Automatic feature extraction method suitable for battery state of health assessment and life prediction

Abstract

The application discloses an automatic feature extraction method suitable for battery state of health (SOH) evaluation and life prediction, and comprises the following steps: S1, performing a cyclic charge-discharge aging test on the battery under a fixed temperature and different charge-discharge rates, and collecting current information and voltage information of the battery in the cyclic process; S2, obtaining the capacity corresponding to the cycle through ampere-hour integration under different charge-discharge rates, and establishing a capacity increment curve corresponding to the working condition; S3, performing Gaussian convolution on the IC curve under a one-dimensional time scale, and establishing a Gaussian space and a Gaussian difference space corresponding to the working condition; S4, constructing towers of different scales, calculating the Gaussian difference space of each tower, and finding local extreme points in adjacent difference spaces; and S5, processing the automatic feature database, selecting similar points as the same sequence, and performing correlation analysis on the battery SOH, and selecting a feature sequence with high correlation as the automatic feature. The method has low calculation complexity, and can directly obtain a feature point sequence with high correlation with the SOH.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of lithium ion battery state of health estimation, and particularly relates to an automatic feature extraction method suitable for battery state of health evaluation and life prediction. BACKGROUND

[0002] The state of health (SOH) estimation of a battery is an important research direction in the field of electric vehicles (EVs), and has a direct impact on the effectiveness of a battery management system (BMS) and the service life of an electric vehicle. At present, the battery SOH evaluation and life prediction based on increment capacity analysis (ICA) are widely researched and applied.

[0003] The increment capacity curve refers to the establishment of an increment curve by measuring the changes of the charging capacity and voltage of a battery over time. In the process of the decrease of the battery capacity with the increase of the cycle number and the use time, the morphology and characteristics of the increment capacity curve also change, and therefore the characteristics hidden in the increment capacity curve can be used to estimate the SOH and the remaining life of the battery.

[0004] The main feature extraction methods of the increment capacity (IC) curve at the present stage can be divided into the following kinds. The first kind is the IC curve analysis based on morphological characteristics, which uses a Gauss model, a Lorentz model, a pseudo-Voigt Peak model, etc. to fit the morphological characteristics of the IC curve of the battery charging process, and reconstructs the IC curve. This method has high precision, but is very susceptible to noise, requires a good initial value of the fitting parameter, and the range of parameter identification will directly affect the accuracy of the characteristics, and the accuracy lacks mathematical verification.

[0005] The second kind is the model-based method, which constructs an increment capacity curve containing corresponding parameters through an electrical model, such as an equivalent circuit model or a Davies model, and then identifies the parameters of each model feature. This method has high robustness, but the accuracy of parameter identification is very sensitive, and the initial value needs to be adjusted actively, and complete charging data is required, and if the data is missing, the accuracy and precision will be greatly affected.

[0006] The third method is a data-driven method, which extracts the characteristic parameters implied in the capacity increment curve, directly extracts the corresponding health factor, and then estimates the SOH and the remaining life of the battery. This method has high precision, but requires good data sets and manual screening and adjustment to obtain better precision and accuracy. SUMMARY

[0007] The present application aims to overcome the shortcomings of the prior art and provide an automatic feature extraction method suitable for battery health state evaluation and life prediction,

[0008] The purpose of the present application is achieved by the following technical scheme: an automatic feature extraction method suitable for battery health state evaluation and life prediction, comprising the following steps:

[0009] S1. Perform cyclic charging and discharging aging test on the battery at a fixed temperature and different charge and discharge rates, and collect current information and voltage information during the battery cycle;

[0010] The step S1 comprises the following sub-steps:

[0011] S101. At a fixed temperature, charge at 1 / 3C rate to cut-off current in constant current and constant voltage mode, stand for 30 minutes, and discharge at 1 / 3C rate to cut-off voltage in constant current mode;

[0012] S102. Charge the battery at the required rate in CCCV mode to the cut-off current, i.e. charge to the cut-off voltage in constant current mode, and then charge in constant voltage mode to the cut-off current;

[0013] S104. Repeat S102-S103 until the capacity of the battery decays to the specified life cutoff point, i.e. 70% of the initial rated capacity; organize the current information and voltage information during the cycle to establish the battery aging cycle data set under the corresponding charge and discharge rate;

[0014] S105. Change the charge and discharge rate, repeat S101-S104, and obtain the battery aging cycle database under different charge and discharge rates at a fixed temperature.

[0015] S2. Under different charge and discharge rates, obtain the capacity corresponding to the cycle by ampere-hour integration method, and then establish the capacity increment curve corresponding to the working condition;

[0016] The step S2 comprises the following sub-steps:

[0017] S201. Select the battery aging cycle data set under the same working condition from the battery aging cycle library, and obtain the charge and discharge capacity Q of each cycle by using the ampere-hour integration method CCCV , and the charge and discharge capacity Q(t0) at each time;

[0018] S202. Differentiate the capacity at each time to obtain the capacity increment curve thereof;

[0019] S203. Repeat S201-S202 to obtain the capacity information Q(t) at each time under different discharge rates, and then analyze the capacity increment curve under each rate to obtain the capacity increment curve under each working condition.

[0020] The capacity differentiation in step S202 uses the average differentiation of the previous point and the next point of the current capacity point as the capacity differentiation of the current point.

[0021] S3. Perform Gaussian convolution on the IC curve under the one-dimensional time scale to establish the Gaussian space and the Gaussian difference space corresponding to the working condition;

[0022] The step S3 includes the following sub-steps:

[0023] S301. Select the IC curve under the same working condition to establish the corresponding Gaussian convolution kernel;

[0024] S302. Perform Gaussian filtering and smoothing on the IC curve to obtain the Gaussian space layer corresponding to the Gaussian convolution kernel;

[0025] S303. Update the Gaussian convolution kernel;

[0026] S304. Repeat S301-S303 to update the Gaussian convolution kernel each time to obtain the Gaussian space corresponding to the corresponding cycle under the corresponding charge and discharge rate;

[0027] S305. Repeat S301-S304 in different cycles under the same working condition to establish the Gaussian space of different cycles;

[0028] S306. Repeat S301-S305 to obtain the Gaussian space layer of different cycles under different working conditions, and construct the Gaussian space of each working condition.

[0029] S4. Construct towers of different scales, calculate the Gaussian difference space of each tower, and find the local extreme points in the adjacent difference spaces;

[0030] The step S4 includes the following sub-steps:

[0031] S401. Differentiate the Gaussian space of the same cycle under the same working condition to obtain the Gaussian difference space of the corresponding cycle;

[0032] S402. Repeat S401 to obtain the DOG space of different working conditions under the IC curve, and combine to construct the tower under the scale;

[0033] S403. After down-sampling the IC curve, repeat S401-S402 to build the tower of the next scale;

[0034] S404. Find the local extreme points in the adjacent DOG space of the same cycle in the same tower, and take them as the candidate points for estimating the battery state of health and evaluating the remaining life, and build an automatic feature database.

[0035] Step S403 builds towers of different scales, and the number of towers is set to three, that is, after twice down-sampling of the IC curve, Gaussian space and Gaussian difference space are established respectively.

[0036] S5. Process the automatic feature database, select similar points as the same sequence, and perform correlation analysis with the battery SOH to select the feature sequence with high correlation as the automatic feature.

[0037] The step S5 includes the following sub-steps:

[0038] S501. Determine the weight matrix, select the point with the minimum weight difference from the initial point value in the cycle as the same sequence, perform weight sequence analysis on the feature points in the initial cycle, and obtain different similar point sequences;

[0039] S502. Perform correlation analysis on each similar point sequence and the battery SOH, and take the top three correlations as the automatically extracted features.

[0040] The present application has the advantages that the present application provides an automatic feature extraction method suitable for battery state of health evaluation and life prediction. By establishing a charge and discharge database under different working conditions, Gaussian space and Gaussian difference space are established for the capacity increment curve, similar feature sequences in the difference space are taken as automatic features, and features with high correlation with SOH are selected as features for evaluating SOH and remaining life. The present application has two advantages, first, the sequence feature with high correlation with SOH can be directly obtained, without the need for parameter identification using complex circuit models or morphological models. Second, the present application has strong applicability and can be used for analysis of different batteries under different working conditions, without the need for adjusting model parameters, avoiding the case of being effective only for a single working condition. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 The method flowchart of the present application;

[0042] Figure 2 The battery capacity increment curve schematic diagram of the embodiment;

[0043] Figure 3 The Gaussian space curve diagram of the battery IC curve of the embodiment;

[0044] Figure 4The battery IC curves are shown in the Gaussian difference space curves of each layer in the embodiment.

[0045] Figure 5 This is a schematic diagram of the local extreme points of the battery IC curve under the same cycle in the embodiment.

[0046] Figure 6 This is a schematic diagram illustrating the automatic feature point extraction of the battery under different cycles in an embodiment.

[0047] Figure 7(a) is a voltage-cycle diagram of the characteristic sequence points in the embodiment;

[0048] Figure 7(b) is a schematic diagram of the IC curve values ​​of the feature sequence points in the embodiment.

[0049] Figure 7(c) is a schematic diagram of the confidence threshold-loop of the feature sequence points in the embodiment. Detailed Implementation

[0050] The following is in conjunction with the appendix Figure 1 The technical solution of the present invention will be described in further detail below, but the scope of protection of the present invention is not limited to the following description.

[0051] Automatic feature extraction methods suitable for battery health status assessment and lifespan prediction, such as Figure 1 As shown, it includes the following steps:

[0052] S1. Conduct cyclic charge-discharge aging tests on the battery at a fixed temperature and different charge-discharge rates, and collect current and voltage information during the battery cycle.

[0053] S101. Place the lithium-ion battery in a constant temperature chamber and charge it to the cutoff current at a constant current and constant voltage condition at a 1 / 3C rate. Let it stand for 30 minutes and then discharge it to the cutoff voltage at a constant current condition at a 1 / 3C rate.

[0054] The temperature fluctuation range and the accuracy requirements for current and voltage acquisition in the test environment shall comply with the national standard GB / T 36276.

[0055] S102. Charge the battery under CCCV conditions to the cutoff current according to the required rate, that is, charge under constant current conditions to the charging cutoff voltage, and then charge under constant voltage conditions to the charging cutoff current.

[0056] S103. Discharge the battery at a constant current until the discharge cutoff voltage is reached, according to the required rate.

[0057] S104. Repeat S102-S103 until the battery capacity decays to the specified end-of-life point, i.e., 70% of the initial rated capacity. Organize the current and voltage information during the cycle to establish a battery aging cycle dataset under the corresponding charge / discharge rates.

[0058] S105. Change the charge / discharge rate, repeat S101-S104, and obtain a database of battery aging cycles at different charge / discharge rates under a fixed temperature.

[0059] S2. Under different charge and discharge rates, the capacity of the corresponding cycle is obtained by the ampere-hour integral method, and then the capacity increment curve of the corresponding operating condition is established.

[0060] S201. Select battery aging cycle datasets under the same operating condition from the battery aging cycle library, and obtain the charge / discharge capacity Q of each cycle using the time-integration method. CCCV And the charge / discharge capacity Q(t0) at each moment, the specific calculation method is as follows:

[0061]

[0062]

[0063] Where n is the time from the constant current and constant voltage charging condition to the charging cutoff current, I(t) is the current value at time t, Δt is the load current sampling time interval, t0 is the value within the charging time range, and t0∈[0,n];

[0064] S202. Differentiate the capacity at each time step to obtain its capacity increment curve. The specific formula is as follows:

[0065]

[0066] Where V(t) represents the voltage value at time t, Q(t) represents the capacity value at time t, and m is the time when the constant current charging condition reaches the cutoff voltage.

[0067] Taking LIB#35 as an example, its operating conditions are 0.5C rate CCCV charging and 100% DOD. The capacity increment curve is as follows: Figure 2 As shown, the most obvious IC peak positions are around 3.8V, 3.9V, and 4.0V.

[0068] S203. Repeat S201-S202 to obtain the capacity information Q(t) at each moment under different discharge rates, and then analyze the capacity increment curves under each rate to obtain the capacity increment curves under each operating condition.

[0069] S3. Perform Gaussian convolution on the IC curve under a one-dimensional time scale to establish the Gaussian space and the difference of Gaussians under the corresponding working conditions;

[0070] S301. Select the IC curve under the same working conditions and establish the corresponding Gaussian convolution kernel. The specific calculation method is as follows:

[0071] The IC curve is convolved using a Gaussian function, as shown below:

[0072]

[0073] Where h represents the convolution interval, set here to h = [1, 10], σ represents the standard deviation of the Gaussian filter function, indicating the smoothness of the curve, and G(h, σ) represents a scale-variable Gaussian function. Specifically, large scales correspond to the overall features of the image, while small scales correspond to the detailed features of the image.

[0074] S302. Smooth the IC curve using Gaussian filtering to obtain the Gaussian spatial layer corresponding to the Gaussian convolution kernel; the specific calculation method is shown below:

[0075] L(IC,σ)=G(h,σ)×I(IC)

[0076] Where L(IC,σ) represents the smoothed IC curve, and I(IC) represents the IC curve before smoothing.

[0077] This yields the Gaussian spatial layer corresponding to the Gaussian convolution kernel.

[0078] S303. Update the Gaussian convolution kernel, defining the kernel standard deviation as:

[0079] σ=(σ,kσ,k 2 σ,…k n-1 σ)

[0080] In this example, k is defined as k = 2 1 / 3 , where n represents the number of layers in the Gaussian space, which is defined here as: n = 6.

[0081] S304. Repeat S301-S303, updating the Gaussian convolution kernel each time to obtain the corresponding Gaussian space for the corresponding charge / discharge rate.

[0082] Taking the second cycle of the 0.5C 100% DOD charge-discharge cycle in this embodiment as an example, the constructed Gaussian space is as follows: Figure 3 As shown, the curve gradually becomes smoother as the Gaussian convolution kernel size increases. Six Gaussian space layers are constructed in this loop, and their combination forms the Gaussian space of the second loop.

[0083] S305. Repeat S301-S304 under the same working condition but different cycles to establish Gaussian spaces for different cycles;

[0084] S306. Repeat S301-S305 to obtain Gaussian space layers for different cycles under different working conditions, and construct Gaussian spaces for each working condition.

[0085] S4. Construct octaves of different scales, calculate the Gaussian difference space for each octave, and find local extrema in adjacent difference spaces.

[0086] S401. Perform finite difference on the Gaussian space of the same cycle under the same working condition to obtain the Gaussian difference space of the corresponding cycle. The specific calculation process is as follows:

[0087] D i (IC,σ i )=(G(h,kσ i )-G(h,σ i ))×I(IC)

[0088] =L(IC,kσ) i )-L(IC,σ i )

[0089] Where D(IC,σ) represents a DOG layer, L(IC,kσ) represents a Gaussian layer with a kσ kernel, and i represents the number of DOG layers, where i = 1, 2, ..., m. Since DOG is derived from Gaussian spatial differences, the number of DOG layers is the number of Gaussian layers minus one, i.e., m = n - 1 = 5.

[0090] Taking the second cycle of the 0.5C 100% DOD charge-discharge cycle in this embodiment as an example, the constructed Gaussian difference space is as follows: Figure 4 As shown, there are very obvious fluctuations around 3.8V, 3.9V, and 4.0V in its DOG space.

[0091] S402. Repeat S401 to obtain the DOG space for different operating conditions under this IC curve, and combine them to construct a tower at this scale; the specific calculation process is as follows:

[0092] Octave num ={DOG cycle,i}

[0093] Where num represents the scale of the tower, cycle represents the cycle number of the DOG space, and i represents the number of DOG space layers under that cycle number.

[0094] S403. After downsampling the IC curve, repeat S401-S402 to construct the tower at the next level. The specific calculation method is as follows:

[0095] The IC curve is downsampled, with the sampling frequency reduced by a factor of 2, specifically as follows:

[0096] IC down =IC(t), t=0,2,…2m

[0097] Among them, ICdown This represents the IC curve after downsampling, where IC represents the IC curve before downsampling, and t represents the sampling time.

[0098] S404. Local extrema are found in the adjacent DOG space of the same cycle within the same tower. These extrema are used as candidate points for estimating battery health status and assessing remaining lifetime, thus establishing an automatic feature database. The specific calculations are as follows:

[0099] We select the three middle layers of the DOG space as the target and check whether they are the same layer and the local extrema of the DOG space layers above and below:

[0100]

[0101] Where k represents the number of DOG layers, and k = 2, 3, 4, and t represents the sampling time.

[0102] Taking the second cycle of the 0.5C 100% DOD charge-discharge cycle in this embodiment as an example, the local extreme point obtained is as follows: Figure 5 As shown.

[0103] S5. Process the automatic feature database, select similar points as the same sequence, and perform correlation analysis with the battery SOH to select feature sequences with high correlation as automatic features.

[0104] S501. Determine the weight matrix, select the point with the smallest weight difference from the initial point value in the loop as the same sequence, perform weight sequence analysis on the feature points in the initial loop to obtain different similar point sequences. The specific calculation process is as follows:

[0105] First, for the feature points in the initial loop, detect the feature points in all subsequent loops, and match the point with the minimum weight error as the similar point for that loop. That is:

[0106] W*(P cycle,lag -P 0,i )=minW*(P cycle,all -P 0,i )

[0107] S i =[P 0,i ,P cycle,lag ]

[0108] Where cycle represents the cycle number, P represents the feature point, W represents the weight matrix, and S i A sequence representing feature points. P cycle,lag This indicates the relationship between the corresponding loop and the initial point P. 0,i The most similar points.

[0109] Taking LIB#35 under 0.5C 100% DOD conditions as an example, the similarity point sequence in this embodiment is as follows: Figure 6 As shown, its sequence represents the starting position of the IC peak at 3.8V.

[0110] S502. Perform correlation analysis on each similarity sequence and the battery SOH, and take the top three correlations as automatically extracted features.

[0111] First, the correlation coefficient between similarity points and battery SOH is defined as follows:

[0112]

[0113] Where ρ is the sample correlation coefficient, μ is the average value of the similar point sequence and the battery cycle SOH, and σ represents the standard deviation of the similar point sequence and the battery cycle SOH.

[0114] Following the steps above, a corresponding Gaussian difference space is constructed based on the Gaussian space model. A feature library is built by selecting local extreme points, and the hidden features in the IC curve are automatically extracted. Based on the battery's SOH correlation coefficient, a feature sequence that is close to SOH is automatically selected as a health factor for evaluating the battery's SOH and remaining life.

[0115] In the embodiments of this patent, a lithium iron phosphate battery (LiFePO4, LFP) with a nominal capacity of 1.1Ah is used as the experimental object. The method steps S1 of this invention are carried out under fixed temperature conditions. Cyclic charge-discharge aging tests are conducted on the battery at fixed temperature and different charge-discharge rates to collect current and voltage information during the battery cycle. Then, at different charge-discharge rates, the capacity of the corresponding cycle is obtained by ampere-hour integration method. Subsequently, the capacity increment curve of the corresponding operating condition is established, and Gaussian convolution is performed on the IC curve under one-dimensional time scale to establish Gaussian space and Gaussian difference space under the corresponding operating condition. Then, towers of different scales are constructed, and the Gaussian difference space of each tower is calculated. Local extreme points in adjacent difference spaces are found. Finally, the automatic feature database is processed, similar points are selected as the same sequence, and correlation analysis is performed with the battery SOH. Feature sequences with high correlation are selected to evaluate the battery SOH and remaining life. The upper limit cutoff voltage for charging during the CC phase of the battery under test is 4.2V, and the lower limit cutoff current for charging during the CV phase is 0.05A. During the discharge test, the discharge current during the CC phase is at the specified discharge rate, and the cutoff voltage for the discharge phase is 2.7V. If the CC phase data is normal, this method can be used to analyze the CC phase and estimate the battery's SOH.

[0116] Specifically, taking battery LIB#35 under the 0.5C100DOD condition as an example, the voltage value of its most correlated feature sequence is shown in Figure 7(a), the peak value of dQ / dV is shown in Figure 7(b), and its confidence threshold represents the relative error from the initial point, as shown in Figure 7(c). The confidence threshold gradually increases with the number of cycles and remains relatively stable before the capacity drops significantly. Among the automatically extracted feature sequences, the top three correlation coefficients with SOH are 0.8110, 0.8025, and 0.7600, respectively. The high correlation allows for a relatively effective assessment of battery SOH and remaining lifespan. Therefore, this invention can estimate battery SOH and lifespan by relying on the capacity increment curve and obtaining its implicit features through the Gaussian difference space for different operating conditions and batteries. It can also handle special situations well, while meeting accuracy requirements. It has low computational complexity, low data requirements, wide applicability, and strong noise resistance.

[0117] The above description represents preferred embodiments of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technical or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. An automatic feature extraction method suitable for battery health status assessment and lifespan prediction, characterized in that, Includes the following steps: S1. Conduct cyclic charge-discharge aging tests on the battery at a fixed temperature and different charge-discharge rates, and collect current and voltage information during the battery cycle process; S2. Under different charge and discharge rates, the capacity of the corresponding cycle is obtained by the ampere-hour integration method, and then the capacity increment curve of the corresponding operating condition is established; S3. Perform Gaussian convolution on the IC curve under a one-dimensional time scale to establish the Gaussian space and Gaussian difference space under the corresponding working conditions; S4. Construct towers of different scales, calculate the Gaussian difference space for each tower, and find local extrema in adjacent difference spaces; Includes the following sub-steps: S401. Difference is performed on the Gaussian space of the same cycle under the same working condition to obtain the Gaussian difference space of the corresponding cycle; S402. Repeat S401 to obtain the DOG space under different working conditions under the IC curve, and combine them to construct a tower at this scale; S403. After downsampling the IC curve, repeat S401-S402 to construct the tower at the next level. S404. Find local extrema in the adjacent DOG space of the same cycle in the same tower, and use them as candidate points to estimate battery health status and assess remaining life, and establish an automatic feature database. S5. Process the automatic feature database, select similar points as the same sequence, and perform correlation analysis with the battery SOH to select feature sequences with high correlation as automatic features; Includes the following sub-steps: S501. Determine the weight matrix, select the point with the smallest weight difference from the initial point value in the loop as the same sequence, perform weight sequence analysis on the feature points in the initial loop to obtain different similar point sequences; S502. Perform correlation analysis on each similarity sequence and the battery SOH, and take the top three correlations as automatically extracted features.

2. The automatic feature extraction method for battery health status assessment and lifespan prediction according to claim 1, characterized in that, Step S1 includes the following sub-steps: S101. Under a fixed temperature, charge the battery to the cutoff current at a constant current and constant voltage at a 1 / 3C rate, let it stand for 30 minutes, and discharge it to the cutoff voltage at a constant current at a 1 / 3C rate. S102. Charge the battery under CCCV conditions to the cutoff current according to the required rate, that is, charge under constant current conditions to the charging cutoff voltage, and then charge under constant voltage conditions to the charging cutoff current. S104. Repeat S102-S103 until the battery capacity decays to the specified end of its lifespan, i.e., 70% of the initial rated capacity; organize the current and voltage information during the cycle and establish a battery aging cycle dataset under the corresponding charge and discharge rates. S105. Change the charge / discharge rate, repeat S101-S104, and obtain a database of battery aging cycles at different charge / discharge rates under a fixed temperature.

3. The automatic feature extraction method for battery health status assessment and lifespan prediction according to claim 1, characterized in that, Step S2 includes the following sub-steps: S201. Select battery aging cycle datasets under the same operating condition from the battery aging cycle database, and obtain the charge and discharge capacity of each cycle using the ampere-hour integration method. And the charge / discharge capacity at each moment. ; S202. Differentiate the capacity at each time point to obtain its capacity increment curve; S203. Repeat S201-S202 to obtain capacity information at each moment under different discharge rates. Then, the capacity increment curves at each multiplier are analyzed to obtain the capacity increment curves under each operating condition.

4. The automatic feature extraction method for battery health status assessment and lifespan prediction according to claim 3, characterized in that, The capacity described in step S202 is differentiated, and the average differential between the previous and next points of the current capacity point is used as the capacity differential of the current point.

5. The automatic feature extraction method for battery health status assessment and lifespan prediction according to claim 1, characterized in that, Step S3 includes the following sub-steps: S301. Select the IC curve under the same working conditions and establish the corresponding Gaussian convolution kernel; S302. Smooth the IC curve using Gaussian filtering to obtain the Gaussian spatial layer corresponding to the Gaussian convolution kernel; S303. Update the Gaussian convolution kernel; S304. Repeat S301-S303, updating the Gaussian convolution kernel each time to obtain the corresponding Gaussian space for the corresponding charge / discharge rate; S305. Repeat S301-S304 under the same working condition but different cycles to establish Gaussian spaces for different cycles; S306. Repeat S301-S305 to obtain Gaussian space layers for different cycles under different working conditions, and construct Gaussian spaces for each working condition.

6. The automatic feature extraction method for battery health status assessment and lifespan prediction according to claim 1, characterized in that, Step S403 involves constructing towers of different scales, with the number of towers set to three. This means that after downsampling the IC curve twice, Gaussian space and Gaussian difference space are established respectively.

Images