Power battery health scoring method and system based on multi-feature fusion

Abstract

The invention discloses a power battery health scoring method and system based on multi-feature fusion. The method comprises the following steps: preprocessing and classifying obtained power battery operation data; constructing a consistency evaluation index system; determining a consistency evaluation index weight based on a combination weighting method of the game theory; calculating a total evaluation score of the power battery by combining each consistency evaluation index and the weight and the threshold value thereof; obtaining each data set and carrying out feature extraction, screening and construction to obtain an optimal feature set of each power battery; the optimal feature set is divided into a training set, a verification set and an evaluation set to be used for training a BP neural network model and a LightGRM model respectively, then evaluation weights of the two models are calculated, and finally a battery health evaluation score is calculated. According to the hybrid modeling method based on the lightweight space-time diagram convolutional network, the precision and robustness of battery health state estimation are remarkably improved by fusing physical mechanism constraints and data driving advantages.

Description

Technical Field

[0001] The present invention relates to a power battery health scoring method in the field of new energy technology, and in particular to a power battery health scoring method based on multi-feature fusion, and also to a power battery health scoring system based on multi-feature fusion. Background Art

[0002] With the development and utilization of new energy technologies, lithium-ion batteries have become a high-quality power source for new energy vehicles due to their superior high energy density, ultra-long cycle life, low self-discharge characteristics, and excellent environmental adaptability. However, in actual use, battery performance gradually degrades due to the combined effects of internal electrochemical side reactions, complex operating conditions, and the environment. When the battery capacity retention rate falls below the 80% threshold or the internal resistance increases by more than 30%, its energy output capacity will show a cliff-like decline. More seriously, localized overcharge / over-discharge may trigger a thermal runaway chain reaction, leading to uncontrollable safety accidents. Therefore, evaluating the reliability and safety of lithium batteries is extremely important.

[0003] Currently, research on lithium-ion battery state of health (SOH) estimation and prediction has gradually shifted from traditional model-based methods to data-driven and hybrid approaches, incorporating advanced algorithms to improve accuracy and adaptability. Traditional model-based methods achieve parameter identification through physical mechanism modeling, but suffer from high computational complexity and lack of real-time performance. Data-driven methods have become mainstream due to their nonlinear fitting capabilities. However, while data-driven methods (such as deep learning models) can capture complex nonlinear relationships, they face challenges such as insufficient feature extraction quality, weak temporal correlation, and poor interpretability. Therefore, existing power battery health scoring methods suffer from low accuracy and poor robustness. Summary of the Invention

[0004] In order to solve the technical problems of low precision and poor robustness of existing power battery health scoring methods, the present invention provides a power battery health scoring method and system based on multi-feature fusion.

[0005] The present invention is implemented by the following technical solution: a power battery health scoring method based on multi-feature fusion, which includes the following steps: S1: preprocessing and classifying the obtained power battery operation data to obtain charging / resting / discharging data; S2: constructing a system of consistency evaluation indicators related to voltage, temperature, capacity, and internal resistance based on the charging / resting / discharging data; S3: determining the weights of the consistency evaluation indicators of the power battery in the charging / resting / discharging state based on a combined weighted method of game theory; S4: calculating the total evaluation score of the power battery by combining each consistency evaluation indicator and its weight and threshold; S5: obtaining each data set based on the charging / resting / discharging data and performing feature extraction, screening, and construction to obtain the optimal feature set of each power battery; S6: first dividing the optimal feature set into a training set, a validation set, and an evaluation set according to a preset ratio, and training the BP neural network model and the LightGRM model respectively according to the total evaluation score, and then calculating the evaluation weights of the two models, and finally calculating the final battery health evaluation score based on the evaluation weights and the scores of the power battery by the two models.

[0006] The present invention constructs a battery consistency evaluation index system after preprocessing and classifying the power battery operation data, and then scientifically determines the weight of each consistency index based on the combined weighted method of game theory. Then, the total evaluation score of the power battery is calculated by combining the weight and threshold of each consistency index, and then feature engineering is performed to finally obtain the optimal feature set. Finally, the BP neural network model and the LightGRM model are trained respectively using the optimal feature set. According to the evaluation weights of the two models and the total evaluation scores obtained by training each model, the final battery health evaluation score is obtained by weighted combination. In this way, the hybrid modeling method based on the lightweight spatiotemporal graph convolutional network significantly improves the accuracy and robustness of battery health status estimation by integrating physical mechanism constraints and data-driven advantages, and solves the technical problems of low accuracy and poor robustness of existing power battery health scoring methods.

[0007] As a further improvement of the above solution, the voltage-related consistency evaluation indicators include voltage range F1, voltage curve distance F2 and voltage variation coefficient F3; the temperature-related consistency evaluation indicators include temperature range F4, temperature curve distance F5 and temperature rise rate F6; the capacity-related consistency evaluation indicators include incremental capacity peak F7 and platform area charging capacity F8; the internal resistance-related consistency evaluation indicators include internal resistance variation coefficient F9;

[0008] Among them, the calculation formulas for each consistency evaluation index are:

[0009]

[0010] Where N represents the length of the specified charging data segment, V t,max Indicates the maximum battery cell voltage at time t, V t,min Indicates the minimum battery cell voltage at time t;

[0011]

[0012] Where B represents the number of battery cells in the battery pack. Indicates the voltage of battery cell No. b at time t;

[0013]

[0014] Where, represents the average voltage of all batteries at the tth moment of the charging process;

[0015]

[0016] Where, T t,max Indicates the maximum temperature value detected by the battery pack temperature sensor at time t, T t,min The minimum temperature value detected by the battery pack temperature sensor at time t;

[0017]

[0018] Where S represents the total number of battery pack temperature sensors. It represents the temperature detected by the s-th temperature sensor at the t-th time. Represents the average temperature of all temperature sensors at time t;

[0019]

[0020] Where, Respectively represent the temperature at the start and end of charging of the sth temperature sensor, Respectively indicate the charging start time and charging end time;

[0021]

[0022] In the formula, Q represents the quantity series, and U represents the voltage series;

[0023]

[0024] Where C0 represents the rated capacity of the battery cell, I i represents the current of the battery pack at the i-th moment, t start , t end Respectively represent the start time and end time of charging for a specified voltage segment;

[0025]

[0026] Where, Indicates the voltage of the bth battery cell at the last moment before the current drops during the charging process; Indicates the voltage of the bth battery cell at the first moment after the current drops during the charging process; I e Indicates the current value at the last moment before the current drops during the charging process; I s Indicates the current value at the first moment after the current drops during the charging process

[0027] Furthermore, the method for determining the weights of the consistency evaluation indicators includes the following steps: S3.1: first construct a judgment matrix, then perform a consistency test on the judgment matrix, and finally calculate the weight vector of the hierarchical analysis method; S3.2: first establish an evaluation matrix, then normalize the evaluation matrix, then calculate the entropy of the consistency evaluation indicator, then calculate the information entropy redundancy, and finally calculate the entropy weight of each consistency evaluation indicator to obtain the entropy weight method weight vector; S3.3: introduce a linear combination coefficient, perform a weighted combination of the hierarchical analysis method weight vector and the entropy weight method weight vector, and calculate the consistency evaluation indicator weight.

[0028] Furthermore, the method for consistency checking the judgment matrix includes the following steps: calculating the maximum eigenvalue of the judgment matrix; calculating a consistency index based on the maximum eigenvalue; calculating a consistency ratio based on the consistency index; judging whether the consistency ratio is less than a preset ratio, if so, passing the consistency check; otherwise, failing the consistency check and adjusting the elements in the judgment matrix;

[0029] The method for calculating the weight vector of the hierarchical analysis method comprises the following steps: calculating the weight of the eigenvalue method; calculating the weight of the arithmetic mean method; calculating the weight of the geometric mean method; calculating the weight vector of the hierarchical analysis method according to the weight of the eigenvalue method, the weight of the arithmetic mean method and the weight of the geometric mean method;

[0030] The method for normalizing the evaluation matrix includes the following steps: first normalizing the positive indicators and the negative indicators, and then obtaining the normalized matrix of all evaluation factors.

[0031] Furthermore, the judgment matrix is:

[0032]

[0033] Wherein, A represents the judgment matrix, a ij Indicates indicator a i Relative to indicator a j The relative importance of , and the following constraints are met:

[0034] a ij >0i,j∈Z∩[1,9]

[0035]

[0036] a ii =1i∈Z∩[1,9]

[0037] a ij ∈{1, 2, 3, 4, 5, 6, 7, 8, 9}i,j∈Z∩[1,9];

[0038] The solution formula for the maximum eigenvalue is: det(A-λ·I)=0

[0039] Where λ represents the eigenvalue;

[0040] The calculation formula of the consistency index is:

[0041] Where, CI represents the consistency index, λ max represents the maximum eigenvalue, and L represents the number of consistency evaluation indicators in the power battery consistency evaluation indicator system;

[0042] The calculation formula of the consistency ratio is:

[0043] Wherein, CR represents the consistency ratio, and RI represents the average random consistency index;

[0044] The calculation formula of the eigenvalue method weight is:

[0045]

[0046] Where, represents the weight of the eigenvalue method, Represents the weight value of the kth feature in the first category of indicators;

[0047] The calculation formula of the arithmetic mean weight is:

[0048]

[0049] Where, represents the arithmetic mean weight, Indicates the consistency evaluation index F calculated using the arithmetic average method i The weight of (i∈Z∩[1,9]), L represents the number of consistency evaluation indicators in the power battery consistency evaluation indicator system;

[0050] The calculation formula of the geometric mean weight is:

[0051]

[0052] Where, represents the geometric mean weight, Represents the consistency evaluation index F calculated using the geometric mean method i The weight of (i∈Z∩[1,9]);

[0053] The calculation formula of the AHP weight vector is:

[0054] Wherein, W1 represents the weight vector of the AHP method;

[0055] The evaluation matrix is:

[0056] The normalization formula of positive indicators is:

[0057] The normalization formula for negative indicators is:

[0058] The normalized matrix is:

[0059]

[0060] The entropy e of the consistency evaluation index j The calculation formula is:

[0061]

[0062] Where, Satisfy e j >0;

[0063] The information entropy redundancy d j The calculation formula is: j =1-e j

[0064] The calculation formula of the entropy weight method weight vector is:

[0065]

[0066] Where W2 represents the weight vector of the entropy weight method,

[0067] The calculation formula of the consistency evaluation index weight is: W = λ1·W1+λ2·W2

[0068] Where W represents the weight of the consistency evaluation index, λ1 and λ2 are linear combination coefficients;

[0069] Establish a model to determine the optimal weights. The objective function of the model is:

[0070]

[0071] The constraints of the model are: λ1+λ2=1λ1,λ2≥0

[0072] The calculation formula of the optimal weight is:

[0073] Where W * represents the optimal weight.

[0074] Furthermore, the calculation formula for each consistency indicator score is:

[0075]

[0076] Where, F i (i∈Z∩[1,9]) represents the i-th consistency evaluation index, P i (i∈Z∩[1,9]) represents the threshold of the i-th consistency evaluation indicator;

[0077] The calculation formula for the total evaluation score is:

[0078] Where G i represents the score of the i-th consistency evaluation index, represents the optimal weight of the i-th consistency indicator.

[0079] Furthermore, during feature extraction, a charging feature set from the charging data, a static feature set from the static data, and a discharge feature set from the discharge data are extracted and obtained; the charging feature set includes voltage-related features, capacity-related features, and charging IC curve-related features of each battery in a charging state; the static feature set includes voltage-related features, temperature-related features, and time-related features of each battery in a static state; and the discharge feature set includes voltage-related features, current-related features, temperature-related features, and resistance-related features of each battery in a discharge state;

[0080] The feature screening method comprises the following steps: S5.2.1: performing feature selection on the charging feature set, the static feature set, and the discharging feature set through data segmentation and model training; S5.2.2: performing an importance assessment on all features in each state based on the model to obtain an importance score; S5.2.3: removing the feature with the lowest contribution from the current model based on the importance score; S5.2.4: retraining the model based on the remaining features; S5.2.5: repeating steps S5.2.2 to S5.2.4 until a preset target number of features is reached, thereby obtaining a preliminary feature set for each state;

[0081] The feature construction method includes the following steps: S5.3.1: randomly combine any two features in all preliminary feature sets; S5.3.2: perform grey correlation analysis to obtain the correlation coefficient; S5.3.3: select features with high correlation coefficients to form the optimal feature set of each battery; S5.3.4: perform Z-score standardization on the feature data of the preliminary feature set under each state, and finally obtain the optimal feature set of each battery.

[0082] Furthermore, the calculation formulas for the evaluation weights of the two models are:

[0083]

[0084] Where W BP Represents the evaluation weight of the BP neural network model, W LightGRM Represents the evaluation weight of the LightGRM model; MSE BP Represents the loss function of the BP neural network model, MSE LightGRM Represents the loss function of the LightGRM model;

[0085] The calculation formula for the battery health evaluation score is:

[0086] y score =W BP ×y BP +W LightGRM ×y LightGRM

[0087] Where y score Represents the battery health evaluation score, y BP represents the rating of the power battery by the trained BP neural network model, y LightGRM It represents the score of the power battery by the trained LightGRM model.

[0088] As a further improvement of the above scheme, the operating data uploaded by the battery management system is extracted from the cloud, and the method for preprocessing the operating data includes the following steps: S1.1: comparing the key data information in each row of data, identifying and deleting duplicate record data; S1.2: checking whether there is data missing based on the storage frequency of the data uploaded by the battery management system, and if so: for individually missing data points, using linear interpolation to estimate based on the adjacent data points before and after the missing point to fill in the missing value; for continuous missing fragments in the data, directly deleting the specified data fragment; for fragments with scattered missing data, using the average value of the same type of data to fill in; S1.3: judging whether the information contained in the data strip has at least one of the following situations: (1) the battery cell voltage is greater than the preset cell voltage maximum value; (2) the battery cell voltage is less than the preset cell voltage minimum value; (3) the temperature is greater than the preset temperature maximum value; if so, deleting the corresponding data strip.

[0089] The present invention also provides a power battery health scoring system based on multi-feature fusion, which applies any of the above-mentioned power battery health scoring methods based on multi-feature fusion; the scoring system includes:

[0090] A data processing and classification module is used to pre-process and classify the acquired power battery operation data to obtain charging / resting / discharging data;

[0091] A system construction module, which is used to construct a system including voltage, temperature, capacity, and internal resistance-related consistency evaluation indicators based on the charging / resting / discharging data;

[0092] A weight determination module, configured to determine the weights of consistency evaluation indicators of the power battery in charging / resting / discharging states using a combined weighting method based on game theory;

[0093] A preliminary scoring calculation module, which is used to calculate the overall evaluation score of the power battery by combining various consistency evaluation indicators and their weights and thresholds;

[0094] A feature engineering module is used to obtain various data sets based on the charging / resting / discharging data and perform feature extraction, screening, and construction to obtain the optimal feature set for each power battery;

[0095] A model training module is used to first divide the optimal feature set into a training set, a validation set, and an evaluation set according to a preset ratio, and then train the BP neural network model and the LightGRM model respectively according to the total evaluation score;

[0096] The final score calculation module is used to first calculate the evaluation weights of the two models, and then calculate the final battery health evaluation score based on the evaluation weights and the scores of the power battery by the two models.

[0097] Compared with existing power battery health scoring methods and systems, the power battery health scoring method and system based on multi-feature fusion of the present invention has the following beneficial effects:

[0098] 1. The power battery health scoring method based on multi-feature fusion preprocesses and classifies the power battery operation data to construct a battery consistency evaluation index system. Then, based on the combined weighted method of game theory, the weight of each consistency index is scientifically determined. Then, the weight and threshold of each consistency index are combined to calculate the total evaluation score of the power battery. Feature engineering is then performed to obtain the optimal feature set. Finally, the BP neural network model and LightGRM model are trained separately using the optimal feature set. According to the evaluation weights of the two models and the total evaluation scores obtained by training each model, the final battery health evaluation score is obtained by weighted combination. This hybrid modeling method based on the lightweight spatiotemporal graph convolutional network significantly improves the accuracy and robustness of battery health status estimation by integrating physical mechanism constraints and data-driven advantages, and solves the technical problems of low accuracy and poor robustness of existing power battery health scoring methods.

[0099] 2. This multi-feature fusion-based power battery health scoring method distinguishes between charging, static, and discharging state data to avoid assessment bias caused by mixed operating conditions and improve adaptability to different operating scenarios. It integrates consistency indicators for key parameters such as voltage, temperature, capacity, and internal resistance, covering the main factors affecting battery degradation (such as lithium plating, SEI growth, and active material loss), providing a more comprehensive and accurate battery health score.

[0100] 3. This multi-feature fusion-based power battery health scoring method utilizes a BP neural network model to handle the nonlinear relationships between continuous variables such as voltage and temperature, and a LightGRM model to efficiently handle discrete features such as capacity and internal resistance. Leveraging the complementary strengths of these two models, this method achieves higher scoring accuracy than a single model alone. Furthermore, this is the first time that game theory weight allocation has been combined with a hybrid machine learning model for battery health scoring. The scoring results can directly guide battery maintenance strategies, significantly extending battery life, and are applicable to a wider range of battery scenarios and types.

[0101] 4. The beneficial effects of the power battery health scoring system based on multi-feature fusion are the same as those of the above-mentioned power battery health scoring method, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0102] Figure 1This is a flowchart of a power battery health scoring method based on multi-feature fusion according to Example 1 of the present invention;

[0103] Figure 2 for Figure 1 Data collection flow chart of the power battery health scoring method in [1];

[0104] Figure 3 for Figure 1 Consistency scoring flow chart in the power battery health scoring method;

[0105] Figure 4 for Figure 1 Flowchart of feature engineering in the power battery health scoring method. DETAILED DESCRIPTION

[0106] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0107] Example 1

[0108] Referring to the figure, this embodiment provides a power battery health scoring method based on multi-feature fusion, which is used to score the health of a power battery. In this embodiment, the power battery is a battery pack for a new energy vehicle, which has multiple cells. Of course, in other embodiments, the power battery can also serve as an energy device for other equipment. The power battery health scoring method includes the following steps (S1-S6).

[0109] S1: Preprocess and classify the acquired power battery operating data to obtain charging / resting / discharging data. In this embodiment, operating data uploaded by the battery management system is extracted from the cloud. The classification of the operating data is primarily based on the battery state (charging / resting / discharging). The method for preprocessing the operating data includes the following steps (S1.1-S1.3).

[0110] S1.1: Compare key data within each row of data, identify, and delete duplicate records. Key data includes battery pack index, timestamp, and battery cell voltage. This step primarily removes duplicate data, reducing data processing and preventing it from interfering with subsequent scoring.

[0111] S1.2: Based on the storage frequency of data uploaded by the battery management system (BMS), check for missing data. If missing data exists, perform the following steps: For single missing data points, use linear interpolation to estimate the missing value based on the adjacent data points before and after the missing point; For consecutive missing data segments, directly delete the specified data segment; For fragmented missing data segments, use the average value of the same type of data to fill the missing data. This step handles multiple types of missing data to ensure data integrity.

[0112] S1.3: Determine whether the information contained in the data strip meets at least one of the following conditions: (1) the battery cell voltage is greater than the preset maximum cell voltage; (2) the battery cell voltage is less than the preset minimum cell voltage; (3) the temperature is greater than the preset maximum temperature; if so, delete the corresponding data strip. The data strip mainly contains information such as voltage and temperature. This step mainly processes abnormal data to ensure the accuracy, reliability, and practicality of the data scoring results.

[0113] S2: Based on the charge / rest / discharge data, a system of consistency evaluation indicators related to voltage, temperature, capacity, and internal resistance is constructed. To ensure that the consistency evaluation indicators are constructed based on sufficient data, this embodiment uses the maximum charging voltage as the benchmark and [V1, V2] as the total voltage selection interval, thereby obtaining a more stable and reliable charging segment. V1 is the effective voltage at the start of charging, and V2 is the voltage before charging is terminated.

[0114] In this embodiment, voltage-related consistency evaluation indicators include voltage range F1, voltage curve distance F2, and voltage variation coefficient F3; temperature-related consistency evaluation indicators include temperature range F4, temperature curve distance F5, and temperature rise rate F6; capacity-related consistency evaluation indicators include incremental capacity peak F7 and plateau area charge capacity F8; and internal resistance-related consistency evaluation indicators include internal resistance variation coefficient F9. The calculation formulas for the above nine consistency evaluation indicators are described below.

[0115] (1) Voltage difference F1

[0116] This embodiment selects the battery cell voltage data of [V3, V4] from the charging segment and calculates the voltage range of the specified charging segment. The specific calculation formula is as follows:

[0117]

[0118] In the above formula, N represents the length of the specified charging data segment, V t,max Indicates the maximum battery cell voltage at time t, V t,min Indicates the minimum battery cell voltage at time t.

[0119] (2) Voltage curve distance F2

[0120] Select the battery cell voltage data of [V3, V4] from the charging segment, calculate the Euclidean distance between each battery cell voltage curve and the average voltage curve, and record the maximum Euclidean distance as the voltage curve distance. The specific calculation formula is as follows:

[0121]

[0122] In the above formula, N represents the charging data length of the specified voltage segment, B represents the number of battery cells in the battery pack, Indicates the voltage of battery cell No. b at time t. Represents the average voltage of all battery cells at time t.

[0123] (3) Voltage variation coefficient F3

[0124] Select the battery cell voltage data [V3, V4] from the charging segment and calculate the root mean square error and coefficient of variation of the extreme voltage curve during the charging process to evaluate the voltage change. The specific calculation formula is as follows:

[0125]

[0126] In the above formula, N represents the charging data length of the specified voltage segment. represents the average voltage of all batteries at the tth moment of the charging process, represents the average voltage of a given charging segment, σ v The F3 value represents the root mean square error between the maximum and minimum voltage curve data during charging. The F3 value represents the coefficient of variation of the charging voltage of all cells and is an indicator for evaluating voltage variation. A larger F3 value indicates poorer voltage consistency.

[0127] (4) Temperature range F4

[0128] Calculate the maximum temperature difference for a specified charging segment. The specific calculation formula is as follows:

[0129]

[0130] In the above formula, N represents the length of the specified charging data segment, T t,max Indicates the maximum temperature value detected by the battery pack temperature sensor at time t, T t,min The minimum temperature value detected by the battery pack temperature sensor at time t.

[0131] (5) Temperature curve distance F5

[0132] Calculate the Euclidean distance between the temperature curve of each temperature sensor and the average temperature curve, and record the average of all Euclidean distances as the curve distance. The specific calculation formula is as follows:

[0133]

[0134] In the above formula, N represents the charging data length of the specified voltage segment, S represents the total number of battery pack temperature sensors, It represents the temperature detected by the s-th temperature sensor at the t-th time. Represents the average temperature of all temperature sensors at time t.

[0135] (6) Temperature rise rate F6

[0136] To reflect the degree of temperature change, the temperature data of each temperature sensor is selected from the charging segment, the amplitude of the temperature change during the charging cycle is calculated, and the average of all temperature rise rates is recorded as the average temperature rise rate. The specific calculation formula is as follows:

[0137]

[0138] In the above formula, S represents the total number of battery pack temperature sensors. Respectively represent the temperature at the start and end of charging of the sth temperature sensor, Indicates the charging start time and charging end time respectively.

[0139] (7) Incremental capacity peak F7

[0140] This embodiment adopts the "fit first, then differentiate" incremental capacity curve extraction approach. Based on the current and voltage signals collected during the constant current charging process, the curve fitting process is used for filtering and smoothing, and the differentiated method is used to reduce the noise caused by the differential. The specific steps for calculating the incremental capacity peak are as follows: the current signal I(t) is integrated in ampere hours to obtain the time series q(t) of the battery cell charged; the collected voltage data u(t) is discretized into a set of fixed-interval voltage points u = (u1, u2, u3, u4, ..., u n ) T , where u n -u n-1 =Δu, Δu is a preset fixed value; use q(t) as Y (electricity sequence), u as X (voltage sequence), and fit a tenth-order polynomial by the polynomial fitting method to obtain the QU curve. Read Q through the QU curve = (Q1, Q2, Q3, Q4, ..., Q n ) T According to the voltage sequence U and the power sequence Q, the incremental capacity curve is obtained by using the successive difference method, and the incremental capacity peak is further obtained. The final calculation formula is:

[0141]

[0142] In the above formula, Q represents the charge sequence and U represents the voltage sequence.

[0143] (8) Platform area charging capacity F8

[0144] In order to reflect the degree of change in battery capacity, the current data of [V3, V4] in the charging segment is selected, and the charged capacity is calculated based on the ampere-hour integration method. The ratio of the charged capacity to the rated capacity is recorded as the capacity change in the platform area, and the calculation formula is obtained as follows:

[0145]

[0146] In the above formula, C0 represents the rated capacity of the battery cell, I i represents the current of the battery pack at the i-th moment, t start , t end Respectively represent the start and end time of charging for the specified voltage segment.

[0147] (9) Internal resistance variation coefficient F9

[0148] Extract the data from the process of current reduction at the charging end, that is, the data from the moment the current at the charging end decreases to the moment the charging ends, and calculate the internal resistance variation coefficient. The calculation formula is as follows:

[0149]

[0150]

[0151] In the above formula, B represents the total number of battery cells in the battery pack, R b Indicates the resistance of the b-th battery cell, in milliohms. It indicates the voltage of the bth battery cell at the last moment before the current drops during the charging process. Indicates the voltage of the bth battery cell at the first moment after the current drops during the charging process. e Indicates the current value at the last moment before the current drops during the charging process. s Indicates the current value at the first moment after the current drops during the charging process. Represents the average value of all battery cell resistances, σ R This value represents the standard deviation of all battery cells in the pack. A larger F9 indicates a greater variation in the internal resistance of the battery pack, indicating poor pack consistency.

[0152] S3: Determine the consistency evaluation index weights of the power battery in the charging / resting / discharging state using a combined weighting method based on game theory. In this embodiment, the method for determining the consistency evaluation index weights includes the following steps (S3.1-S3.3).

[0153] S3.1: First, construct a judgment matrix, then perform a consistency check on the judgment matrix, and finally calculate the weight vector of the hierarchical analysis method. In this embodiment, the method for consistency checking the judgment matrix includes the following steps: calculating the maximum eigenvalue of the judgment matrix; calculating a consistency index based on the maximum eigenvalue; calculating a consistency ratio based on the consistency index; and determining whether the consistency ratio is less than a preset ratio. If so, the consistency check is passed; otherwise, the consistency check fails and the elements in the judgment matrix are adjusted.

[0154] The judgment matrix is:

[0155]

[0156] Among them, A represents the judgment matrix, a ij Indicates indicator a i Relative to indicator a j The relative importance of , and the following constraints are met:

[0157] a ij >0i,j∈Z∩[1,9]

[0158]

[0159] a ii =1i∈Z∩[1,9]

[0160] a ij ∈{1, 2, 3, 4, 5, 6, 7, 8, 9}i,j∈Z∩[1,9];

[0161] The solution formula for the maximum eigenvalue is:

[0162] det(A-λ·I)=0

[0163] In the above formula, λ represents the eigenvalue. Solving the above formula can obtain the maximum eigenvalue λ of matrix A max .

[0164] The calculation formula of consistency index is:

[0165]

[0166] In the formula, CI represents the consistency index, λ max represents the maximum eigenvalue, and L represents the number of consistency evaluation indicators in the power battery consistency evaluation index system.

[0167] The formula for calculating the consistency ratio is:

[0168]

[0169] Where CR is the consistency ratio and RI is the average random consistency index. According to the order of the judgment matrix A, the table below shows that RI = 1.46.

[0170] Table 1 Comparison table of matrix order and RI value

[0171] Matrix order 1 2 3 4 5 6 RI value 0 0 0.52 0.89 1.12 1.26 Matrix order 7 8 9 10 11 12 13 RI value 1.36 1.41 1.46 1.49 1.51 1.54 1.56

[0172] Among them, if CR<0.1, it is considered that the judgment matrix A has passed the consistency test; if CR≥0.1, the elements in matrix A need to be adjusted.

[0173] In this embodiment, the method for calculating the weight vector of the hierarchical analysis method includes the following steps: calculating the eigenvalue method weight; calculating the arithmetic mean method weight; calculating the geometric mean method weight; and calculating the hierarchical analysis method weight vector based on the eigenvalue method weight, the arithmetic mean method weight and the geometric mean method weight.

[0174] When calculating the eigenvalue weight, first solve det(A-λ·I)=0 to get the largest eigenvalue λ max , and further solve the following formula:

[0175]

[0176] Finally, the calculation formula of the eigenvalue method weight is:

[0177]

[0178] Where, represents the eigenvalue method weight, Indicates the weight value of the kth feature in the first category of indicators.

[0179] When calculating the arithmetic mean weight, first calculate the weight of the consistency evaluation index:

[0180]

[0181] In the above formula, Indicates the consistency evaluation index F calculated using the arithmetic average method i (i∈Z∩[1,9]), and L represents the number of consistency evaluation indicators in the power battery consistency evaluation index system.

[0182] The calculation formula of the arithmetic mean weight is further obtained as follows:

[0183]

[0184] When calculating the weight of the geometric mean method, first calculate the weight of the consistency evaluation index:

[0185]

[0186] In the above formula, Represents the consistency evaluation index F calculated using the geometric mean method i The weight of (i∈Z∩[1,9]).

[0187] The calculation formula of the geometric mean weight is further obtained as follows:

[0188]

[0189] Where, represents the geometric mean weight, Represents the consistency evaluation index F calculated using the geometric mean method i The weight of (i∈Z∩[1,9]).

[0190] After the above three weight calculations are completed, the final weight vector is calculated, and the calculation formula of the AHP weight vector is obtained as follows:

[0191]

[0192] Where W1 represents the weight vector of the AHP method, represents the weight vector calculated by the eigenvalue method, represents the weight vector calculated by the arithmetic mean method, Represents the weight vector for geometric mean calculation.

[0193] S3.2: First, establish an evaluation matrix, then normalize the evaluation matrix, calculate the entropy of the consistency evaluation indicators, then calculate the information entropy redundancy, and finally calculate the entropy weight of each consistency evaluation indicator to obtain the entropy weight method weight vector. The normalization method for the evaluation matrix includes the following steps: first normalize the positive indicators and normalize the negative indicators, and then obtain the normalized matrix of all evaluation factors.

[0194] The evaluation matrix is:

[0195]

[0196] The normalization formula of positive indicators is:

[0197]

[0198] The normalization formula for negative indicators is:

[0199]

[0200] The normalized matrix is:

[0201]

[0202] Entropy e of consistency evaluation index j The calculation formula is:

[0203]

[0204] Where, Satisfy e j >0.

[0205] Information entropy redundancy d j The calculation formula is:

[0206] d j =1-e j

[0207] The calculation formula of the entropy weight method weight vector is:

[0208]

[0209] Where W2 represents the entropy weight method weight vector,

[0210] S3.3: Introduce the linear combination coefficient, perform weighted combination of the AHP weight vector and the entropy weight method weight vector, and calculate the consistency evaluation index weight. In this embodiment, the calculation formula of the consistency evaluation index weight is:

[0211]

[0212] Where W represents the consistency evaluation index weight, and λ1 and λ2 are linear combination coefficients.

[0213] Establish a model to determine the optimal weights. The objective function of the model is:

[0214]

[0215] The constraints of the model are:

[0216] λ1+λ2=1

[0217] λ1,λ2≥0

[0218] CPLEX solves the model and obtains the optimal weight calculation formula:

[0219]

[0220] Where W * represents the optimal weight.

[0221] S4: Calculate the total evaluation score of the power battery by combining the various consistency evaluation indicators and their weights and thresholds. i The calculation formula for (i∈Z∩[1,9]) is:

[0222]

[0223] Where, F i (i∈Z∩[1,9]) represents the i-th consistency evaluation index, P i (i∈Z∩[1,9]) represents the threshold of the i-th consistency evaluation indicator.

[0224] The calculation formula for the total evaluation score is further obtained as follows:

[0225]

[0226] Where G i represents the score of the i-th consistency evaluation index, Represents the optimal weight of the i-th consistency index. The total evaluation score G 总 It can reflect the actual score of a certain battery evaluation and provide a reference and comparison for the scores obtained in subsequent model training. The closer the scores are, the more accurate the trained model is.

[0227] S5: Based on the charging / resting / discharging data, obtain each data set and perform feature extraction, screening, and construction to obtain the optimal feature set for each power battery. in represents the charging data set of the mth battery; represents the static data set of the mth battery; represents the discharge data set of the mth battery.

[0228] In this embodiment, when extracting features, the charging feature set in the charging data, the static feature set in the static data, and the discharge feature set in the discharge data are extracted and obtained. in represents the charging data set of the mth battery, represents the static data set of the mth battery, represents the discharge data set of the mth battery.

[0229] The charging feature set includes voltage-related features, capacity-related features, and charging IC curve-related features of each battery under charging state. A series of calculations are performed on the voltage, temperature, time and other data of each segment in the charging process to obtain a charging feature set. in, Indicates the voltage-related characteristics of the mth battery under charging state; Indicates the capacity-related characteristics of the mth battery under charging state; Indicates the characteristics of the charging IC curve of the mth battery under charging state. Includes: maximum voltage, minimum voltage, voltage difference, voltage variation coefficient, and charging voltage change rate under charging state; Includes: charging time, charging rate, total charging capacity, and constant current stage charging capacity under charging state; Includes: the peak value of the IC curve in the charging state, the voltage corresponding to the peak value of the IC curve, and the IC peak area.

[0230] The static feature set includes voltage-related features, temperature-related features, and time-related features of each battery in a static state. The voltage, temperature and other data in the static feature set are obtained in, represents the voltage-related characteristics of the mth battery in a static state; represents the temperature-related characteristics of the mth battery in a static state; represents the time-dependent characteristics of the mth battery in a static state. Includes: initial voltage, stable voltage, voltage recovery amplitude, and self-discharge rate in static state; Includes: initial temperature, stable temperature, and temperature difference in static state; Includes: the time it takes for the voltage and temperature to stabilize and the length of time it takes to stand still.

[0231] The discharge feature set includes voltage-related features, current-related features, temperature-related features, and resistance-related features of each battery in the discharge state. A series of calculations are performed on the voltage, temperature, time and other data of each segment in the charging process to obtain a charging feature set. in, Represents the voltage-related characteristics of the mth battery in the discharge state; Represents the current-related characteristics of the mth battery in the discharge state; Represents the temperature-related characteristics of the mth battery in the discharge state; Represents the resistance-related characteristics of the mth battery in the discharge state. Contains: starting voltage, ending voltage, voltage drop rate, and average voltage in the discharge state; Includes: average current and maximum discharge current in discharge state; Includes: temperature difference and temperature change rate under discharge state; Includes: dynamic internal resistance and low-frequency impedance in discharge state.

[0232] The feature screening method includes the following steps: S5.2.1: charging feature set Static feature set and discharge feature set Perform feature selection through data segmentation and model training; S5.2.2: Evaluate the importance of all features in each state based on the model to obtain an importance score; S5.2.3: Based on the importance score, remove the feature with the lowest contribution from the current model; S5.2.4: Retrain the model based on the remaining features; S5.2.5: Repeat steps S5.2.2 to S5.2.4 until the preset target number of features is reached, and obtain a preliminary feature set for each state.

[0233] In this embodiment, data segmentation is to randomly divide the feature data in each state into a training set and a validation set at a ratio of 70% and 30%, respectively. Model training uses the XGBoost algorithm to train the regression model on the training set and calculate the contribution (gain) of each feature to the model's prediction ability. Finally, the preliminary feature set in each state is obtained. Where P represents the number of target features in each state; represents the preliminary feature set under the mth battery charging state; represents the preliminary feature set of the mth battery in static state; Represents the preliminary feature set under the m-th battery discharge state.

[0234] The feature construction method includes the following steps: S5.3.1: For all preliminary feature sets Randomly combine any two features in the feature set of the mth battery, that is, randomly combine two features in the feature set of the mth battery; S5.3.2: Perform grey correlation analysis to obtain the correlation coefficient; S5.3.3: Select features with high correlation coefficients to form the optimal feature set of each battery; S5.3.4: Perform Z-score standardization on the feature data of the preliminary feature set under each state, and finally obtain the optimal feature set of each battery.

[0235] In this embodiment, first, features with high correlation coefficients are selected to form the optimal feature set for the mth battery, namely:

[0236]

[0237] in, represents the optimal feature set of the mth battery charging state; represents the kth optimal feature of the mth battery charge state, where K represents the number of optimal features; represents the optimal feature set of the mth battery in the static state; represents the kth optimal feature of the mth battery at rest, where K represents the number of optimal features; represents the optimal feature set of the mth battery discharge state; represents the kth optimal feature of the mth battery discharge state, and K represents the number of optimal features.

[0238] Secondly, respectively The characteristic data in is Z-score standardized, that is:

[0239]

[0240] In the above formula, X m,k represents the kth characteristic of the mth battery under charging / resting / discharging state; μ k represents the mean of the kth feature; σ k represents the standard deviation of the k-th feature.

[0241] Finally, the optimal feature set of the mth battery is obtained:

[0242]

[0243] In the above formula, Represents the optimal feature set of the mth battery.

[0244] S6: First, the optimal feature set is divided into a training set, a validation set, and an evaluation set according to a preset ratio. The BP neural network model and the LightGRM model are trained separately based on the total evaluation score. The evaluation weights of the two models are then calculated. Finally, the final battery health evaluation score is calculated based on the evaluation weights and the scores of the power battery from the two models. The closer the battery health evaluation scores obtained by training the two models are to the total evaluation score calculated previously, the more accurate the trained model is and the more suitable it is for evaluating the quality of the power battery.

[0245] In this embodiment, The feature data in the dataset is divided into training, validation, and evaluation sets in a ratio of 8:1:1. The training set is used for model training, the validation set is used for hyperparameter tuning and model selection, and the test set is used for final evaluation of model performance. The BP neural network is trained using the training set data, and the backpropagation algorithm is used to optimize the network parameters. The Sigmoid function is selected as the activation function, and the mean square error (MSE) is selected as the loss function. The same training set data (features and target scores) is input into the LightGBM model, and the mean square error and MSE are selected as the loss functions.

[0246] The calculation formulas for the evaluation weights of the two models are:

[0247]

[0248] Where W BP Represents the evaluation weight of the BP neural network model, W LightGRM Represents the evaluation weight of the LightGRM model; MSE BP Represents the loss function of the BP neural network model, MSE LightGRM Represents the loss function of the LightGRM model;

[0249] The calculation formula for the battery health evaluation score is:

[0250] y score =W BP ×y BP +W LightGRM ×y LightGRM

[0251] Where y score Indicates the battery health evaluation score, y BP represents the rating of the power battery by the trained BP neural network model, y LightGRM It represents the score of the power battery by the trained LightGRM model.

[0252] In summary, compared with existing power battery health scoring methods, the power battery health scoring method based on multi-feature fusion in this embodiment has the following beneficial effects:

[0253] 1. The power battery health scoring method based on multi-feature fusion preprocesses and classifies the power battery operation data to construct a battery consistency evaluation index system. Then, based on the combined weighted method of game theory, the weight of each consistency index is scientifically determined. Then, the weight and threshold of each consistency index are combined to calculate the total evaluation score of the power battery. Feature engineering is then performed to obtain the optimal feature set. Finally, the BP neural network model and LightGRM model are trained separately using the optimal feature set. According to the evaluation weights of the two models and the total evaluation scores obtained by training each model, the final battery health evaluation score is obtained by weighted combination. This hybrid modeling method based on the lightweight spatiotemporal graph convolutional network significantly improves the accuracy and robustness of battery health status estimation by integrating physical mechanism constraints and data-driven advantages, and solves the technical problems of low accuracy and poor robustness of existing power battery health scoring methods.

[0254] 2. This multi-feature fusion-based power battery health scoring method distinguishes between charging, static, and discharging state data to avoid assessment bias caused by mixed operating conditions and improve adaptability to different operating scenarios. It integrates consistency indicators for key parameters such as voltage, temperature, capacity, and internal resistance, covering the main factors affecting battery degradation (such as lithium plating, SEI growth, and active material loss), providing a more comprehensive and accurate battery health score.

[0255] 3. This multi-feature fusion-based power battery health scoring method utilizes a BP neural network model to handle the nonlinear relationships between continuous variables such as voltage and temperature, and a LightGRM model to efficiently handle discrete features such as capacity and internal resistance. Leveraging the complementary strengths of these two models, this method achieves higher scoring accuracy than a single model alone. Furthermore, this is the first time that game theory weight allocation has been combined with a hybrid machine learning model for battery health scoring. The scoring results can directly guide battery maintenance strategies, significantly extending battery life, and are applicable to a wider range of battery scenarios and types.

[0256] Example 2

[0257] This embodiment provides a power battery health scoring system based on multi-feature fusion, which applies the power battery health scoring method based on multi-feature fusion in Example 1, and includes a data processing and classification module, a system construction module, a weight determination module, a preliminary score calculation module, a feature engineering module, a model training module, and a final score calculation module.

[0258] The data processing and classification module is used to pre-process and classify the obtained power battery operation data to obtain charging / resting / discharging data. The system construction module is used to construct a system including consistency evaluation indicators related to voltage, temperature, capacity, and internal resistance based on the charging / resting / discharging data. The weight determination module is used to determine the weights of the consistency evaluation indicators of the power battery in the charging / resting / discharging state based on the combined weighted method of game theory. The preliminary score calculation module is used to combine the various consistency evaluation indicators, their weights, and thresholds to calculate the total evaluation score of the power battery.

[0259] The feature engineering module is used to obtain each data set based on the charging / resting / discharging data and perform feature extraction, screening, and construction to obtain the optimal feature set of each power battery. The model training module is used to first divide the optimal feature set into a training set, a validation set, and an evaluation set according to a preset ratio, and then train the BP neural network model and the LightGRM model separately according to the total evaluation score. The final score calculation module is used to first calculate the evaluation weights of the two models, and then calculate the final battery health evaluation score based on the evaluation weights and the scores of the power battery by the two models.

[0260] Example 3

[0261] This embodiment provides a computer terminal including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the power battery health scoring method based on multi-feature fusion according to embodiment 1 are implemented.

[0262] The method of Example 1 can be implemented in the form of software, such as a standalone program installed on a computer terminal, which can be a computer, a smartphone, a control system, or other IoT device. The method of Example 1 can also be implemented as an embedded program installed on a computer terminal, such as a single-chip microcomputer.

[0263] Example 4

[0264] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps of the power battery health scoring method based on multi-feature fusion in embodiment 1 are implemented.

[0265] When the method of Example 1 is applied, it can be applied in the form of software, such as a program designed as a computer-readable storage medium that can run independently. The computer-readable storage medium can be a USB flash drive designed as a USB shield, and the USB flash drive is designed to start the program of the entire method through external triggering.

[0266] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A power battery health scoring method based on multi-feature fusion, characterized in that: It includes the following steps: S1: Preprocess and classify the acquired power battery operation data to obtain charging / resting / discharging data; S2: Based on the charging / resting / discharging data, a system of consistency evaluation indicators including voltage, temperature, capacity, and internal resistance is constructed; S3: Determine the weights of the consistency evaluation indicators of the power battery in the charging / stationary / discharging states using a combined weighted method based on game theory; S4: Calculate the total evaluation score of the power battery by combining various consistency evaluation indicators, their weights, and thresholds; S5: Obtaining each data set based on the charging / resting / discharging data and performing feature extraction, screening, and construction to obtain an optimal feature set for each power battery; S6: First, the optimal feature set is divided into a training set, a validation set, and an evaluation set according to a preset ratio, and the BP neural network model and the LightGRM model are trained respectively according to the total evaluation score. Then, the evaluation weights of the two models are calculated. Finally, the final battery health evaluation score is calculated based on the evaluation weights and the scores of the power battery by the two models.

2. The power battery health scoring method based on multi-feature fusion according to claim 1 is characterized in that: The voltage-related consistency evaluation indicators include voltage range F1, voltage curve distance F2 and voltage variation coefficient F3; the temperature-related consistency evaluation indicators include temperature range F4, temperature curve distance F5 and temperature rise rate F6; the capacity-related consistency evaluation indicators include incremental capacity peak F7 and platform area charging capacity F8; the internal resistance-related consistency evaluation indicators include internal resistance variation coefficient F9; Among them, the calculation formulas for each consistency evaluation index are: Where N represents the length of the specified charging data segment, V t,max Indicates the maximum battery cell voltage at time t, V t,min Indicates the minimum battery cell voltage at time t; Where B represents the number of battery cells in the battery pack. Indicates the voltage of battery cell No. b at time t; Where, represents the average voltage of all batteries at the tth moment of the charging process; Where, T t,max Indicates the maximum temperature value detected by the battery pack temperature sensor at time t, T t,min The minimum temperature value detected by the battery pack temperature sensor at time t; Where S represents the total number of battery pack temperature sensors. It represents the temperature detected by the s-th temperature sensor at the t-th time. Represents the average temperature of all temperature sensors at time t; Where, Respectively represent the temperature at the start and end of charging of the sth temperature sensor, Respectively indicate the charging start time and charging end time; In the formula, Q represents the quantity series, and U represents the voltage series; Where C0 represents the rated capacity of the battery cell, I i represents the current of the battery pack at the i-th moment, t start , t end Respectively represent the start time and end time of charging for a specified voltage segment; Where, Indicates the voltage of the bth battery cell at the last moment before the current drops during the charging process; Indicates the voltage of the bth battery cell at the first moment after the current drops during the charging process; I e Indicates the current value at the last moment before the current drops during the charging process; I s Indicates the current value at the first moment after the current drops during the charging process.

3. The power battery health scoring method based on multi-feature fusion according to claim 2 is characterized in that: The method for determining the consistency evaluation index weights comprises the following steps: S3.1: First, construct a judgment matrix, then perform a consistency check on the judgment matrix, and finally calculate the weight vector of the hierarchical analysis method; S3.2: First, establish an evaluation matrix, then normalize the evaluation matrix, then calculate the entropy of the consistency evaluation index, then calculate the information entropy redundancy, and finally calculate the entropy weight of each consistency evaluation index to obtain the entropy weight method weight vector; S3.3: Introduce a linear combination coefficient, perform a weighted combination of the AHP weight vector and the entropy weight method weight vector, and calculate the consistency evaluation index weight.

4. The power battery health scoring method based on multi-feature fusion according to claim 3 is characterized in that: The method for consistency checking the judgment matrix comprises the following steps: calculating the maximum eigenvalue of the judgment matrix; calculating a consistency index based on the maximum eigenvalue; calculating a consistency ratio based on the consistency index; judging whether the consistency ratio is less than a preset ratio, if so, passing the consistency check; otherwise, failing the consistency check and adjusting the elements in the judgment matrix; The method for calculating the weight vector of the hierarchical analysis method comprises the following steps: calculating the weight of the eigenvalue method; calculating the weight of the arithmetic mean method; calculating the weight of the geometric mean method; calculating the weight vector of the hierarchical analysis method according to the weight of the eigenvalue method, the weight of the arithmetic mean method and the weight of the geometric mean method; The method for normalizing the evaluation matrix includes the following steps: first normalizing the positive indicators and the negative indicators, and then obtaining the normalized matrix of all evaluation factors.

5. The power battery health scoring method based on multi-feature fusion according to claim 4 is characterized in that: The judgment matrix is: Wherein, A represents the judgment matrix, a ij Indicates indicator a i Relative to indicator a j The relative importance of , and the following constraints are met: a ij >0i,j∈Z∩[1,9] a ii =1i∈Z∩[1,9] a ij ∈{1,2,3,4,5,6,7,8,9}i,j∈Z∩[1,9]; The solution formula for the maximum eigenvalue is: det(A-λ·I)=0 Where λ represents the eigenvalue; The calculation formula of the consistency index is: Where, CI represents the consistency index, λ max represents the maximum eigenvalue, and L represents the number of consistency evaluation indicators in the power battery consistency evaluation indicator system; The calculation formula of the consistency ratio is: Wherein, CR represents the consistency ratio, and RI represents the average random consistency index; The calculation formula of the eigenvalue method weight is: Where, represents the weight of the eigenvalue method, Represents the weight value of the kth feature in the first category of indicators; The calculation formula of the arithmetic mean weight is: Where, represents the arithmetic mean weight, Indicates the consistency evaluation index F calculated using the arithmetic average method i The weight of (i∈Z∩[1,9]), L represents the number of consistency evaluation indicators in the power battery consistency evaluation indicator system; The calculation formula of the geometric mean weight is: Where, represents the geometric mean weight, Represents the consistency evaluation index F calculated using the geometric mean method i The weight of (i∈Z∩[1,9]); The calculation formula of the AHP weight vector is: Wherein, W1 represents the weight vector of the AHP method; The evaluation matrix is: The normalization formula of positive indicators is: The normalization formula for negative indicators is: The normalized matrix is: The entropy e of the consistency evaluation index j The calculation formula is: Where, Satisfy e j >0; The information entropy redundancy d j The calculation formula is: d j =1-e j The calculation formula of the entropy weight method weight vector is: Where W2 represents the weight vector of the entropy weight method, The calculation formula for the consistency evaluation index weight is: W=λ1·W1+λ2·W2 Where W represents the weight of the consistency evaluation index, λ1 and λ2 are linear combination coefficients; Establish a model to determine the optimal weights. The objective function of the model is: The constraints of the model are: λ1+λ2=1 λ1,λ2≥0 The calculation formula of the optimal weight is: Where W * represents the optimal weight.

6. The power battery health scoring method based on multi-feature fusion according to claim 5, characterized in that: The calculation formula for each consistency indicator score is: Where, F i (i∈Z∩[1,9]) represents the i-th consistency evaluation index, P i (i∈Z∩[1,9]) represents the threshold of the i-th consistency evaluation indicator; The calculation formula for the total evaluation score is: Where G i represents the score of the i-th consistency evaluation index, represents the optimal weight of the i-th consistency indicator.

7. The power battery health scoring method based on multi-feature fusion according to claim 6, characterized in that: During feature extraction, a charging feature set from the charging data, a static feature set from the static data, and a discharge feature set from the discharge data are extracted and obtained; the charging feature set includes voltage-related features, capacity-related features, and charging IC curve-related features of each battery in a charging state; the static feature set includes voltage-related features, temperature-related features, and time-related features of each battery in a static state; and the discharge feature set includes voltage-related features, current-related features, temperature-related features, and resistance-related features of each battery in a discharge state; The feature screening method comprises the following steps: S5.2.1: performing feature selection on the charging feature set, the static feature set, and the discharging feature set through data segmentation and model training; S5.2.2: performing an importance assessment on all features in each state based on the model to obtain an importance score; S5.2.3: removing the feature with the lowest contribution from the current model based on the importance score; S5.2.4: retraining the model based on the remaining features; S5.2.5: repeating steps S5.2.2 to S5.2.4 until a preset target number of features is reached, thereby obtaining a preliminary feature set for each state; The feature construction method includes the following steps: S5.3.1: randomly combine any two features in all preliminary feature sets; S5.3.2: perform grey correlation analysis to obtain the correlation coefficient; S5.3.3: select features with high correlation coefficients to form the optimal feature set of each battery; S5.3.4: perform Z-score standardization on the feature data of the preliminary feature set under each state, and finally obtain the optimal feature set of each battery.

8. The power battery health scoring method based on multi-feature fusion according to claim 1, characterized in that: The calculation formulas for the evaluation weights of the two models are: Where W BP Represents the evaluation weight of the BP neural network model, W LightGRM Represents the evaluation weight of the LightGRM model; MSE BP Represents the loss function of the BP neural network model, MSE LightGRM Represents the loss function of the LightGRM model; The calculation formula for the battery health evaluation score is: y score =W BP ×y BP +W LightGRM ×y LightGRM Where y score Represents the battery health evaluation score, y BP represents the rating of the power battery by the trained BP neural network model, y LightGRM It represents the score of the power battery by the trained LightGRM model.

9. The power battery health scoring method based on multi-feature fusion according to claim 1, characterized in that: The method for extracting the operating data uploaded by the battery management system from the cloud and preprocessing the operating data includes the following steps: S1.1: Compare key data information in each row of data, identify and delete duplicate records; S1.2: Check whether there is any missing data based on the storage frequency of the data uploaded by the battery management system. If so, the following steps are performed: For single missing data points, use linear interpolation to estimate the missing value based on the adjacent data points before and after the missing point; For consecutive missing segments in the data, directly delete the specified data segment; For fragments with scattered missing data, use the average value of the same type of data to fill the missing data; S1.3: Determine whether the information contained in the data strip contains at least one of the following situations: (1) the battery cell voltage is greater than the preset maximum cell voltage; (2) the battery cell voltage is less than the preset minimum cell voltage; (3) the temperature is greater than the preset maximum temperature; if so, delete the corresponding data strip.

10. A power battery health scoring system based on multi-feature fusion, characterized in that: The application is a power battery health scoring method based on multi-feature fusion as described in any one of claims 1 to 9; the scoring system includes: A data processing and classification module is used to pre-process and classify the acquired power battery operation data to obtain charging / resting / discharging data; A system construction module, which is used to construct a system including voltage, temperature, capacity, and internal resistance-related consistency evaluation indicators based on the charging / resting / discharging data; A weight determination module, configured to determine the weights of consistency evaluation indicators of the power battery in charging / resting / discharging states using a combined weighting method based on game theory; A preliminary scoring calculation module, which is used to calculate the overall evaluation score of the power battery by combining various consistency evaluation indicators and their weights and thresholds; A feature engineering module is used to obtain various data sets based on the charging / resting / discharging data and perform feature extraction, screening, and construction to obtain the optimal feature set for each power battery; A model training module is used to first divide the optimal feature set into a training set, a validation set, and an evaluation set according to a preset ratio, and then train the BP neural network model and the LightGRM model respectively according to the total evaluation score; The final score calculation module is used to first calculate the evaluation weights of the two models, and then calculate the final battery health evaluation score based on the evaluation weights and the scores of the power battery by the two models.

Images