n+1th interpolation moment, U is the data set consisting of n+1 interpolation voltages, i n is the interpolation current corresponding to the n+1th interpolation moment, I is the data set consisting of n+1 interpolation currents, t n is the interpolation temperature corresponding to the n+1th interpolation moment, and T is the data set consisting of n+1 interpolation temperatures;

[0016] Afterwards, the unified reference Base1 corresponding to the constant current charging stage, the unified reference Base2 corresponding to the constant voltage charging stage, or the unified reference Base3 corresponding to the constant current discharging stage are interpolated, and the sampling time is re-divided into τ interp1 ={τ0,τ s1 , τ 2·s1 , ...τ k1·s1}、τ interp2={τ0,τ s2 , τ 2·s2 , ...τ k2·s2} or τ interp3 ={τ0,τ s1 , τ 2·s3 ,…τ k3·s3}, where s1, s2, s3 are the resampling steps used in different stages, k1, k2, k3 are the number of resamplings in different stages minus one, and the numbers 1, 2, 3 after in erp, s and k represent constant current charging, constant voltage charging and constant current discharging respectively; τ interp1 is the interpolation time data set of the constant current charging stage, τ k1·s1 is the time s1 of the constant current charging stage k1, τ interp2 is the interpolation time data set of the constant voltage charging stage, τ k2·s2 is the time s2 of the constant voltage charging stage k2, τ interp3 is the interpolation time data set of the constant current discharge stage, τ k3•s3 It is the moment s3 of the constant current discharge stage k3.

[0017] Furthermore, by interp1 , τ interp2 and τ interp3 Substitute them into the corresponding extreme learning machine respectively to predict the voltage, current or temperature U corresponding to the resampling time interp1 = {v0, v s1 , v 2·s1 ,…v k1·s1}、U interp3 = {v0, v s3 , v 2·s3 ,...v k3·s3}、I interp2 ={i0,i s2 ,i 2·s2 ,...i k2·s2}、T interp1 ={t0, t s1 , t 2·s1 ,…t k1·s1}、T interp2 ={t0, t s2 , t 2·s2 ,...t k2·s2}、T interp3 ={t0, t s3 , t 2·s3 ,...t k3·s3}, since constant current means constant current and constant voltage means constant voltage, there is no need to interpolate the current in the constant current stage. Similarly, there is no need to interpolate the voltage in the constant voltage stage;

[0018] Among them, U interp1is the voltage interpolation data set in the constant current charging stage, v k1·s1 is the interpolated voltage at time s1 in the constant current charging stage k1, U interp3 The interpolated voltage data set, v, for the constant current discharge phase k3•s3 is the interpolation voltage at time s3 in the constant current discharge stage k3, I interp2 Interpolation current data set for constant voltage charging stage, i k2·s2 is the interpolated current at time s2 in the constant voltage charging stage k2, T interp1 is the interpolation time data set of the constant current charging stage, t k1·s1 is the time s1 of the constant current charging stage k1, T interp2 is the interpolation time data set of the constant voltage charging stage, t k2·s2 is the s2 moment of the constant voltage charging stage k2, T interp3 is the interpolation time data set of the constant current discharge stage, t k3•s3 It is the moment s3 of the constant current discharge stage k3.

[0019] Furthermore, in step S4, the interpolation moments corresponding to different stages are used as graph nodes, the corresponding voltages, currents or temperatures are used as attributes / features of the nodes, graph edges are established between each node (moment) and the nodes (moments) after the node (moment), and the charging capacity, discharging capacity or health status of this charge and discharge cycle are used as graph labels, thereby completing the composition of new fragments for each charge and discharge stage, wherein the health status of the charging process is the ratio of the charging capacity of the charging process to the rated capacity; the health status of the discharging process is the ratio of the discharge capacity of the discharging process to the rated capacity.

[0020] Furthermore, in step S4, according to the composition of the new fragments for each charging and discharging stage, a graph learning method applicable to space is adopted to perform graph-level learning to complete the regression task.

[0021] Furthermore, the spatially applicable graph learning method includes a graph attention mechanism.

[0022] Compared with the prior art, the present invention has the following advantages and effects:

[0023] (1) No need to perform complex feature engineering work.

[0024] (2) The selection of the resampling step size can provide guidance for the sampling work in the actual process.

[0025] (3) This is the first attempt to apply graph representation learning to the battery field, providing new research ideas for battery health status assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art.

[0027] Figure 1 is an implementation flow chart of an embodiment of the present invention;

[0028] Figure 2 It is a constant current discharge time decay diagram during the aging process in an embodiment of the present invention;

[0029] Figure 3 Schematic diagram of constant current discharge voltage during the aging process in an embodiment of the present invention;

[0030] Figure 4 It is a schematic diagram of a time decay segment of a constant current discharge during an aging process in an embodiment of the present invention;

[0031] Figure 5 A graph constructed for learning to represent graphs in an embodiment of the present invention;

[0032] Figure 6 A schematic diagram showing a curve diagram in an embodiment of the present invention;

[0033] Figure 7 This is a schematic diagram of the structure of the map learning model in an embodiment of the present invention;

[0034] Figure 8 It is a schematic diagram of the prediction effect in the embodiment of the present invention. DETAILED DESCRIPTION

[0035] The core of the present invention is to provide a lithium-ion battery health state estimation method based on graph representation learning, which avoids feature extraction work, provides guidance for sampling work, and provides a new idea for lithium-ion battery health state estimation.

[0036] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0037] Embodiment 1:

[0038] A method for estimating the health status of lithium-ion batteries based on graph representation learning, such as Figure 1 As shown, the following steps are included:

[0039] S1. First, collect the complete curve of the constant current charging stage, constant voltage charging stage or constant current discharging stage in each charge and discharge cycle of the lithium-ion battery;

[0040] Collect the current-time curve, voltage-time curve and temperature-time curve in the constant current charging stage, constant voltage charging stage or constant current discharging stage;

[0041] Collect the charging capacity of the charging process or the discharging capacity of the discharging process in the charge and discharge cycle, as well as the rated capacity of the lithium-ion battery;

[0042] In this embodiment, taking the Oxford University lithium-ion battery public data set as an example, for the constant current discharge stage, the collected complete curve includes the discharge voltage-time curve;

[0043] S2, intercepting the complete curve of the constant current charging stage, the constant voltage charging stage or the constant current discharging stage in each charge and discharge cycle to form a new segment of the constant current charging stage, the constant voltage charging stage or the constant current discharging stage in each charge and discharge cycle;

[0044] As lithium-ion batteries age, the time spent in the constant current discharge phase tends to decrease, so the time spent in the constant current discharge phase in the last charge and discharge cycle is used as the corresponding unified benchmark;

[0045] The complete curves of different charge and discharge cycles in the constant current discharge phase collected in step S1 are clipped according to the corresponding unified benchmark, so as to obtain the maximum curve segment of the constant current discharge phase of each cycle that is not greater than the unified benchmark time length, i.e., the new segment.

[0046] In this embodiment, for each charge and discharge cycle m, the constant current discharge time l is recorded. m , thus forming a set L = {l1, l2, ..., l m ,...l M}, as the battery ages, l m It shows a decreasing trend, m≤M, M is the total number of cycles, such as Figure 2 As shown in the figure, the time taken for the last charge and discharge cycle constant current discharge is the shortest, about 2730.052s. Figure 3 It also shows that with the aging process, the discharge time is getting shorter and shorter, so the last time l can be taken M As a benchmark, all discharge curves in the charge and discharge cycles are discarded. M The subsequent part makes the length of each discharge curve basically the same, such as Figure 4 shown.

[0047] S3. For new segments, an extreme learning machine is used to perform interpolation prediction and establish a database required for graph representation learning;

[0048] For the voltage-time curves at different stages, different sampling moments correspond to different voltages;

[0049] Due to sampling reasons, the voltage at the required time may not be available, so an extreme learning machine is used for interpolation prediction, and the sampling time data set τ = {τ0, τ1, τ2, ...τ n} as input, the voltage data set U at the corresponding moment = {v0, v1, v2, ...v n} as the output, the gradient descent method is used for training, and the trained extreme learning machine is obtained, where n is the nth sampling moment; in addition, τ n is the n+1th interpolation moment, τ is the data set consisting of n+1 interpolation moments, v n is the interpolation voltage corresponding to the n+1th interpolation moment, and U is the data set consisting of n+1 interpolation voltages;

[0050] Afterwards, the unified benchmark Base3 corresponding to the constant current discharge stage is interpolated and the sampling time is re-divided into τ interp3 ={τ0,τ s1 , τ 2·s3 , ...τ k3·s3}, where s3 is the interpolation step size used in the constant current discharge stage, k3 is the number of resamples minus one in the constant current discharge stage, and the number 3 after int erp, s and k represents constant current discharge; τ interp3 is the interpolation time data set of the constant current discharge stage, τ k3•s3 It is the s3 moment of the constant current discharge stage k3;

[0051] By dividing τ interp3 Substitute them into the corresponding extreme learning machine respectively to predict the voltage U corresponding to the resampling time interp3 = {v0, v s3 , v 2·s3 ,...v k3·s3}, since constant current means that the current is constant, there is no need to interpolate the current during the constant current stage;

[0052] Among them, U interp3 The interpolated voltage data set, v, for the constant current discharge phase k3•s3 is the interpolation voltage at time s3 in the constant current discharge stage k3, T interp3 is the interpolation time data set of the constant current discharge stage, t k3•s3 It is the moment s3 of the constant current discharge stage k3.

[0053] S4, perform graph representation and graph learning;

[0054] The figure shows Figure 5 and Figure 6As shown, the interpolation time corresponding to the constant current discharge stage is used as the graph node, the corresponding voltage is used as the attribute / feature of the node, each node (time) is connected to the node (time) after the node (time) to establish an edge, and the charging capacity, discharge capacity or health status of this charge and discharge cycle is used as the label of the graph, thereby completing the composition of the new fragment for each charge and discharge stage, wherein the health status of the charging process is the ratio of the charging capacity of the charging process to the rated capacity; the health status of the discharge process is the ratio of the discharge capacity of the discharge process to the rated capacity. In this embodiment, taking the deep learning pyg framework programming as an example, a graph learning model is established, such as Figure 7 As shown in the figure, the implementation process is as follows: two layers of attention mechanism GATConv, followed by pooling global_mean_pool, and finally connected to the fully connected layer Linear to get the output, and the loss (using the mse loss function) between the output and the actual value (that is, the health status of the cycle) is back-propagated, and the network weights are updated using gradient descent, and the iteration is continuously updated to gradually optimize the model.

[0055] Since the Oxford University lithium-ion battery public data set has a total of 8 batteries from Cell1 to Cell8, Cell2 is used as the training sample, Cell1 is the verification sample, and the rest are test samples. Figure 8 The prediction effect of the test sample taking Cell3 as an example is shown. It can be seen that the prediction result is consistent with the actual result. In order to quantitatively describe the prediction effect, the relative percentage error (RPE) and maximum error (ME) are used to quantitatively evaluate the prediction result:

[0056]

[0057]

[0058] Table 1 gives the prediction results of Cell3 to Cell8:

[0059] Table 1 Error of SOH prediction results during constant current discharge of Oxford battery

[0060] Cell3 Cell4 Cell5 Cell6 Cell7 Cell8 ME lower limit -0.014 -0.006 -0.002 -0.003 -0.005 -0.005 ME upper limit 0.028 0.026 0.038 0.026 0.028 0.020 Lower RPE limit 0.006％ 0.104％ 0.036％ 0.010％ 0.001％ 0.022％ RPE upper limit 2.915％ 2.668％ 3.859％ 2.766％ 2.938％ 2.128％

[0061] Embodiment 2:

[0062] Similar to Example 1, Example 2 uses the voltage-time curve in the constant current charging stage for graphical representation to estimate the health status.

[0063] Embodiment 3:

[0064] Similar to Example 1, Example 3 is aimed at the constant voltage charging stage, but the difference is that the current-time curve of the constant voltage stage is used for graphical representation to estimate the health status.

[0065] The above embodiments are only specific embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are included in the protection scope of the present invention.

Claims

1. A method for estimating the health status of lithium-ion batteries based on graph representation learning, characterized in that: The following steps are involved: S1. First, collect the complete curve of the constant current charging stage, constant voltage charging stage or constant current discharging stage in each charge and discharge cycle of the lithium-ion battery; Collect the current-time curve, voltage-time curve and temperature-time curve in the constant current charging stage, constant voltage charging stage or constant current discharging stage; S2, intercepting the complete curve of the constant current charging stage, the constant voltage charging stage or the constant current discharging stage in each charge and discharge cycle to form a new segment of the constant current charging stage, the constant voltage charging stage or the constant current discharging stage in each charge and discharge cycle; The time spent in the constant current charging stage, constant voltage charging stage or constant current discharging stage in the last charge and discharge cycle is used as the corresponding unified benchmark; The complete curves of different charge and discharge cycles of the constant current charging stage, the constant voltage charging stage or the constant current discharging stage collected in step S1 are cut according to the corresponding unified benchmark, so as to obtain the maximum curve segment of the constant current charging stage, the constant voltage charging stage or the constant current discharging stage of each cycle that is not greater than the unified benchmark time length, i.e., the new segment; S3. For new segments, an extreme learning machine is used to perform interpolation prediction and establish a database required for graph representation learning; For the current-time curve, voltage-time curve or temperature-time curve at different stages, different sampling moments correspond to different voltages, currents or temperatures; The extreme learning machine is used for interpolation prediction, and the sampling time data set τ={τ0,τ1,τ2,...τ n } as input, the voltage data set U at the corresponding moment is U={v0,v1,v2,...v n }、Current data set I={i0,i1,i2,...i n }、Temperature data set T={t0,t1,t2,...t n } are used as outputs, respectively, and the gradient descent method is used for training, and the trained extreme learning machine is obtained, where n is the nth sampling moment; in addition, τ n is the n+1th interpolation moment, τ is the data set consisting of n+1 interpolation moments, v n is the interpolation voltage corresponding to the n+1th interpolation moment, U is the data set consisting of n+1 interpolation voltages, i n is the interpolation current corresponding to the n+1th interpolation moment, I is the data set consisting of n+1 interpolation currents, t n is the interpolation temperature corresponding to the n+1th interpolation moment, and T is the data set consisting of n+1 interpolation temperatures; Afterwards, the unified reference Base1 corresponding to the constant current charging stage, the unified reference Base2 corresponding to the constant voltage charging stage, or the unified reference Base3 corresponding to the constant current discharging stage are interpolated, and the sampling time is re-divided into τ interp1 ={τ0,τ s1 ,τ 2·s1 ,...τ k1·s1 }、τ interp2 ={τ0,τ s2 ,τ 2·s2 ,...τ k2·s2 } or τ interp3 ={τ0,τ s1 ,τ 2·s3 ,...τ k3·s3 }, where s1, s2, s3 are the resampling steps used in different stages, k1, k2, k3 are the number of resampling in different stages minus one, and the numbers 1, 2, 3 after interp, s and k represent constant current charging, constant voltage charging and constant current discharging respectively; τ interp1 is the interpolation time data set of the constant current charging stage, τ k1·s1 is the time s1 of the constant current charging stage k1, τ interp2 is the interpolation time data set of the constant voltage charging stage, τ k2·s2 is the time s2 of the constant voltage charging stage k2, τ interp3 is the interpolation time data set of the constant current discharge stage, τ k3·s3 It is the s3 moment of the constant current discharge stage k3; By dividing τ interp1 , τ interp2 and τ interp3 Substitute them into the corresponding extreme learning machine respectively to predict the voltage, current or temperature U corresponding to the resampling time interp1 ={v0,v s1 ,v 2·s1 ,...v k1·s1 }、U interp3 ={v0,v s3 ,v 2·s3 ,...v k3·s3 }、I interp2 ={i0,i s2 ,i 2·s2 ,...i k2·s2 }、T interp1 ={t0,t s1 ,t 2·s1 ,...t k1·s1 }、T interp2 ={t0,t s2 ,t 2·s2 ,...t k2·s2 }、T interp3 ={t0,t s3 ,t 2·s3 ,...t k3·s3 }, since constant current means constant current and constant voltage means constant voltage, there is no need to interpolate the current in the constant current stage. Similarly, there is no need to interpolate the voltage in the constant voltage stage; Among them, U interp1 is the voltage interpolation data set in the constant current charging stage, v k1·s1 is the interpolated voltage at time s1 in the constant current charging stage k1, U interp3 The interpolated voltage data set, v, for the constant current discharge phase k3·s3 is the interpolation voltage at time s3 in the constant current discharge stage k3, I interp2 Interpolation current data set for constant voltage charging stage, i k2·s2 is the interpolated current at time s2 in the constant voltage charging stage k2, T interp1 is the interpolation time data set of the constant current charging stage, t k1·s1 is the time s1 of the constant current charging stage k1, T interp2 is the interpolation time data set of the constant voltage charging stage, t k2·s2 is the s2 moment of the constant voltage charging stage k2, T interp3 is the interpolation time data set of the constant current discharge stage, t k3·s3 It is the s3 moment of the constant current discharge stage k3; S4. Perform graph representation and graph learning.

2. A method for estimating the health status of a lithium-ion battery based on graph representation learning according to claim 1, characterized in that: In step S1, the charging capacity of the charging process or the discharging capacity of the discharging process in the charge and discharge cycle, and the rated capacity of the lithium-ion battery are collected.

3. The method for estimating the health status of a lithium-ion battery based on graph representation learning according to claim 1, characterized in that: In step S4, the interpolation moments corresponding to different stages are used as graph nodes, the corresponding voltage, current or temperature is used as the attribute or feature of the node, the edge of the graph is established between each node and the node after the node, and the charging capacity, discharge capacity or health status of this charge and discharge cycle is used as the label of the graph, thereby completing the composition of the new fragment for each charge and discharge stage.

4. A method for estimating the health status of a lithium-ion battery based on graph representation learning according to any one of claims 1 to 3, characterized in that: in, The health status of the charging process is the ratio of the charging capacity to the rated capacity during the charging process; The health status of the discharge process is the ratio of the discharge capacity to the rated capacity during the discharge process.

5. The method for estimating the health status of a lithium-ion battery based on graph representation learning according to claim 3, characterized in that: In step S4, according to the composition of the new fragments for each charging and discharging stage, a graph learning method applicable to space is used to perform graph-level learning to complete the regression task.

6. A method for estimating the health status of a lithium-ion battery based on graph representation learning according to claim 5, characterized in that: The spatially applicable graph learning method includes a graph attention mechanism.

Images