Method for quantifying battery consistency score

Abstract

The invention relates to a method for quantifying a battery consistency score, and aims to solve the problems of data processing and weight distribution of battery consistency assessment in the prior art, and the method comprises the following steps: firstly, carrying out Z-score standardization processing on voltage, internal resistance and capacity data, and eliminating abnormal values through a quartile method; normalized data is subjected to non-linear adjustment by applying an exponential function with a natural constant e as a base number, a principal component analysis method is further adopted to extract principal components of consistency scores, weight coefficients of all dimensions are calculated and dynamically updated, finally, comprehensive consistency scores are calculated through a weighted average method, quantification of multi-dimensional representation to a single index is achieved, and the method is suitable for the multi-dimensional representation. The invention further relates to a method for dynamically adjusting a normalization interval according to historical data of the battery, the adaptability and robustness of an evaluation process are improved, an objective and accurate evaluation tool is provided for quality control of the battery, and the method is suitable for evaluation of batteries of different batches and models.

Description

Technical Field

[0001] The present application relates to the field of battery technology, and in particular to a method for quantifying a battery consistency score. Background Art

[0002] In the battery production process, battery consistency is a key quality indicator. Battery consistency involves data in multiple dimensions, such as voltage, internal resistance, and capacity. These data have different dimensions and ranges. Direct comparison and calculation will lead to distorted results. In order to quantify battery consistency, these data need to be normalized to eliminate dimensional differences. However, the choice of normalization method directly affects the accuracy of subsequent consistency scores. Maximum and minimum value normalization is a commonly used normalization method, but its results are sensitive to outliers in the data, resulting in uneven distribution of normalized data.

[0003] In addition, the standardized data needs to be further processed to generate a consistency score. Using an exponential function to process the standardized data can amplify the differences between the data and make the consistency score more sensitive. However, the parameter selection and calculation process of the exponential function need to be precisely controlled to ensure the rationality and stability of the scoring results. In different production batches and battery types, the maximum and minimum value settings also need to be adjusted according to actual conditions, which increases the complexity of the operation. At the same time, the consistency scores of multiple dimensions need to be calculated comprehensively, and the distribution of weights directly affects the final overall consistency score.

[0004] Therefore, how to simplify the calculation process and improve the objectivity and accuracy of the scoring results while ensuring the effectiveness of the normalization method is a technical problem that needs to be solved urgently. Summary of the Invention

[0005] In order to solve the problems existing in the above-mentioned prior art, the present application aims to provide a method for quantifying the battery consistency score.

[0006] A method for quantifying a battery consistency score described in this application includes the following steps:

[0007] S101. For the data of the three dimensions of voltage, internal resistance and capacity, the Z-score normalization method is used to perform normalization processing to obtain normalized data with unified dimensions;

[0008] S102. In the normalized data, outlier data points of each dimension are automatically identified using the interquartile range method and are removed from the subsequent analysis process;

[0009] S103. For the normalized data after removing outliers, a nonlinear adjustment is performed using an exponential function with the natural constant e as the base, where the exponent is the normalized data value. Through the exponential base selection and exponential dynamic adjustment mechanism, new data with amplified inter-dimensional differences is obtained;

[0010] S104: During the exponential amplification process, the fluctuation of the current data is determined based on the data fluctuation threshold. If the fluctuation exceeds the preset threshold, the exponential parameter adjustment process is triggered to adjust the calculation accuracy by optimizing the exponential function parameters while controlling the consumption of computing resources.

[0011] S105. Use principal component analysis to extract principal components from the consistency scores of the three dimensions, condense the dimension information to avoid information redundancy, and calculate the weight coefficient of each principal component;

[0012] S106. During the calculation of the principal component weights, dynamically updating the principal component weights through statistical analysis of historical data;

[0013] S107. Based on the weight coefficients of each dimension obtained by principal component analysis, a weighted average method is used to calculate the comprehensive consistency score, and the consistency performance of each dimension is quantified into a comprehensive evaluation index;

[0014] S108. For batteries of different batches and models, dynamically adjust the normalization processing interval boundaries according to the maximum and minimum value ranges of their historical data.

[0015] Preferably, in step S101, the Z-score normalization method is used for normalization processing, including: calculating the mean and standard deviation of the voltage value, internal resistance value and capacity value;

[0016] The data of each dimension is standardized according to the mean and standard deviation to eliminate dimensional differences.

[0017] Preferably, in step S102, identifying and eliminating outliers by using the interquartile range method includes: calculating the first quartile and the third quartile of each dimension;

[0018] Determine upper and lower thresholds based on the first quartile and the third quartile;

[0019] Data points outside the threshold range are removed to generate anomaly-free dataset.

[0020] Preferably, in step S103, the nonlinear adjustment using an exponential function with a natural constant e as a base includes: calculating an exponential function value with e as a base for the normalized data of each dimension;

[0021] The exponential parameter of the exponential function is dynamically adjusted to obtain new data with significantly amplified differences.

[0022] Preferably, in step S105, extracting the principal components and calculating the weight coefficients by principal component analysis includes: constructing a scoring matrix of voltage value, internal resistance value and capacity value; calculating the covariance matrix of the scoring matrix and performing eigenvalue decomposition;

[0023] Sort by eigenvalue and select the first two principal components;

[0024] The loading matrix of the principal components is calculated to generate weight coefficients.

[0025] Preferably, in step S107, the weighted average method is used to calculate the comprehensive consistency score, including: multiplying the standardized data by the weight matrix to calculate the projection value of the sample in the principal component space;

[0026] A comprehensive consistency score is generated based on the projection values ​​and the weight coefficients by a weighted average method.

[0027] Preferably, in step S108, the dynamically adjusting the normalization processing interval boundary includes: obtaining the maximum and minimum values ​​in the historical data of batteries of different batches and models;

[0028] Calculating the normalized interval boundary according to the maximum value and the minimum value;

[0029] Based on the interval boundaries, the data of the current batch of batteries is normalized to generate an evaluation result that adapts to production conditions.

[0030] The method for quantifying battery consistency scores described in the present application has the advantage of performing a comprehensive analysis of data in three dimensions: voltage, internal resistance, and capacity. First, Z-score standardization is used to eliminate data distribution differences caused by different dimensions, and the interquartile range method is used to identify and eliminate outliers. Then, an exponential function with the natural constant e as the base is used for nonlinear adjustment, and the exponential parameters are dynamically adjusted according to data fluctuations, and the calculation accuracy is adjusted. Next, the principal component analysis method is applied to extract the principal components of the consistency score, calculate the weight coefficients of each dimension, and dynamically update the weights through historical data statistics. Finally, the comprehensive consistency score is calculated based on the weighted average method, and the multi-dimensional performance is quantified into a single indicator.

[0031] The present invention can also dynamically adjust the normalization processing interval based on the historical data of different batches and models of batteries to improve the adaptability and robustness of the evaluation process. This method effectively solves problems such as data processing, outlier identification, and dimensional weight allocation in battery consistency evaluation, providing an objective and accurate evaluation tool for battery quality control. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1This is a process of a method for quantifying battery consistency score described in this application Figure 1 ;

[0033] Figure 2 This is a process of a method for quantifying battery consistency score described in this application Figure 2 . DETAILED DESCRIPTION

[0034] like Figure 1-Figure 2 As shown, a method for quantifying a battery consistency score described in this application includes the following steps:

[0035] S101. For the data of the three dimensions of voltage, internal resistance and capacity, the Z-score normalization method is used to perform normalization processing to obtain normalized data with unified dimensions;

[0036] S102. In the normalized data, outlier data points of each dimension are automatically identified using the interquartile range method and are removed from the subsequent analysis process;

[0037] S103. For the normalized data after removing outliers, a nonlinear adjustment is performed using an exponential function with the natural constant e as the base, where the exponent is the normalized data value. Through the exponential base selection and exponential dynamic adjustment mechanism, new data with amplified inter-dimensional differences is obtained;

[0038] S104: During the exponential amplification process, the fluctuation of the current data is determined based on the data fluctuation threshold. If the fluctuation exceeds the preset threshold, the exponential parameter adjustment process is triggered to adjust the calculation accuracy by optimizing the exponential function parameters while controlling the consumption of computing resources.

[0039] S105. Use principal component analysis to extract principal components from the consistency scores of the three dimensions, condense the dimension information to avoid information redundancy, and calculate the weight coefficient of each principal component;

[0040] S106. During the calculation of the principal component weights, dynamically updating the principal component weights through statistical analysis of historical data;

[0041] S107. Based on the weight coefficients of each dimension obtained by principal component analysis, a weighted average method is used to calculate the comprehensive consistency score, and the consistency performance of each dimension is quantified into a comprehensive evaluation index;

[0042] S108. For batteries of different batches and models, dynamically adjust the normalization processing interval boundaries according to the maximum and minimum value ranges of their historical data.

[0043] like Figure 1-Figure 2As shown, in step S101, the data of the three dimensions of voltage, internal resistance and capacity are normalized using the Z-score normalization method to eliminate the data distribution differences caused by different dimensions and obtain normalized data of unified dimensions.

[0044] Furthermore, in step S101, the original data set of voltage value, internal resistance value and capacity value is obtained, and the mean and standard deviation of each dimension data are calculated;

[0045] The Z-score standardization method is used to normalize the voltage value to eliminate the data distribution differences caused by dimension differences;

[0046] The Z-score standardization method is used to normalize the internal resistance value to eliminate the data distribution differences caused by dimension differences;

[0047] The Z-score standardization method was used to normalize the capacity values ​​to eliminate the data distribution differences caused by dimension differences;

[0048] Based on the standardized data set, a linear regression model of voltage and capacity values ​​is constructed to predict the battery capacity change trend;

[0049] Using the normalized data set, a support vector machine model of internal resistance and voltage values ​​was established to determine the degree of battery performance degradation.

[0050] The standardized data set is used to train a random forest model of voltage, internal resistance, and capacity to evaluate the battery health status.

[0051] Based on the prediction results of the random forest model, the correlation between voltage, internal resistance and capacity is identified, and the variation pattern of battery performance is analyzed;

[0052] The prediction results of the machine learning model are used to generate a battery performance evaluation report and provide battery maintenance and replacement recommendations.

[0053] Specifically, in step S101, the original data sets of voltage, internal resistance and capacity are obtained, and the mean of the voltage is calculated to be 3.7V, with a standard deviation of 0.1V, the mean of the internal resistance is 0.02Ω, with a standard deviation of 0.005Ω, and the mean of the capacity is 3000mAh, with a standard deviation of 200mAh.

[0054] The voltage value was normalized using the Z-score standardization method, converting the voltage value of 3.6V to (3.6-3.7) / 0.1=-1 to eliminate the data distribution differences caused by the dimension difference;

[0055] The Z-score normalization method is used to normalize the internal resistance value, converting the internal resistance value of 0.025Ω to (0.025-0.02) / 0.005=1 to eliminate the data distribution differences caused by the dimension difference;

[0056] The capacity value was normalized using the Z-score standardization method, converting the capacity value of 3200mAh to (3200-3000) / 200=1 to eliminate the data distribution differences caused by the dimension difference.

[0057] Based on the standardized data set, a linear regression model of voltage and capacity values ​​is constructed, and the least squares method is used to fit the regression equation y = ax + b to predict the battery capacity change trend;

[0058] Using the normalized data set, a support vector machine model of internal resistance and voltage values ​​was established. The radial basis kernel function was selected and the penalty parameter C = 1 was set to determine the degree of battery performance degradation.

[0059] The standardized dataset was used to train a random forest model of voltage, internal resistance, and capacity values. The number of trees was set to 100 and the Gini coefficient was used as the splitting criterion to evaluate the battery health status.

[0060] According to the prediction results of the random forest model, the correlation coefficient between voltage and capacity is calculated to be 0.85, and the correlation coefficient between internal resistance and voltage is -0.73. The correlation between voltage, internal resistance and capacity is identified, and the variation pattern of battery performance is analyzed.

[0061] The prediction results of the machine learning model are used to generate a battery performance evaluation report. Combined with the current battery capacity prediction value below 80% and the internal resistance value above 0.03Ω, battery maintenance and replacement recommendations are provided.

[0062] Specifically also includes: Z-score standardization formula:

[0063]

[0064] X: a single measurement value of the battery's original parameter (voltage, internal resistance, or capacity) (unit: V, Ω, mAh);

[0065] μ: The average value of the parameter in the current batch (e.g., the average voltage is 3.7V);

[0066] σ: The standard deviation of the parameter in the current batch (e.g. voltage standard deviation 0.1V);

[0067] Z: dimensionless value after normalization, eliminating dimension differences.

[0068] like Figure 1-Figure 2As shown, in step S102, in the normalized data, the interquartile range method is used to automatically identify the outlier data points in each dimension, and they are removed from the subsequent analysis process to ensure the stability and reliability of the processed data.

[0069] Furthermore, in step S102, the original data is linearly transformed using the minimum-maximum method to obtain the value range of each feature column;

[0070] According to the numerical range of each feature column, calculate the first quartile and third quartile of each dimension data;

[0071] The interquartile range method was used to determine the upper and lower thresholds of each dimension data;

[0072] Identify outlier data points in each dimension based on upper and lower thresholds;

[0073] Remove the identified outlier data points from the dataset to construct an outlier-free dataset;

[0074] The natural constant e is used as the fixed base of the exponential function to establish a nonlinear amplification model;

[0075] Dynamically adjust the exponential parameter of the exponential function based on the distribution characteristics of the data set without abnormalities;

[0076] For each data dimension, calculate the amplification value after exponential function processing;

[0077] Through the preset threshold range, it is determined whether it is necessary to further adjust the index parameters or reselect the index base for secondary amplification processing, so as to obtain a new data set in which the differences between dimensions are significantly amplified.

[0078] Specifically, in step S102, the original data is linearly transformed using the minimum-maximum method to map the value range of each feature column to the interval [0, 1], including converting the value of feature column A from [10, 100] to [0, 1];

[0079] Based on the numerical range of each feature column, calculate the first quartile and third quartile of each dimension data, including feature column A, where the first quartile is 30 and the third quartile is 70;

[0080] The interquartile range method is used to determine the upper and lower bounds of each dimension data, including the upper and lower bounds of feature column A, which are 15 and 85 respectively, calculated using the formulas Q1-1.5*IQR and Q3+1.5*IQR;

[0081] Where, Q1: first quartile, the value of the 25% quantile after data sorting (e.g., voltage Q1 = 3.6V);

[0082] Q3: The third quartile, the value of the 75% quantile after data sorting (such as voltage Q3 = 3.8V),

[0083] IQR: interquartile range, Q3-Q1 (e.g., voltage IQR = 0.2V), identifies abnormal data points in voltage, internal resistance, or capacity (e.g., voltage <3.3V or >4.1V);

[0084] Based on the upper and lower thresholds, outlier data points of each dimension are identified, including data points with a value of 5 or 90 in feature column A, which are marked as outliers;

[0085] Remove the identified outlier data points from the dataset and construct a non-outlier dataset, including the remaining data after removing 5 and 90 from feature column A;

[0086] The natural constant e is used as the fixed base of the exponential function to establish a nonlinear amplification model, which includes applying the exponential function e to the normalized value 0.5 of the feature column A. 0.5 ;

[0087] According to the distribution characteristics of the data set without abnormalities, the exponential parameter of the exponential function is dynamically adjusted, including adjusting the exponential parameter of feature column A to 1.2 to amplify the difference;

[0088] For each data dimension, calculate the amplification value after the exponential function processing, including 0.5 of feature column A after e 0.6 Get 1.8221;

[0089] Through the preset threshold range, it is determined whether it is necessary to further adjust the exponential parameters or reselect the exponential base for secondary amplification processing. For example, if the amplification value of feature column A is less than 2, the exponential parameter is adjusted to 1.5, and finally a new data set is obtained in which the differences between dimensions are significantly amplified.

[0090] like Figure 1-Figure 2 As shown, in step S103, for the normalized data after removing outliers, an exponential function with the natural constant e as the base is used for nonlinear adjustment, and the exponent is the normalized data value. Through the exponential base selection and exponential dynamic adjustment mechanism, new data with amplified inter-dimensional differences is obtained.

[0091] Furthermore, in step S103, a normalized data set after removing outliers is obtained, and the dimensional range to be processed is determined;

[0092] The natural constant e is used as the fixed base of the exponential function to establish a nonlinear amplification model;

[0093] Dynamically adjust the exponential parameter of the exponential function according to the distribution characteristics of the normalized data;

[0094] For each data dimension, calculate the amplification value after exponential function processing;

[0095] Determine whether further adjustment of index parameters is required based on the preset threshold range;

[0096] If there is insufficient difference between dimensions, the exponential base is reselected for secondary amplification;

[0097] Obtain a new dataset in which the differences between dimensions are significantly amplified;

[0098] Obtain a three-dimensional consistency score dataset and construct a score matrix, where the number of rows is the number of samples and the number of columns is the number of dimensions;

[0099] Calculate the covariance matrix of the rating matrix, which reflects the linear relationship between dimensions.

[0100] Specifically, in step S103, a normalized data set is obtained after removing outliers, and the dimensional range to be processed is determined, including normalizing the data of the three dimensions of voltage, internal resistance, and capacity using the Z-score normalization method to eliminate the data distribution differences caused by different dimensions and obtain normalized data with unified dimensions;

[0101] A nonlinear amplification model is established by using the natural constant e as the fixed base of the exponential function, including constructing an exponential function with e as the base and the exponent as the normalized data value;

[0102] Dynamically adjusting the exponential parameter of the exponential function according to the distribution characteristics of the normalized data, including determining the adjustment range of the exponential parameter by performing statistical analysis on the data distribution;

[0103] For each data dimension, calculating an amplified value after the exponential function processing, including applying the exponential function to the voltage dimension data to obtain an amplified voltage value;

[0104] Determine whether further adjustment of index parameters is needed through a preset threshold range, including setting a threshold of 0.1. If the difference between dimensions is less than this value, parameter adjustment is triggered;

[0105] If the difference between dimensions is insufficient, the exponential base is reselected for secondary amplification, including adjusting the base from e to 2 and performing the amplification calculation again;

[0106] Obtain a new data set with significantly amplified differences between dimensions, including generating new data with significant differences in voltage, internal resistance, and capacity;

[0107] Obtain a three-dimensional consistency score dataset and construct a score matrix. The number of rows in the matrix is ​​the number of samples, and the number of columns is the number of dimensions. This includes constructing a score matrix for 100 samples with 3 columns.

[0108] The covariance matrix of the scoring matrix is ​​calculated. The covariance matrix reflects the linear relationship between dimensions, including analyzing the correlation between voltage and internal resistance through the covariance matrix.

[0109] Specifically also includes: exponential amplification function: Y=e kx

[0110] X: standardized data (Z-score result), ranging from [-3, 3];

[0111] k: A dynamically adjusted exponential parameter (e.g., k = 1.2) used to amplify the differences between dimensions;

[0112] Y: Amplified data, enhancing the sensitivity differences between voltage, internal resistance, and capacity.

[0113] like Figure 1-Figure 2 As shown, in step S104, in the exponential amplification processing, the fluctuation of the current data is judged according to the data fluctuation threshold. If it exceeds the preset threshold, the exponential parameter adjustment process is triggered, and the calculation accuracy is adjusted by optimizing the exponential function parameters, while controlling the consumption of computing resources.

[0114] Furthermore, in step S104, data fluctuation is monitored in real time, a current data fluctuation value is obtained, and a determination is made based on a preset threshold whether the current data fluctuation value exceeds the threshold;

[0115] If the data fluctuation exceeds the preset threshold, the exponential function parameter optimization process is initiated;

[0116] Get the current exponential function parameters, optimize the parameters using the gradient descent method, and get new exponential function parameters;

[0117] Applying the new exponential function parameters to the exponential amplification process to obtain processed data;

[0118] Calculate the accuracy of the processed data and determine whether it meets the accuracy requirements;

[0119] If the accuracy meets the requirements, retain the new exponential function parameters;

[0120] If the accuracy does not meet the requirements, continue to optimize the parameters until the accuracy requirements are met;

[0121] During the parameter optimization process, a resource monitoring module is used to monitor computing resource consumption in real time. If resource consumption exceeds the preset upper limit, the optimization algorithm is adjusted to reduce resource consumption.

[0122] Specifically, in step S104, data fluctuations are monitored in real time, and the current data fluctuation value is obtained by calculating the standard deviation, including calculating the standard deviation of the past 10 data points in the time series as 0.15;

[0123] Based on the preset threshold of 0.2, determine whether the current data fluctuation value exceeds the threshold. If the standard deviation of 0.15 is less than 0.2, the adjustment process will not be triggered;

[0124] If the standard deviation is 0.25 and exceeds the threshold, the exponential function parameter optimization process is started;

[0125] Get the current exponential function parameter α = 1.5, use the gradient descent method to iteratively optimize with a learning rate of 0.01, and get the new exponential function parameter α = 1.8;

[0126] The new exponential function parameter α = 1.8 is applied to the exponential amplification process, and the normalized data x = 0.7 is calculated to obtain the processed data e (1.8*0.7) =3.62;

[0127] The accuracy of the processed data is calculated and evaluated by the mean square error. If the mean square error is 0.02, which is less than the preset accuracy of 0.05, the new exponential function parameter α = 1.8 is retained;

[0128] If the mean square error is 0.06, which is greater than the preset accuracy, continue to optimize the parameters until the mean square error is below 0.05;

[0129] During the parameter optimization process, the resource monitoring module is used to monitor the CPU usage in real time. If the CPU usage exceeds 80%, the learning rate of the optimization algorithm is adjusted from 0.01 to 0.005 to reduce resource consumption.

[0130] Specifically also includes: Gradient descent parameter optimization formula:

[0131]

[0132] k: dynamic parameter of the exponential function (e.g., initial k = 1.0);

[0133] α: learning rate (e.g. α = 0.01), controls the parameter adjustment step size;

[0134] Loss: Loss function (such as mean square error) measures the accuracy of processed data.

[0135] like Figure 1-Figure 2 As shown, in step S105, principal component analysis is used to extract the principal components of the consistency scores of the three dimensions, and information redundancy is avoided by concentrating the dimension information. The weight coefficient of each principal component is calculated to objectively reflect the importance of each dimension.

[0136] Furthermore, in step S105, a three-dimensional consistency scoring dataset is obtained and a scoring matrix is ​​constructed, where the number of rows in the matrix is ​​the number of samples and the number of columns is the number of dimensions;

[0137] Calculate the covariance matrix of the rating matrix, which reflects the linear relationship between dimensions;

[0138] Perform eigenvalue decomposition on the covariance matrix, and the size of the eigenvalue represents the contribution rate of the principal component;

[0139] Sort by eigenvalue from largest to smallest and select the first two principal components whose cumulative contribution rate exceeds the preset threshold;

[0140] Calculate the principal component loading matrix, and the loading coefficient reflects the contribution of the original dimension to the principal component;

[0141] The principal component weights are calculated based on the loading matrix, and the weight values ​​reflect the importance of the principal components;

[0142] The principal component score formula is used to calculate the projection value of the sample on the principal component;

[0143] Through statistical analysis of historical data, the principal component weights are dynamically updated to reduce the impact of subjective factors on weight distribution;

[0144] Based on the dynamic indicators, determine whether the weight adjustment amount is within a reasonable range. If it is out of the range, recalculate it.

[0145] Specifically, in step S105, a three-dimensional consistency scoring dataset is obtained, and a scoring matrix is ​​constructed. The number of rows in the matrix is ​​the number of samples, and the number of columns is the number of dimensions. For example, if the dataset contains 100 samples and three dimensions, the scoring matrix is ​​a 100×3 matrix.

[0146] Calculate the covariance matrix of the rating matrix. The covariance matrix reflects the linear relationship between dimensions, including an element in the matrix of 0.85, indicating that there is a strong positive correlation between two dimensions;

[0147] The covariance matrix is ​​decomposed by eigenvalues. The size of the eigenvalues ​​represents the contribution rate of the principal component. The eigenvalues ​​are 3.2, 1.5, and 0.3 respectively. The first principal component has the largest contribution rate.

[0148] Sorting by eigenvalue from largest to smallest, the first two principal components are selected, and the cumulative contribution rate exceeds the preset threshold, including the threshold being set at 85%, and the cumulative contribution rate of the first two principal components is 92%;

[0149] The principal component loading matrix was calculated. The loading coefficient reflects the contribution of the original dimension to the principal component, including the first principal component whose loadings on the three dimensions were 0.78, 0.65, and 0.72, respectively;

[0150] The principal component weights are calculated based on the loading matrix. The weight values ​​reflect the importance of the principal components, including the first principal component weight of 0.62 and the second principal component weight of 0.38;

[0151] The principal component score formula is used to calculate the projection value of the sample on the principal component, including a sample with a score of 1.25 on the first principal component and 0.87 on the second principal component;

[0152] Through statistical analysis of historical data, the principal component weights are dynamically updated to reduce the impact of subjective factors on weight distribution. For example, based on 1,000 pieces of historical data, the weight of the first principal component was adjusted to 0.58, and the weight of the second principal component was adjusted to 0.42.

[0153] Based on the dynamic indicators, determine whether the weight adjustment amount is within a reasonable range. If it exceeds the range, recalculate it, including the dynamic indicators being ±10%. The adjusted weight is within a reasonable range and does not need to be recalculated.

[0154] Specifically also includes: Covariance matrix calculation:

[0155]

[0156] X, Y: parameters of different dimensions (such as voltage and capacity);

[0157] μX, μY: the mean of the two dimensions;

[0158] n: number of samples (e.g. 100 battery samples);

[0159] Principal component analysis (PCA) eigenvalue decomposition: covariance matrix → eigenvalue λ and eigenvector;

[0160] λ: eigenvalue, indicating the variance contribution of the principal component (e.g., λ1 = 3.2, contribution rate 60%);

[0161] Eigenvector: principal component direction (for example, the first principal component direction is [0.7, 0.5, 0.3], indicating that voltage has the highest weight).

[0162] like Figure 1-Figure 2 As shown, in step S106, during the calculation of the principal component weights, the principal component weights are dynamically updated through statistical analysis of historical data to reduce the impact of human subjective factors on weight distribution and ensure the rationality of weight distribution.

[0163] Furthermore, in step S106, a historical data set is obtained and an initial scoring matrix is ​​constructed, where the number of rows in the matrix is ​​the number of samples and the number of columns is the number of dimensions;

[0164] Calculate the covariance matrix of the initial rating matrix. The covariance matrix reflects the linear relationship between dimensions.

[0165] Perform eigenvalue decomposition on the covariance matrix, and the size of the eigenvalue represents the contribution rate of the principal component;

[0166] Sort by eigenvalue from largest to smallest and select the first two principal components whose cumulative contribution rate exceeds the preset threshold;

[0167] Calculate the principal component loading matrix, and the loading coefficient reflects the contribution of the original dimension to the principal component;

[0168] Calculate the dynamic change characteristics of historical data based on statistical methods and determine the range of weight adjustment;

[0169] If the amount of data exceeds the preset threshold, the principal component eigenvector is updated using an analytical method to calculate a new weight distribution scheme;

[0170] Based on the dynamic indicators, determine whether the weight adjustment amount is within a reasonable range. If it exceeds the range, recalculate;

[0171] The distribution method is used to apply the adjusted weight values ​​to the principal components to generate new weight distribution results.

[0172] Specifically, in step S106, the historical data set is obtained and the initial scoring matrix is ​​constructed. The number of rows in the matrix is ​​the number of samples and the number of columns is the number of dimensions. 1000 samples are extracted from the three-dimensional consistency scoring data set. Each sample contains the scoring values ​​of three dimensions. The covariance matrix of the initial scoring matrix is ​​calculated. The covariance matrix reflects the linear relationship between the dimensions. The formula is used. Calculate the covariance between dimensions;

[0173] The covariance matrix is ​​decomposed into eigenvalues. The eigenvalues ​​represent the contribution rate of the principal components. The eigenvalues ​​obtained after decomposition are 2.5, 1.8, and 0.7 respectively.

[0174] Sorting by eigenvalue from largest to smallest, the first two principal components are selected, and the cumulative contribution rate exceeds the preset threshold of 80%;

[0175] Calculate the principal component loading matrix. The loading coefficient reflects the contribution of the original dimension to the principal component. The loading coefficient of the first principal component is [0.6, 0.7, 0.3], and the loading coefficient of the second principal component is [0.4, 0.2, 0.8].

[0176] Calculate the dynamic change characteristics of historical data based on statistical methods and determine the range of weight adjustment, including calculating the standard deviation and mean to determine the range of weight adjustment to be ±0.1;

[0177] If the amount of data exceeds a preset threshold, an analytical method is used to update the principal component eigenvectors and calculate a new weight distribution scheme, including recalculating the eigenvectors using the singular value decomposition method;

[0178] Based on the dynamic indicators, determine whether the weight adjustment is within a reasonable range. If it is out of range, recalculate, including setting a reasonable range of 0.05 to 0.15;

[0179] The distribution method is used to apply the adjusted weight values ​​to the principal components to generate new weight distribution results, including adjusting the weight of the first principal component to 0.6 and the weight of the second principal component to 0.4.

[0180] like Figure 1-Figure 2 As shown, in step S107, based on the weight coefficients of each dimension obtained by principal component analysis, a weighted average method is used to calculate the comprehensive consistency score, and the consistency performance of each dimension is quantified into a comprehensive evaluation index.

[0181] Furthermore, in step S107, a three-dimensional consistency scoring dataset is obtained and a scoring matrix is ​​constructed, where the number of rows in the matrix is ​​the number of samples and the number of columns is the number of dimensions;

[0182] Calculate the covariance matrix of the rating matrix, which reflects the linear relationship between dimensions;

[0183] Perform eigenvalue decomposition on the covariance matrix, and the size of the eigenvalue represents the contribution rate of the principal component;

[0184] Sort by eigenvalue from largest to smallest and select the first two principal components whose cumulative contribution rate exceeds the preset threshold;

[0185] Calculate the principal component loading matrix, and the loading coefficient reflects the contribution of the original dimension to the principal component;

[0186] The principal component weights are calculated based on the loading matrix, and the weight values ​​reflect the importance of the principal components;

[0187] The principal component score formula is used to calculate the projection value of the sample on the principal component;

[0188] Based on the weight coefficients of each dimension obtained by principal component analysis, the weighted average method was used to calculate the comprehensive consistency score;

[0189] Compare the comprehensive consistency score with the preset threshold to judge the consistency performance of the sample and generate the final evaluation result.

[0190] Specifically, in step S107 , a three-dimensional consistency scoring data set is obtained, including scoring data of 100 samples and 3 dimensions, and a 100×3 scoring matrix is ​​constructed;

[0191] Calculate the covariance matrix of the rating matrix using Reflects the linear relationship between dimensions;

[0192] Perform eigenvalue decomposition on the covariance matrix, including obtaining eigenvalues ​​λ1 = 2.5, λ2 = 1.8, λ3 = 0.7, which represent the principal component contribution rate;

[0193] Sorting by eigenvalue from largest to smallest, the first two principal components are selected, and the cumulative contribution rate exceeds the preset threshold of 85%;

[0194] The principal component loading matrix was calculated, including the loading coefficients of 0.6, 0.3, and 0.1, reflecting the contribution of the original dimensions to the principal components;

[0195] The principal component weights were calculated based on the loading matrix, including weight values ​​of 0.7 and 0.3, reflecting the importance of the principal components;

[0196] The principal component score formula is used to calculate the projection value of the sample on the principal component;

[0197] Based on the weight coefficients of each dimension obtained by principal component analysis, the weighted average method was used to calculate the comprehensive consistency score, and the formula was Score = 0.7 × PC1 + 0.3 × PC2;

[0198] The comprehensive consistency score is compared with the preset threshold of 0.8 to judge the consistency performance of the sample and generate the final evaluation result.

[0199] Specifically, it also includes: Weighted average comprehensive score formula: Score = ω1×PC1+ω2×PC2

[0200] w1, w2: principal component weights (e.g. w1 = 0.7, w2 = 0.3), determined by the proportion of eigenvalues;

[0201] PC1, PC2: Projection values ​​of the sample on the principal components (e.g. PC1 = 1.5, PC2 = 0.8).

[0202] like Figure 1-Figure 2 As shown, in step S108, for batteries of different batches and models, the interval boundaries of the normalization processing are dynamically adjusted according to the maximum and minimum value ranges of their historical data to ensure the adaptability and robustness of the evaluation process to changes in production conditions.

[0203] Furthermore, in step S108, the batch numbers and model values ​​of different batches of batteries are obtained, and the maximum and minimum values ​​corresponding to each batch model are extracted from the historical data;

[0204] The normalized interval boundaries are calculated based on the maximum and minimum values, and the interval values ​​are updated using a dynamic adjustment algorithm;

[0205] Based on the updated interval value, the historical value of the current batch of batteries is normalized to obtain a normalized value;

[0206] Determine whether the normalized value is within the preset boundary value range. If not, readjust the interval boundary.

[0207] According to the adjusted interval boundaries, the production values ​​are normalized to generate the adaptation values;

[0208] The robustness test algorithm is used to evaluate the fitness value and determine whether the robustness value meets the preset standard;

[0209] If the robust value meets the standard, the final normalized result is output;

[0210] If it does not meet the requirements, recalculate the interval boundaries and repeat the above steps;

[0211] Use historical data to construct the initial data set and obtain the principal component eigenvectors and corresponding weight values;

[0212] Calculate the dynamic change characteristics of historical data based on statistical methods and determine the range of weight adjustment.

[0213] Specifically, in step S108, historical data of battery batch number B001 and model number T01 is obtained through database query, and the maximum value of its voltage parameter is extracted as 4.2V and the minimum value is 3.0V;

[0214] Using a dynamic adjustment algorithm, the normalized interval boundary is set to [3.0, 4.2] and the interval value is updated;

[0215] Based on the updated interval boundaries, the historical voltage values ​​of the current batch of batteries are normalized to obtain a normalized value of 0.85;

[0216] Determine whether the normalized value is within the preset boundary value range [0,1]. If not, readjust the interval boundary to [2.8,4.3].

[0217] The production batch voltage values ​​were normalized according to the adjusted interval boundaries, generating an adaptation value of 0.92;

[0218] The Z-score method in the robustness test algorithm is used to calculate the robustness value of the fitness value, which is 1.2, to determine whether it meets the preset standard threshold of 1.5;

[0219] If the robust value meets the standard, the final normalized result is output; if not, the interval boundary is recalculated to [2.9, 4.2] and the above steps are repeated;

[0220] The initial data set was constructed using historical data, and the principal component eigenvectors [0.7, 0.3] and corresponding weight values ​​[0.6, 0.4] were extracted;

[0221] The dynamic change characteristics of historical data are calculated based on statistical methods, and the range of weight adjustment is determined to be ±0.1.

[0222] Specifically also includes: Dynamic normalization interval adjustment formula:

[0223]

[0224] X: original parameters of the current batch of batteries (such as voltage 3.6V);

[0225] Xmin, Xmax: The minimum and maximum values ​​of the parameter in the historical data (e.g. voltage range 3.0V-4.2V).

[0226] Those skilled in the art can make various other corresponding changes and deformations based on the technical solutions and concepts described above, and all of these changes and deformations should fall within the scope of protection of the claims of this application.

Claims

1. A method for quantifying a battery consistency score, characterized in that: The following steps are involved: S101. For the data of the three dimensions of voltage, internal resistance and capacity, the Z-score normalization method is used to perform normalization processing to obtain normalized data with unified dimensions; S102. In the normalized data, outlier data points of each dimension are automatically identified using the interquartile range method and are removed from the subsequent analysis process; S103. For the normalized data after removing outliers, a nonlinear adjustment is performed using an exponential function with the natural constant e as the base, where the exponent is the normalized data value. Through the exponential base selection and exponential dynamic adjustment mechanism, new data with amplified inter-dimensional differences is obtained; S104: In the exponential amplification process, the fluctuation of the current data is determined according to the data fluctuation threshold. If the fluctuation exceeds the preset threshold, the exponential parameter adjustment process is triggered to adjust the calculation accuracy by optimizing the exponential function parameters. S105. Use principal component analysis to extract principal components from the consistency scores of the three dimensions, condense the dimension information to avoid information redundancy, and calculate the weight coefficient of each principal component; S106. During the calculation of the principal component weights, dynamically updating the principal component weights through statistical analysis of historical data; S107. Based on the weight coefficients of each dimension obtained by principal component analysis, a weighted average method is used to calculate the comprehensive consistency score, and the consistency performance of each dimension is quantified into a comprehensive evaluation index; S108. For batteries of different batches and models, dynamically adjust the normalization processing interval boundaries according to the maximum and minimum value ranges of their historical data.

2. The method for quantifying a battery consistency score according to claim 1, wherein: Step S101 obtains the original data set of voltage value, internal resistance value and capacity value, and calculates the mean and standard deviation of each dimension data; The Z-score normalization method is used to normalize the voltage values ​​to eliminate the data distribution differences caused by dimension differences; The Z-score normalization method is used to normalize the internal resistance value to eliminate the data distribution differences caused by dimension differences; The Z-score standardization method was used to normalize the capacity values ​​to eliminate the data distribution differences caused by dimension differences; Based on the standardized data set, a linear regression model of voltage and capacity values ​​is constructed to predict the battery capacity change trend; Using the normalized data set, a support vector machine model of internal resistance and voltage values ​​was established to determine the degree of battery performance degradation. The standardized data set is used to train a random forest model of voltage, internal resistance, and capacity to evaluate the battery health status. Based on the prediction results of the random forest model, the correlation between voltage, internal resistance and capacity is identified to analyze the variation pattern of battery performance; The prediction results of the machine learning model are used to obtain a battery performance evaluation report.

3. The method for quantifying a battery consistency score according to claim 1, wherein: The step S102 uses the minimum-maximum method to perform linear transformation on the original data to obtain the value range of each feature column; According to the numerical range of each feature column, calculate the first quartile and third quartile of each dimension data; The interquartile range method was used to determine the upper and lower thresholds of each dimension data; Identify outlier data points in each dimension based on upper and lower thresholds; Remove the identified outlier data points from the dataset to construct an outlier-free dataset; The natural constant e is used as the fixed base of the exponential function to establish a nonlinear amplification model; Dynamically adjust the exponential parameter of the exponential function based on the distribution characteristics of the data set without abnormalities; For each data dimension, calculate the amplification value after exponential function processing; Through the preset threshold range, it is determined whether it is necessary to further adjust the index parameters or reselect the index base for secondary amplification processing, so as to obtain a new data set in which the differences between dimensions are significantly amplified.

4. The method for quantifying a battery consistency score according to claim 1, wherein: The step S103 obtains a normalized data set after removing outliers and determines the dimensional range that needs to be processed; The natural constant e is used as the fixed base of the exponential function to establish a nonlinear amplification model; Dynamically adjust the exponential parameter of the exponential function according to the distribution characteristics of the normalized data; For each data dimension, calculate the amplification value after exponential function processing; Determine whether further adjustment of index parameters is required based on the preset threshold range; If there is insufficient difference between dimensions, the exponential base is reselected for secondary amplification; Obtain a new dataset in which the differences between dimensions are significantly amplified; Obtain a three-dimensional consistency score dataset and construct a score matrix, where the number of rows is the number of samples and the number of columns is the number of dimensions; Calculate the covariance matrix of the rating matrix, which reflects the linear relationship between dimensions.

5. The method for quantifying a battery consistency score according to claim 1, wherein: The step S104 monitors data fluctuations in real time and obtains the current data fluctuation value; According to the preset threshold, determine whether the current data fluctuation value exceeds the threshold; If the data fluctuation exceeds the preset threshold, the exponential function parameter optimization process is initiated; Get the current exponential function parameters, optimize the parameters using the gradient descent method, and get new exponential function parameters; Applying the new exponential function parameters to the exponential amplification process to obtain processed data; Calculate the accuracy of the processed data and determine whether it meets the accuracy requirements; If the accuracy meets the requirements, retain the new exponential function parameters; If the accuracy does not meet the requirements, continue to optimize the parameters until the accuracy requirements are met; During the parameter optimization process, a resource monitoring module is used to monitor computing resource consumption in real time. If resource consumption exceeds the preset upper limit, the optimization algorithm is adjusted.

6. The method for quantifying a battery consistency score according to claim 1, wherein: In step S105, a three-dimensional consistency scoring data set is obtained and a scoring matrix is ​​constructed, where the number of rows in the matrix is ​​the number of samples and the number of columns is the number of dimensions; Calculate the covariance matrix of the rating matrix, which reflects the linear relationship between dimensions; Perform eigenvalue decomposition on the covariance matrix, and the size of the eigenvalue represents the contribution rate of the principal component; Sort by eigenvalue from largest to smallest and select the first two principal components whose cumulative contribution rate exceeds the preset threshold; Calculate the principal component loading matrix, and the loading coefficient reflects the contribution of the original dimension to the principal component; The principal component weights are calculated based on the loading matrix, and the weight values ​​reflect the importance of the principal components; The principal component score formula is used to calculate the projection value of the sample on the principal component; Through statistical analysis of historical data, the principal component weights are dynamically updated to reduce the impact of subjective factors on weight distribution; Based on the dynamic indicators, determine whether the weight adjustment amount is within a reasonable range. If it is out of the range, recalculate it.

7. The method for quantifying a battery consistency score according to claim 1, wherein: In step S106, a historical data set is obtained and an initial scoring matrix is ​​constructed, where the number of rows in the matrix is ​​the number of samples and the number of columns is the number of dimensions; Calculate the covariance matrix of the initial rating matrix. The covariance matrix reflects the linear relationship between dimensions. Perform eigenvalue decomposition on the covariance matrix, and the size of the eigenvalue represents the contribution rate of the principal component; Sort by eigenvalue from largest to smallest and select the first two principal components whose cumulative contribution rate exceeds the preset threshold; Calculate the principal component loading matrix, and the loading coefficient reflects the contribution of the original dimension to the principal component; Calculate the dynamic change characteristics of historical data based on statistical methods and determine the range of weight adjustment; If the amount of data exceeds the preset threshold, the principal component eigenvector is updated using an analytical method to calculate a new weight distribution scheme; Based on the dynamic indicators, determine whether the weight adjustment amount is within a reasonable range. If it exceeds the range, recalculate; The distribution method is used to apply the adjusted weight values ​​to the principal components to generate new weight distribution results.

8. The method for quantifying a battery consistency score according to claim 1, wherein: The step S107 obtains a three-dimensional consistency scoring data set and constructs a scoring matrix, where the number of rows in the matrix is ​​the number of samples and the number of columns is the number of dimensions; Calculate the covariance matrix of the rating matrix, which reflects the linear relationship between dimensions; Perform eigenvalue decomposition on the covariance matrix, and the size of the eigenvalue represents the contribution rate of the principal component; Sort by eigenvalue from largest to smallest and select the first two principal components whose cumulative contribution rate exceeds the preset threshold; Calculate the principal component loading matrix, and the loading coefficient reflects the contribution of the original dimension to the principal component; The principal component weights are calculated based on the loading matrix, and the weight values ​​reflect the importance of the principal components; The principal component score formula is used to calculate the projection value of the sample on the principal component; Based on the weight coefficients of each dimension obtained by principal component analysis, the weighted average method was used to calculate the comprehensive consistency score; Compare the comprehensive consistency score with the preset threshold to judge the consistency performance of the sample and generate the final evaluation result.

9. The method for quantifying a battery consistency score according to claim 1, wherein: In step S108, the batch numbers and model values ​​of different batches of batteries are obtained, and the maximum and minimum values ​​corresponding to each batch model are extracted from the historical data; The normalized interval boundaries are calculated based on the maximum and minimum values, and the interval values ​​are updated using a dynamic adjustment algorithm; Based on the updated interval value, the historical value of the current batch of batteries is normalized to obtain a normalized value; Determine whether the normalized value is within the preset boundary value range. If not, readjust the interval boundary. According to the adjusted interval boundaries, the production values ​​are normalized to generate the adaptation values; The robustness test algorithm is used to evaluate the fitness value and determine whether the robustness value meets the preset standard; If the robust value meets the standard, the final normalized result is output; If it does not meet the requirements, recalculate the interval boundaries and repeat the above steps; Use historical data to construct the initial data set and obtain the principal component eigenvectors and corresponding weight values; Calculate the dynamic change characteristics of historical data based on statistical methods and determine the range of weight adjustment.

Images