A lithium battery state of health estimation method based on statistical feature analysis

Abstract

The application provides a lithium battery health state estimation method based on statistical characteristic analysis, comprising the following steps: S1, obtaining N initial characteristics according to the charging voltage curve, the charging capacity curve, the IC curve and the temperature curve of the battery, and obtaining six statistical characteristics through statistical analysis based on the N initial characteristics; S2, constructing a single-core Gaussian process regression model, and training the model by using the six statistical characteristics through permutation and combination to obtain a plurality of characteristic combinations; S3, estimating the lithium battery health state by using the trained single-core Gaussian process regression model and each characteristic combination, so as to determine the optimal characteristic combination; S4, constructing a double-core Gaussian process regression model, searching and optimizing the hyperparameters of the double-core Gaussian process regression model by using the Northern Goshawk algorithm, and obtaining a final prediction model; and S5, estimating the lithium battery health state according to the prediction model and the optimal characteristic combination. The application can effectively improve the estimation accuracy.

Description

Technical Field

[0001] This invention belongs to the field of battery state estimation, specifically relating to a method for estimating the health state of lithium batteries based on statistical feature analysis. Background Technology

[0002] Existing battery state of health (SOH) estimation methods can be broadly categorized into two types: model-based methods and data-driven methods. Model-based methods typically employ multiple models, such as equivalent circuit models (ECMs) and electrochemical models. The challenge with electrochemical models lies in the fact that battery SOH degradation can be caused by various failure mechanisms, and dimensionality reduction is required in subsequent calculations, making the establishment of a comprehensive electrochemical model nearly impossible. Compared to electrochemical models, ECMs are more concise. However, simple model structures struggle to reflect complex battery aging mechanisms, posing a challenge to accurate SOH estimation.

[0003] With the rapid development of artificial intelligence and machine learning, data-driven methods, which do not need to consider the internal electrochemical reactions and failure mechanisms of lithium batteries, have become a research hotspot for SOH estimation in recent years, compared to model-based methods. Data-driven methods include neural networks, support vector machines, and Gaussian filters. Compared to other data-driven methods, neural networks have stronger generalization capabilities and can achieve higher estimation accuracy. In data-driven neural network methods, the SOH of the battery is estimated by extracting health features (HF) from the aging process.

[0004] However, in existing technologies, some HF samples have low correlation with SOH during HF extraction, resulting in an inability to comprehensively characterize the battery aging process and making it difficult to obtain high estimation accuracy. Furthermore, existing technologies use all HF samples as input to estimate SOH, which can easily lead to overfitting, negatively impacting estimation accuracy. Additionally, some existing technologies use single-kernel Gaussian process regression (GPR) models for SOH estimation, but single-kernel models suffer from insufficient prediction accuracy and generalization ability. Moreover, existing technologies require manual calculation of GPR hyperparameters, which cannot guarantee that the hyperparameters will reach optimal values ​​or avoid errors caused by human intervention, thus resulting in insufficient estimation accuracy. Summary of the Invention

[0005] The purpose of this invention is to propose a method for estimating the health status of lithium batteries based on statistical feature analysis, which can effectively improve the estimation accuracy.

[0006] This invention is achieved through the following technical solution:

[0007] A method for estimating the health status of lithium batteries based on statistical feature analysis includes the following steps:

[0008] Step S1: Obtain N initial features based on the battery's charging voltage curve, charging capacity curve, IC curve, and temperature curve, and then obtain six statistical features based on these N initial features: mean feature F. A Median characteristics F M Lower quartile characteristics F LQ Upper quartile characteristics F UQ Range characteristic F R Standard deviation characteristics F SD ;

[0009] Step S2: Construct a single-kernel Gaussian process regression model. The six statistical features obtained in step S1 are permuted and combined to obtain multiple feature combinations. These multiple feature combinations are used to train the single-kernel Gaussian process regression model. The feature combination includes one or more statistical features.

[0010] Step S3: Estimate the health status of the lithium battery using the trained single-core Gaussian process regression model and the feature combinations from step S2, thereby determining the optimal feature combination.

[0011] Step S4: Construct a dual-core Gaussian process regression model, use the Northern Eagle algorithm to search and optimize the hyperparameters of the dual-core Gaussian process regression model, and input the obtained optimal hyperparameters into the dual-core Gaussian process regression model to obtain the final prediction model.

[0012] Step S5: Estimate the health status of the lithium battery based on the prediction model obtained in step S4 and the optimal feature combination obtained in step S3.

[0013] Furthermore, in step S1, the mean feature The median feature The lower quartile feature The upper quartile feature The range feature F R =HI N -HI1, the standard deviation characteristic Among them, HI i Let i represent the i-th initial feature, where i = 1, 2, ..., N.

[0014] Furthermore, in step S2, the feature combinations obtained by arranging and combining the six statistical features are as follows: feature combination including only one statistical feature, feature combination including two different statistical features, feature combination including three different statistical features, feature combination including four different statistical features, feature combination including five different statistical features, and feature combination including six statistical features.

[0015] Furthermore, in step S2, the kernel function of the constructed dual-kernel Gaussian process regression model is K = Ma3 + Ma5, where, In the formula, δ f1 δ represents the signal variance of the kernel function Ma3. l1 δ represents the length scale of the kernel function Ma3. f2 δ represents the signal variance of the kernel function Ma5. l2 x represents the length scale of the kernel function Ma5. i Represents the observed data, x j This represents the input feature vector.

[0016] Furthermore, in step S1, when obtaining the initial features based on the charging voltage curve, the interval of the charging voltage curve containing the voltage inflection point is selected, and values ​​are taken in this interval at the same voltage interval to obtain a certain amount of time data as the initial features.

[0017] Furthermore, in step S1, when obtaining initial features based on the charging capacity curve, the interval of the charging capacity curve containing the voltage inflection point is selected, and several capacity data are obtained in this interval at the same voltage interval as initial features.

[0018] Furthermore, in step S1, when obtaining initial features based on the IC curve, an interval containing voltage peaks in the IC curve is selected, and several IC data points are obtained within this interval at the same voltage interval as initial features.

[0019] Furthermore, in step S1, when obtaining the initial characteristics based on the temperature curve, the formula is used... Calculate the temperature change curve corresponding to the temperature curve, and obtain a number of temperature change data on the temperature change curve with the same voltage interval as the initial feature, where T(k) is the temperature value corresponding to voltage k, and Δ is the sampling interval.

[0020] The present invention has the following beneficial effects:

[0021] This invention obtains N initial features based on the battery's charging voltage curve, charging capacity curve, IC curve, and temperature curve. Six statistical features are then derived from the distribution of these N initial features. These six statistical features are further combined to obtain multiple feature combinations. A conventional single-core Gaussian process regression model is first constructed. This model is then trained using these feature combinations. The optimal feature combination is obtained from the trained single-core Gaussian process regression model. A dual-core Gaussian process regression model is then constructed. The hyperparameters of this dual-core Gaussian process regression model are searched and optimized using the Northern Eagle algorithm to obtain the final prediction model. Finally, based on this prediction model and... The optimal feature combination is used to estimate the health status of lithium batteries. In this process, initial features are extracted from multiple aspects to comprehensively characterize the battery aging process. The optimal statistical features are first selected using a conventional single-core Gaussian process regression model for the final estimation, which avoids the overfitting problem. Furthermore, a dual-core Gaussian process regression model is used in the final estimation to avoid the problems of insufficient prediction accuracy and generalization ability of a single core. The use of the Northern Eagle hyperparameter optimization can ensure that the hyperparameters reach the optimal value while avoiding errors caused by human operation. In summary, this invention combines optimal processing from various perspectives to effectively improve the accuracy of battery health status estimation. Attached Figure Description

[0022] The present invention will now be described in further detail with reference to the accompanying drawings.

[0023] Figure 1 This is a flowchart of the present invention.

[0024] Figure 2 This is a graph showing the charging voltage variation of the present invention.

[0025] Figure 3 This is a graph showing the change in charging capacity according to the present invention.

[0026] Figure 4 This is an IC curve diagram of the present invention.

[0027] Figure 5 This is a temperature curve diagram of the present invention.

[0028] Figure 6 This is a temperature change curve of the present invention.

[0029] Figure 7 This invention is compared with the estimation performance of single-core GPR and dual-core GPR. Detailed Implementation

[0030] like Figure 1 As shown, the lithium battery health status estimation method includes the following steps:

[0031] Step S1: Obtain N initial features based on the battery's charging voltage curve, charging capacity curve, IC curve, and temperature curve, and then obtain six statistical features based on these N initial features: mean feature F. A Median characteristics F M Lower quartile characteristics F LQ Upper quartile characteristics F UQ Range characteristic F R Standard deviation characteristics F SD ;

[0032] The datasets used in this embodiment include the NASA dataset and the Oxford dataset. For the NASA dataset, the experimental dataset used comes from NASA's Ames Center of Excellence for Prediction, and the three batteries (labeled B0005, B0006, and B0007) are second-generation 18650 batteries manufactured by Idaho National Laboratory. All batteries were repeatedly tested at room temperature (24 degrees Celsius) using three different operating modes: constant current and constant voltage charge (CC-CV) mode, constant current (CC) discharge, and impedance measurement mode. The CC charging current was 1.5A, the upper voltage limit was 4.2V, the CV charging voltage was 4.2V, and CV ended when the current dropped to 50mA. CC discharge and CV charging were continuously repeated. The experiment was stopped when the batteries reached the end of their lifespan.

[0033] For the Oxford dataset, eight KoKam lithium-ion batteries (Cell1-Cell8) with a nominal capacity of 740 mAh were tested in a 40°C hot chamber for the Oxford battery degradation dataset. These batteries used graphite as the negative electrode and LiMO2 as the positive electrode. Battery capacity was measured every 100 aging cycles to obtain the true state of equilibrium (SOH). All batteries were charged at a constant current (CC) of 740 mA until the voltage reached 4.2 V, then depleted at a rate of 740 mA until the voltage reached 2.7 V. Voltage, current, surface temperature, and capacity values ​​were recorded. In this embodiment, three batteries (Cell1-Cell3) were selected for subsequent experiments.

[0034] For the extraction of initial features:

[0035] In this embodiment, battery health status is defined in terms of capacity decay: Among them, C i This represents the maximum capacity in this loop; C o Indicates the initial capacity.

[0036] When obtaining initial features from the charging voltage curve, the interval of the charging voltage curve containing the voltage inflection point is selected, and values ​​are taken within this interval at the same voltage interval to obtain a certain amount of time data as the initial features. Specifically:

[0037] The charging process of a lithium battery includes two stages: CC charging and CV charging. The lithium battery first charges in CC mode. When the voltage reaches 4.2V, it switches to CV charging mode, where the current gradually decreases until the battery is fully charged and charging stops. Figure 2 In the charging voltage variation curve shown, the horizontal axis represents time and the vertical axis represents voltage. It can be seen that when the voltage is in the 3.8V-4.1V range, there is an inflection point in the voltage variation curve. Before this inflection point, the curve slope is relatively large, and the time difference for the same voltage is small. After this inflection point, the slope of the voltage curve begins to decrease, and the time difference for the same voltage gradually increases. When the voltage fluctuates significantly between 3.8V and 3.9V, the time difference corresponding to the same voltage is small. Therefore, the charging voltage curve between 3.9V and 4.1V is selected for initial feature extraction. The same voltage interval used is 50mV, that is, the time difference calculated for every 50mV between 3.9V and 4.1V is used as the initial feature corresponding to the charging voltage. These initial features are denoted as HI1, HI2, HI3, and HI4, respectively.

[0038] When obtaining initial features from the charging capacity curve, the interval of the charging capacity curve containing the voltage inflection point is selected, and several capacity data points are obtained within this interval at the same voltage interval as initial features. Specifically:

[0039] As batteries age, their capacity gradually decreases at the same voltage. For example... Figure 3 In the charging capacity change curve shown, the horizontal axis represents voltage and the vertical axis represents capacitance. It can be seen that the charging capacity curve has an inflection point near the voltage value of 4V. Therefore, the capacitance value in the range of 3.9V-4.1V is selected. Within this range, the capacitance value corresponding to every 50mV voltage interval is taken as the initial characteristic of the capacitance. These initial characteristics are denoted as HI5, HI6, HI7, HI8, and HI9 respectively.

[0040] When obtaining initial features from the IC curve, the interval of the IC curve containing the voltage peak is selected, and several IC data points are obtained within this interval at the same voltage interval as initial features. Specifically:

[0041] like Figure 4In the IC curve graph shown, the horizontal axis represents voltage and the vertical axis represents capacity increment. It can be seen that the two peaks of the IC curve are between 3.8-4.1V. However, as the battery ages, the first peak gradually disappears, and the second peak gradually decreases. The changes in the IC curve are most significant between 3.8-4.1V. Within this range, as the battery ages, the IC value at the same voltage gradually decreases. However, in the voltage ranges close to 3.8V and 4.1V, the difference in IC values ​​between different cycles at the same voltage is not as significant as near the second peak. Considering that the difference in IC values ​​is not significant in these two segments, we selected the IC values ​​corresponding to every 50mV voltage interval within the 3.8-4.1V voltage range as the initial characteristics of the IC curve, denoted as HI. 10 HI 11 HI 12 HI 13 HI 14 HI 15 HI 16 HI 17 .

[0042] When obtaining initial characteristics from temperature curves, according to the formula... Calculate the temperature change curve corresponding to the temperature curve, and obtain several temperature change data points from the temperature change curve as initial features. Here, T(k) is the temperature value corresponding to voltage k, and Δ is the sampling interval, which determines the time interval between each sampling point. Specifically:

[0043] like Figure 5 In the temperature curve graph shown, the horizontal axis represents voltage and the vertical axis represents temperature. Figure 6 In the temperature change curve shown, the horizontal axis represents voltage, and the vertical axis represents the temperature change. For the sampling interval Δ, if it is too large, it may be unable to capture short-term temperature changes; if it is too small, it may be affected by noise, resulting in greater errors. In this embodiment, the sampling interval is 30mV. Figure 6 The temperature change curves were obtained, and the temperature change values ​​corresponding to voltages of 4.08V, 4.1V, and 4.12V were selected as initial features. These initial features are denoted as HI. 18 HI 19 HI 20 HI 21 .

[0044] After collecting N=21 initial features, in order to obtain higher estimation accuracy, these initial features are processed as follows:

[0045] According to the formula Obtain the mean feature F A The mean characteristic can reflect the central tendency of a set of data; according to the formula Obtain the median feature F M The median is a characteristic that can represent the general level of a set of data; according to the formula Obtain the lower quartile feature F LQ The lower quartiles can represent the degree of dispersion of a set of data; according to the formula The upper quartile feature F is obtained UQ The upper quartiles can represent the degree of dispersion of a set of data; according to the formula F... R =HI N -HI1 yields the range characteristic F R The range can represent the magnitude of a set of data; according to the formula Obtain the standard deviation characteristic F SD Standard deviation represents the degree of fluctuation in a set of data; where HI i Let i represent the i-th initial feature, where i = 1, 2, ..., N.

[0046] Table 1 shows the correlation coefficients between the six statistical features and SOH. As can be seen from Table 1, the Pearson correlation coefficient between each statistical feature and SOH is above 0.95, indicating that each statistical feature has good predictive ability and importance.

[0047] Table 1

[0048]

[0049]

[0050] Step S2: Construct a single-core Gaussian process regression model. The six statistical features obtained in step S1 are permuted and combined to obtain multiple feature combinations. These multiple feature combinations are used to train the dual-core Gaussian process regression model. The feature combination includes one or more statistical features. The single-core Gaussian process regression model is a conventional model, and its specific parameters can be recorded using existing parameters.

[0051] Specifically:

[0052] Gaussian process regression (GPR) is a machine learning method used to solve problems involving high dimensionality, small sample sizes, and nonlinearity. For regression problems, the input feature x and output y can be modeled as an implicit function plus independent Gaussian noise, as shown in the following equation: y = f(x) + ε, ε ~ N(0, δ) 2 ), where ε is Gaussian noise, and x is a feature combination consisting of one or more statistical features. For any input, the GPR is completely composed of its mean function m(x) and covariance function K(x). i ,x jThe probability distribution function f(x) is determined by the kernel function and the corresponding probability distribution function f(x), which follows a Gaussian distribution: Normally, the mean m(x) is zero.

[0053] The feature combinations obtained by arranging six statistical features are as follows: feature combinations containing only one statistical feature, feature combinations containing two different statistical features, feature combinations containing three different statistical features, feature combinations containing four different statistical features, feature combinations containing five different statistical features, and feature combinations containing six statistical features. For example, a feature combination containing only one statistical feature is: (F A ), (F M ), (F LQ ), (F UQ ), (F R ), (F SD ); including feature combinations (F) with two different statistical characteristics A F M ), (F A F LQ ), (F A F UQ ), (F A F R ), (F A F SD ), (F M F LQ ), (F M F UQ ), (F M F R ), (F M F SD ), (F LQ F UQ ), (F LQ F R ), (F LQ F SD ), (F UQ F R ), (F UQ F SD ), (F R F SD ).

[0054] After constructing the single-core GPR, the single-core GPR is trained using the above feature combination.

[0055] Step S3: Estimate the health status of the lithium battery using the trained single-core Gaussian process regression model and the feature combinations from step S2, thereby determining the optimal feature combination.

[0056] The estimation results are evaluated using squared absolute error (MAE) and mean squared error (RMSE), and the feature combination corresponding to the minimum squared absolute error (MAE) and mean squared error (RMSE) is selected as the optimal feature combination.

[0057] Step S4: Construct a dual-core Gaussian process regression model, use the Northern Eagle algorithm to search and optimize the hyperparameters of the dual-core Gaussian process regression model, and input the obtained optimal hyperparameters into the dual-core Gaussian process regression model to obtain the final prediction model.

[0058] The kernel function of the dual-core GPR in this embodiment is the composite kernel function K(x). i ,x j The composite kernel function K(x) i ,x j ) = Ma3 + Ma5, where, This Ma3 is able to capture smooth, short-term changes. The Ma5 function can capture a smooth long-term downward trend, therefore this composite kernel function can effectively track the long-term downward trend and short-term changes in battery capacity. In the formula, δ... f1 δ represents the signal variance of the kernel function Ma3. l1 δ represents the length scale of the kernel function Ma3. f2 δ represents the signal variance of the kernel function Ma5. l2 x represents the length scale of the kernel function Ma5. i Represents the observed data, x j This represents the input feature vector.

[0059] If m(x) is 0, then the prior distribution of the observations considering noise can be expressed as: Among them, I N It is an N-dimensional identity matrix. This represents the noise covariance matrix.

[0060] The joint probability distribution of the observed and predicted values ​​is represented as follows:

[0061] Where f* represents the estimated value of y, and x* represents the test dataset.

[0062] The prediction of GPR can be expressed as:

[0063]

[0064] in, Let cov(f*) be the predicted mean vector, and Θ2 = [δ] be the predicted variance. f1 ,δ l1 ,δ f2 ,δl2 ,δ N ] is the hyperparameter set.

[0065] For the Northern Eagle algorithm:

[0066] Initialization phase:

[0067] In the Northern Goshawk optimization algorithm, the Northern Goshawk population can be represented by the following population matrix: In the formula, X is the population matrix of the Northern Goshawk; X i It is the position of the i-th Northern Goshawk; x i,j Let be the position of the i-th Northern Goshawk in the j-th dimension; N is the number of Northern Goshawks; m is the dimension of the problem to be solved.

[0068] The objective function value of the problem can be represented by an objective function value vector as follows: In the formula, M is the objective function vector of the Northern Goshawk population; M i Let be the objective function value of the i-th Northern Eagle.

[0069] Phase 1: Prey Identification (Exploration Phase)

[0070] In the first stage of its hunt, the northern goshawk randomly selects prey and swiftly attacks it. The selection of prey is random. This stage involves a global search of the search space to determine the optimal region. This stage can be represented by the following formula:

[0071] P i =X k k = 1, 2, ..., i-1, i+1, ... N

[0072]

[0073] In the formula: P i Let i be the location of the prey of the i-th northern eagle; is the objective function value of the location of the i-th Northern Goshawk's prey; k is a random number in the range [1, N]. This is the new location of the i-th Northern Goshawk; It is the new position of the i-th Northern Eagle in the j-th dimension; is the objective function value of the i-th Northern Eagle after the first stage update; r is a random number in the range [0,1]; H is a random integer of 1 or 2.

[0074] Phase Two: Chase and Escape (Development Phase)

[0075] After a northern goshawk attacks its prey, the prey will attempt to escape. Therefore, during the prey-chasing phase, the northern goshawk needs to continue pursuing the prey. Because of their high speed, northern goshawks can chase prey in almost any situation and eventually successfully capture it. Simulating this behavior improves the algorithm's ability to perform local searches within the search space. This phase can be represented by the following formula:

[0076]

[0077] In the formula, t and T are the current iteration number and the maximum iteration number, respectively; R is the attack radius, which decreases as the number of iterations increases; Let be the value of the solution for the i-th eagle at its new position in the j-th dimension; This represents the objective function value for this stage.

[0078] Step S5: Estimate the health status of the lithium battery based on the prediction model obtained in step S4 and the optimal feature combination obtained in step S3.

[0079] Figure 7 This graph compares the estimation performance of the present invention with that of single GPR and dual-core GPR. The horizontal axis represents the number of cycles, and the vertical axis represents the health status. It can be seen that the estimation performance of the present invention is better and closer to the reference value. Furthermore, for MAE and RMSE, the present invention controls the MAE to within 0.6% and the RMSE to within 1% for all three battery cells in the NASA dataset, while the MAE and RMSE of single-core GPR and dual-core GPR are both within 1.5%.

[0080] The above description is merely a preferred embodiment of the present invention and should not be construed as limiting the scope of the present invention. All equivalent changes and modifications made in accordance with the scope of the patent application and the contents of the specification of the present invention should still fall within the scope of the patent of the present invention.

Claims

1. A method for estimating the health status of lithium batteries based on statistical feature analysis, characterized in that: Includes the following steps: Step S1: Obtain N initial features based on the battery's charging voltage curve, charging capacity curve, IC curve, and temperature curve, and then obtain six statistical features based on these N initial features: Mean characteristic F A Median characteristics F M Lower quartile characteristics F LQ Upper quartile characteristics F UQ Range characteristic F R Standard deviation characteristic F SD ; Step S2: Construct a single-kernel Gaussian process regression model. The six statistical features obtained in step S1 are permuted and combined to obtain multiple feature combinations. These multiple feature combinations are used to train the single-kernel Gaussian process regression model. The feature combination includes one or more statistical features. Step S3: Estimate the health status of the lithium battery using the trained single-core Gaussian process regression model and the feature combinations from step S2, thereby determining the optimal feature combination. Step S4: Construct a dual-core Gaussian process regression model, use the Northern Eagle algorithm to search and optimize the hyperparameters of the dual-core Gaussian process regression model, and input the obtained optimal hyperparameters into the dual-core Gaussian process regression model to obtain the final prediction model. Step S5: Estimate the health status of the lithium battery based on the prediction model obtained in step S4 and the optimal feature combination obtained in step S3. Furthermore, in step S1, the mean feature The median feature The lower quartile feature The upper quartile feature The range feature F R =HI N -HI1, the standard deviation characteristic Among them, HI i Let i represent the i-th initial feature, where i = 1, 2, ..., N; Furthermore, in step S2, the feature combinations obtained by arranging and combining the six statistical features are as follows: feature combination including only one statistical feature, feature combination including two different statistical features, feature combination including three different statistical features, feature combination including four different statistical features, feature combination including five different statistical features, and feature combination including six statistical features. Furthermore, in step S4, the kernel function of the constructed dual-kernel Gaussian process regression model is K = Ma3 + Ma5, where, In the formula, δ f1 δ represents the signal variance of the kernel function Ma3. l1 δ represents the length scale of the kernel function Ma3. f2 δ represents the signal variance of the kernel function Ma5. l2 x represents the length scale of the kernel function Ma5. i Represents the observed data, x j This represents the input feature vector.

2. The method for estimating the health status of a lithium battery based on statistical feature analysis according to claim 1, characterized in that: In step S1, when obtaining initial features based on the charging voltage curve, the interval of the charging voltage curve containing the voltage inflection point is selected, and values ​​are taken in this interval at the same voltage interval to obtain a number of time data as initial features.

3. The method for estimating the health status of a lithium battery based on statistical feature analysis according to claim 1, characterized in that: In step S1, when obtaining initial features based on the charging capacity curve, the interval of the charging capacity curve containing the voltage inflection point is selected, and several capacity data are obtained in this interval with the same voltage interval as initial features.

4. A method for estimating the health status of a lithium battery based on statistical feature analysis according to claim 1, 2, or 3, characterized in that: In step S1, when obtaining initial features based on the IC curve, the interval of the IC curve containing voltage peaks is selected, and several IC data are obtained in this interval with the same voltage interval as initial features.

5. A method for estimating the health status of a lithium battery based on statistical feature analysis according to claim 1, 2, or 3, characterized in that: In step S1, when obtaining the initial characteristics based on the temperature curve, the formula is used. Calculate the temperature change curve corresponding to the temperature curve, and obtain a number of temperature change data on the temperature change curve with the same voltage interval as the initial feature, where T(k) is the temperature value corresponding to voltage k, and Δ is the sampling interval.

Images