Classification method and system for re-matching applicable scenes for retired batteries

Abstract

The invention relates to the technical field of echelon utilization of decommissioned batteries, in particular to a classification method and system for re-matching application scenes for decommissioned batteries. According to the proposed two-layer model structure, firstly, preliminary screening is carried out through SVM, whether the battery is suitable for echelon utilization or not is determined, and then finer scene classification is carried out through Softmax regression. The structure not only embodies a hierarchical decision-making process, but also allows more detailed purpose matching of the battery. The method has the uniqueness that the method combines the advantages of two different types of models, the high efficiency of the SVM for the dichotomy problem, and the Softmax regression model can reflect the flexibility in the multi-classification task and can perform scene classification on the decommissioned battery meeting the echelon utilization condition, so that the re-applicable application scene of the decommissioned battery is determined. Therefore, the retired batteries can be effectively distributed to the most suitable application scene, and the maximum utilization of resources is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of cascade utilization of retired batteries, and in particular to a classification method and system for re-matching applicable scenarios for retired batteries. Background Art

[0002] The cascade utilization of retired batteries can reduce the production cost of the new energy vehicle industry and is conducive to the large-scale promotion and application of other battery usage scenarios. The cascade utilization and recycling of batteries can create economic value and improve the efficiency of resource utilization. Retired batteries can be used in energy storage, base station backup power, two / three-wheel electric vehicles, photovoltaic street lights, portable mobile power supplies and UPS power supplies. Research on how to effectively classify these batteries and use them in appropriate application scenarios can meet market demand and enhance their application value.

[0003] By properly classifying and treating retired batteries, heavy metals and chemicals contained in retired batteries can be prevented from polluting the environment. If retired batteries are not properly handled, they may cause long-term pollution to soil, water sources, etc. At the same time, re-matching retired batteries with cascade utilization scenarios can greatly reduce the production demand for new batteries, thereby reducing the exploitation of natural resources and environmental damage. The existing classification matching adopts a single-layer classification scheme, and uses the classification results of the SVM model based on three different kernel functions for decision fusion, and the model is single; secondly, the scheme proposes to classify by weight distribution through decision fusion, and the results are simply excellent, good, and general. The categories are limited, and no specific cascade utilization scenarios are divided. That is to say, even if three levels of battery quality are divided, it is not known which scenarios these three levels of classification can be reused; thirdly, the scheme uses the same classification model (SVM) based on three different kernel functions to jointly decide the classification results, which means that a battery has to be tested using three functions, which increases the steps and time of classification. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is to solve the problem of retired batteries that meet the requirements of cascade utilization and re-match applicable scenarios to enter the market for the second time.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a classification method for re-matching applicable scenarios for retired batteries, which includes collecting and collating data on the cascade utilization of retired batteries studied by previous scholars to construct a new data set that meets the cascade utilization requirements;

[0008] After preprocessing the data set, a layer of SVM model training is performed;

[0009] Export the trained SVM model to a file and apply it to the classification system;

[0010] Determine the applicable scenarios for retired batteries, and establish a data set for second-layer model training by collecting national standards on battery usage requirements for each scenario;

[0011] After data preprocessing, the dataset is trained with a two-layer Softmax regression model;

[0012] Export the trained Softmax regression model to a file and use it in the classification system.

[0013] As a preferred solution of the classification method for re-matching the applicable scenarios of retired batteries described in the present invention, the collected data feature values ​​include basic information, usage history, performance parameters, and safety of retired batteries, and the corresponding data categories are divided into those that meet the requirements of cascade utilization and those that do not meet the requirements of cascade utilization, and then the data set is cleaned to process missing values ​​and abnormal values;

[0014] The data set is preprocessed and then trained with a layer of SVM model. The specific implementation of the layer of SVM model adopts the following steps:

[0015] Data preparation: Collect basic information including but not limited to retired batteries, usage history, performance parameters, and safety, and then clean the data set to handle missing values ​​and outliers;

[0016] Data cleaning: Clean the data to remove invalid or abnormal data points;

[0017] Feature selection: Select the most relevant features from the integrated dataset for input to the SVM model;

[0018] The Z-score normalization method is used to scale the features to the same scale. For each feature, the normalized feature can be expressed as:

[0019]

[0020] Among them, X j ′ represents the standardized feature; X j Indicates characteristics; μ j Represents feature X j The average value, σ j Represents feature X j The standard deviation of

[0021] Then, the correlation analysis between the features is performed and the correlation coefficient matrix between the features is calculated to evaluate the relationship between the many features. The correlation coefficient indicates the degree of correlation between features i and j:

[0022]

[0023] Among them, r ij represents the correlation coefficient, that is, the degree of correlation between features i and j; x ik 、x jk Respectively represent the values ​​of feature i and feature j in the kth sample; represents the sample mean of feature i and feature j; n represents the number of samples, that is, the total number of observations in the data set; k represents the sample index, a cyclic variable from 1 to n;

[0024] According to the correlation coefficient matrix, features with high correlation are selected for retention, and a threshold is set to 0.5, that is, only features with correlation coefficients greater than 0.5 are retained. According to the results of correlation analysis, the features are ranked by importance, and then relatively important features are selected to form a dataset subset for input to the SVM binary classification model. The remaining capacity, maximum voltage, and internal resistance are selected as the main feature vectors, and the feature vector of each battery can be expressed as:

[0025] X=(x1,x2,x3)=(remaining capacity, maximum voltage, internal resistance)

[0026] Build the SVM model:

[0027] For binary classification, that is, linear separability, the objective function of the SVM model can be expressed as:

[0028]

[0029] Among them, W is the weight vector, b is the bias term, C is the penalty parameter, and T represents the transpose of the W weight vector; in order to allow some feature points to violate the classification boundary, ε is added i is the slack variable;

[0030] The constraints of the above objective function can be expressed as:

[0031] y i (W T X i +b)≥1-ε i ,i=1,2,...,n

[0032] Since the data is linearly separable, there is no need to use a complex kernel function, only a simple linear kernel function is needed, so the decision function of the binary classification model is:

[0033] f(x)=sign(WT X+b)

[0034] Classification Decision:

[0035] After model training, for any new battery feature vector xj, the SVM binary classification decision model can be used to determine whether the battery meets the basic conditions for cascade utilization;

[0036] If f(x)′>0, the battery meets the requirements of cascade utilization, and can enter the subsequent scene judgment classification; if f(x)′<0, the battery does not meet the requirements of cascade utilization, and is discarded, and the metal is recycled for reuse;

[0037] In a single-layer SVM model, retired batteries whose output results do not meet the conditions for cascade utilization will be directly eliminated and will no longer enter the next stage.

[0038] As a preferred solution of the classification method for re-matching applicable scenarios for retired batteries described in the present invention, the training includes: training the data set after data preprocessing for a two-layer Softmax regression model; training the two-layer Softmax regression model, and continuously optimizing the model accuracy through parameter tuning;

[0039] The features are standardized using the Z-score standardization method to ensure that all features have the same scale. The standardized features can be expressed as:

[0040]

[0041] Among them, X ij ′ represents the jth feature of each i-th sample, μ j represents the average value of the jth feature, σ j represents the standard deviation of the jth feature;

[0042] SoftMax regression model:

[0043] Initialize the weight matrix:

[0044] W∈R d×K

[0045] Where d is the number of features and K is the number of categories;

[0046] b represents the bias vector:

[0047] b∈R K

[0048] Forward propagation:

[0049] After calculating the linear combination of the input features and model weights for each category, plus the bias vector, the expression can be:

[0050]

[0051] Among them, W is the model weight, b represents the bias vector, i represents the category, and x represents the feature;

[0052] SoftMax function:

[0053] Use the SoftMax function to transform the result of the linear combination into a normalized overview distribution:

[0054]

[0055] Where p(y=i|x) represents the probability that a sample belongs to category i given an input x, and K is the total number of categories;

[0056] Loss function:

[0057] Cross entropy loss function: For each sample, the loss function can be expressed as:

[0058] L i = -log(p(y = y i |x i ))

[0059] Total loss function:

[0060] Based on the entire training set, the total loss function is the average of all cross entropy functions, that is, the average of all sample losses:

[0061]

[0062] Backward Propagation:

[0063] Calculate gradients: Calculate the gradient of the loss function with respect to each weight wij and bias:

[0064]

[0065] Among them: W ij represents each individual weight; b i represents the bias gradient; δ(y k ,j) is the indicator function, and when y k =i is 1, otherwise it is 0;

[0066] Parameter update:

[0067] Gradient Descent: Update weights and biases using gradient descent:

[0068]

[0069] Where η is the learning rate.

[0070] As a preferred solution of the classification method for re-matching applicable scenarios for retired batteries described in the present invention, the trained two-layer Softmax regression model can perform scenario classification on retired batteries that meet the requirements of cascade utilization, that is, re-match the applicable scenarios of retired batteries, and then re-enter the market, determine the data type of the data feature set in the data set, and facilitate the subsequent output of the classification results.

[0071] As a preferred solution of the classification method for re-matching the applicable scenarios of retired batteries described in the present invention, wherein: the retired batteries to be tested are sent into the system, and the appearance of the retired batteries to be classified is inspected by a high-definition camera;

[0072] The retired batteries that have passed the appearance inspection are then sent to the battery parameter inspection stage, which is mainly used to detect various parameters of the retired batteries, including battery performance and reliability;

[0073] After passing the battery detection system, all detection parameters of the retired battery are collected and input into a first-layer binary classification model to make a cascade utilization judgment;

[0074] Then the detection parameters of retired batteries that meet the requirements of cascade utilization are input into the two-layer multi-classification model;

[0075] According to the classification results, the retired batteries are sorted into corresponding applicable scenarios.

[0076] As a preferred solution of the classification method for re-matching the applicable scenarios of retired batteries described in the present invention, the detection appearance has no scratches, deformation, damage and leakage, the positive and negative electrodes have no rust, and the markings are correct and clear; if the conditions are met, the next step is entered, if not, the batteries are discarded and no longer reused;

[0077] Through the battery detection system, the system is mainly used to detect various parameters of retired batteries, including battery performance and reliability;

[0078] Collect all the detection parameters of the retired battery, input the parameters into the first-layer binary classification model, and output the retired batteries that do not meet the conditions for cascade utilization, which will be directly eliminated and will not enter the next stage;

[0079] The battery performance test parameter range includes charge and discharge voltage and capacity at different temperatures, charge and discharge energy at different rates, DC internal resistance, storage performance, standard cycle life, SOC calibration, peak power, and energy efficiency;

[0080] The reliability testing parameters range includes electrical insulation, flame retardancy, thermal management, high temperature and high humidity.

[0081] As a preferred solution of the classification method for re-matching applicable scenarios for retired batteries described in the present invention, wherein: the detection parameters of retired batteries that meet the requirements of cascade utilization are input into a two-layer multi-classification model, and the classification results of the retired batteries can be output, thereby obtaining applicable scenarios for battery cascade utilization, and then entering the market for the second time;

[0082] After passing through the battery detection system, the detection parameters of the retired battery are collected and numbered, and the series of parameters are passed through a one-layer binary classification model to screen the battery for cascade utilization;

[0083] The detection parameters of the screened retired batteries are input into a two-layer multi-classification model to obtain the re-matching results of the retired batteries with the same number, so that they can re-enter the market.

[0084] In a second aspect, an embodiment of the present invention provides a classification system for re-matching applicable scenarios for retired batteries, which includes a construction module, which collects and organizes data on the cascade utilization of retired batteries by previous scholars to construct a new data set that meets the cascade utilization requirements;

[0085] The training module performs a layer of SVM model training on the data set after preprocessing;

[0086] The output module exports the trained SVM model to a file and applies it to the classification system;

[0087] The matching module determines the applicable scenarios for retired batteries and establishes a data set for second-layer model training by collecting national standards on battery usage requirements for each scenario;

[0088] The processing module trains the two-layer Softmax regression model after data preprocessing.

[0089] The classification module exports the trained Softmax regression model to a file and applies it to the classification system.

[0090] In a third aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of a classification method for re-matching applicable scenarios for retired batteries as described in the first aspect of the present invention are implemented.

[0091] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of a classification method for re-matching applicable scenarios for retired batteries as described in the first aspect of the present invention are implemented.

[0092] The beneficial effects of the present invention are: supplementing the back-end deficiencies of the prior art. At present, most scholars study whether retired batteries meet the cascade utilization standards; or improve the accuracy of cascade utilization judgment parameters; based on this research, the present invention uses a layer of SVM two-classification model to determine whether retired batteries meet the cascade utilization standards, and then inputs batteries that meet the cascade utilization standards into a two-layer Softmax regression multi-classification model, and then outputs the applicable scenarios of retired batteries, so that they can re-enter the market and be reused.

[0093] The system structure is clear and the algorithm is robust. The one-layer SVM binary classification model has been proven to provide high-accuracy classification results on a variety of data sets, and the theoretical basis of the SVM algorithm is structural risk minimization, which makes it perform well in avoiding overfitting and underfitting. Compared with other complex machine learning algorithms, SVM has only a few parameters that need to be adjusted (such as C, kernel function and its parameters), which simplifies the model selection and parameter adjustment process. The SVM algorithm can also perform well on small sample data, which is very useful when data acquisition costs are high or data is scarce, and its optimization problem has a mature solution (SMO); the two-layer Softmax regression multi-classification model has a simple structure, is easy to understand and implement, and unlike binary classification algorithms (such as logistic regression), Softmax regression does not require an expansion strategy and can be directly applied to multi-classification problems. At the same time, with the continuous update of market battery utilization scenarios, Softmax regression can also be efficiently expanded to large-scale data sets. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0095] Figure 1 A flowchart of a classification method for re-matching applicable scenarios for retired batteries;

[0096] Figure 2 A computer equipment diagram for a classification method for re-matching applicable scenarios for retired batteries;

[0097] Figure 3 A system-level SVM two-classification model flow chart for a classification method for re-matching applicable scenarios for retired batteries;

[0098] Figure 4 A flow chart of a systematic two-layer Softmax regression multi-classification model for a classification method for re-matching applicable scenarios for retired batteries. DETAILED DESCRIPTION

[0099] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0100] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0101] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0102] Example 1

[0103] Reference Figure 1-2 , which is the first embodiment of the present invention, and provides a classification method for re-matching applicable scenarios for retired batteries, including:

[0104] S100: By collecting and arranging the data of previous scholars on the cascade utilization of retired batteries, a new dataset that meets the requirements of cascade utilization is constructed;

[0105] S101: Collect data feature values ​​including basic information, usage history, performance parameters, and safety of retired batteries. The corresponding data categories are divided into those that meet the requirements for cascade utilization and those that do not meet the requirements for cascade utilization. Then, the data set is cleaned to process missing values ​​and outliers.

[0106] Preferably, the preliminary screening aims to quickly identify whether the retired batteries are suitable for cascade utilization in accordance with the national standard GB / T340152017, so as to ensure the safety and effectiveness of the subsequent utilization process.

[0107] Appearance inspection: Visually inspect the battery for obvious physical damage, leakage, deformation or other abnormalities.

[0108] Performance test: Use professional testing equipment to perform battery charge and discharge tests in accordance with the procedures specified in GB / T340152017 standard, and record the battery performance parameters.

[0109] Qualification criteria: The criteria that define whether a battery is qualified, for example, the battery SOH needs to be above a certain threshold.

[0110] Unqualified treatment: Batteries that do not meet the qualified standards will be marked, excluded from subsequent cascade utilization, and directly discarded.

[0111] The main tests and records are as follows:

[0112] Charging test: Charge the battery according to the charging current, voltage and other conditions recommended by the standard.

[0113] Discharge test: Discharge the battery according to the discharge current, voltage and other conditions recommended by the standard.

[0114] Data recording: Record key performance parameters such as the battery's maximum voltage, discharge capacity, internal resistance, etc. to represent the battery's SOH parameters.

[0115]

[0116] The data collection format at this time is:

[0117]

[0118] Battery number: used to uniquely identify each battery.

[0119] New Capacity: The discharge capacity of the battery in its new state.

[0120] Current discharge capacity: The discharge capacity of the battery in its current state.

[0121] SOC: The battery state of charge percentage calculated based on the current discharge capacity and the rated capacity of a new product.

[0122] SOH: Calculated by comparing the current performance of the battery with that of a new battery or evaluated by parameters such as the battery's internal resistance, capacity, and self-discharge rate.

[0123] During the in-depth inspection phase, more battery information, such as maximum voltage, internal resistance, etc., is collected through electrical performance detectors and merged into a new data set for training and testing the SVM model to determine whether the battery meets the basic conditions for cascade utilization.

[0124] S200: After preprocessing the data set, a layer of SVM model training is performed;

[0125] Preferably, many domestic retired battery detection and recycling companies can collect enough data sets based on this, and divide the validation set and training set into 3:7, and start training the SVM binary classification model. Considering that the application of this model in this system is to determine whether the battery meets the cascade utilization standard, a linear kernel is used to process the binary classification problem.

[0126] The linear SVM binary classification process is as follows:

[0127] Data integration: Integrate the remaining capacity data collected in the preliminary screening phase with the electrical performance data collected in the in-depth inspection phase to form a complete data set in the following data format:

[0128]

[0129] S201: After the data set is preprocessed, a layer of SVM model training is performed. The specific implementation of the layer of SVM model adopts the following steps:

[0130] Data preparation: Collect basic information including but not limited to retired batteries, usage history, performance parameters, and safety, and then clean the data set to handle missing values ​​and outliers;

[0131] Data cleaning: Clean the data to remove invalid or abnormal data points;

[0132] Feature selection: Select the most relevant features from the integrated dataset for input to the SVM model;

[0133] The Z-score normalization method is used to scale the features to the same scale. For each feature, the normalized feature can be expressed as:

[0134]

[0135] Among them, X j ′ represents the standardized feature; X j Indicates characteristics; μ j Represents feature X j The average value, σ j Represents feature X j The standard deviation of

[0136] Then, the correlation analysis between the features is performed and the correlation coefficient matrix between the features is calculated to evaluate the relationship between the many features. The correlation coefficient indicates the degree of correlation between features i and j:

[0137]

[0138] Among them, r ij represents the correlation coefficient, that is, the degree of correlation between features i and j; x ik 、x jk Respectively represent the values ​​of feature i and feature j in the kth sample; represents the sample mean of feature i and feature j; n represents the number of samples, that is, the total number of observations in the data set; k represents the sample index, a cyclic variable from 1 to n;

[0139] According to the correlation coefficient matrix, features with high correlation are selected for retention, and a threshold is set to 0.5, that is, only features with correlation coefficients greater than 0.5 are retained. According to the results of correlation analysis, the features are ranked by importance, and then relatively important features are selected to form a dataset subset for input to the SVM binary classification model. The remaining capacity, maximum voltage, and internal resistance are selected as the main feature vectors, and the feature vector of each battery can be expressed as:

[0140] X=(x1,x2,x3)=(remaining capacity, maximum voltage, internal resistance)

[0141] Build the SVM model:

[0142] For binary classification, that is, linear separability, the objective function of the SVM model can be expressed as:

[0143]

[0144] Among them, W is the weight vector, b is the bias term, C is the penalty parameter, and T represents the transpose of the W weight vector; in order to allow some feature points to violate the classification boundary, ε is added i is the slack variable;

[0145] The constraints of the above objective function can be expressed as:

[0146] y i (W T X i +b)≥1-ε i ,i=1,2,...,n

[0147] Since the data is linearly separable, there is no need to use a complex kernel function, only a simple linear kernel function is needed, so the decision function of the binary classification model is:

[0148] f(x)=sign(W T X+b)

[0149] Classification Decision:

[0150] After model training, for any new battery feature vector xj, the SVM binary classification decision model can be used to determine whether the battery meets the basic conditions for cascade utilization;

[0151] If f(x)′>0, the battery meets the requirements of cascade utilization, and can enter the subsequent scene judgment classification; if f(x)′<0, the battery does not meet the requirements of cascade utilization, and is discarded, and the metal is recycled for reuse;

[0152] In a single-layer SVM model, retired batteries whose output results do not meet the conditions for cascade utilization will be directly eliminated and will no longer enter the next stage.

[0153] Through this link, it is possible to deeply determine whether the battery that meets the residual energy detection meets the requirements of cascade utilization, avoiding the subsequent waste of resources due to unstable prediction, and at the same time improving the effectiveness of cascade utilization of retired batteries. So far, the present invention has obtained all the battery information that can be used for cascade utilization, and will then enter the final link of re-matching the usage scenario.

[0154] S300: export the trained SVM model to a file and apply it to the classification system;

[0155] S400: Determine the applicable scenarios for retired batteries, and establish a data set for second-layer model training by collecting national standards on battery usage requirements for each scenario;

[0156] S500: training a two-layer Softmax regression model on the data set after data preprocessing;

[0157] S501: training the data set after data preprocessing to train a two-layer Softmax regression model; training the two-layer Softmax regression model, and continuously optimizing the model accuracy through parameter tuning;

[0158] The two-layer multi-classification model is implemented using the Softmax regression model as follows:

[0159] The features are standardized using the Z-score standardization method to ensure that all features have the same scale. The standardized features can be expressed as:

[0160]

[0161] Among them, X ij ′ represents the jth feature of each i-th sample, μ j represents the average value of the jth feature, σ j represents the standard deviation of the jth feature;

[0162] SoftMax regression model:

[0163] Initialize the weight matrix:

[0164] W∈R d×K

[0165] Where d is the number of features and K is the number of categories;

[0166] b represents the bias vector:

[0167] b∈R K

[0168] Forward propagation:

[0169] After calculating the linear combination of the input features and model weights for each category, plus the bias vector, the expression can be:

[0170] z i =W i T x+b i

[0171] Among them, W is the model weight, b represents the bias vector, i represents the category, and x represents the feature;

[0172] SoftMax function:

[0173] Use the SoftMax function to transform the result of the linear combination into a normalized overview distribution:

[0174]

[0175] Where p(y=i|x) represents the probability that a sample belongs to category i given an input x, and K is the total number of categories;

[0176] Loss function:

[0177] Cross entropy loss function: For each sample, the loss function can be expressed as:

[0178] L i = -log(p(y = y i |x i ))

[0179] Total loss function:

[0180] Based on the entire training set, the total loss function is the average of all cross entropy functions, that is, the average of all sample losses:

[0181]

[0182] Backward Propagation:

[0183] Calculate gradients: Calculate the gradient of the loss function with respect to each weight wij and bias:

[0184]

[0185] Among them: W ij represents each individual weight; b i represents the bias gradient; δ(y k ,j) is the indicator function, and when y k =i is 1, otherwise it is 0;

[0186] Parameter update:

[0187] Gradient Descent: Update weights and biases using gradient descent:

[0188]

[0189] Where η is the learning rate.

[0190] S600: Export the trained Softmax regression model to a file and apply it to the classification system.

[0191] S601: The trained two-layer Softmax regression model can classify retired batteries that meet the requirements of second-life utilization, that is, re-match the applicable scenarios of retired batteries, and then re-enter the market to determine the data type of the data feature set in the data set, so as to facilitate the subsequent output of classification results.

[0192] Preferably, the Softmax multi-classification regression model includes 1. Data preparation: Data set construction: According to the national standard requirements for battery use in various scenarios, a data set is constructed that includes batteries that meet the conditions for cascade utilization and requirements for different application scenarios.

[0193] Assume that the requirements for each scenario are as follows:

[0194] Energy storage: Remaining capacity>2400mAh, internal resistance<0.06Ω

[0195] Base station backup power: remaining capacity>2200mAh, internal resistance<0.07Ω

[0196] Portable power bank: Remaining capacity>2000mAh, internal resistance<0.08Ω

[0197] Battery car: Remaining capacity>1800mAh, internal resistance<0.1Ω

[0198] Dataset example:

[0199] Battery ID Remaining capacity (mAh) Maximum voltage(V) Internal resistance(Ω) Scenario B001 2500 3.7 0.05 Energy Storage B002 2200 3.6 0.07 Base station backup power B003 2000 3.5 0.06 Portable Power B004 1800 3.4 0.08 Battery car

[0200] Feature selection: Select key performance parameters related to the application scenario, such as battery capacity, maximum voltage, internal resistance, etc.

[0201] Data preprocessing: Clean and standardize data to ensure data quality.

[0202] 2. Feature correlation observation and analysis

[0203] Feature selection: Refer to the feature engineering in the SVM model for feature selection, and select the most relevant features for model training based on the results of correlation analysis.

[0204] 3. Softmax regression model training

[0205] Data Splitting: Divide the dataset into training and testing sets.

[0206] Standardization: The features are standardized using the Z-score standardization method.

[0207] Model initialization: Initialize the model’s weight matrix W and bias vector b.

[0208] Model training: Use the Softmax regression model for training. The training process includes forward propagation, application of the Softmax function, loss function calculation, back propagation, and parameter update.

[0209] Loss Function: The cross entropy loss function is used to measure the difference between the probability distribution predicted by the model and the true label.

[0210] Parameter Update: Use an optimization algorithm such as gradient descent or Adam optimizer to update the weights and biases to reduce the value of the loss function.

[0211] 4. Model Evaluation

[0212] Test set evaluation: Use the trained model to make predictions on the test set and use appropriate evaluation metrics (such as accuracy, precision, recall, etc.) to evaluate model performance.

[0213] 5. Model Application

[0214] Model deployment: Add the trained Softmax regression model to the three-layer multi-classification link of the system.

[0215] Scenario classification: The detection parameters of retired batteries that meet the conditions for cascade utilization screened out by the SVM model are input into the Softmax regression model, and the classification results of retired batteries are output to determine the application scenarios in which retired batteries can be reused.

[0216] S602: Send the retired batteries to be tested into the system, and use a high-definition camera to perform appearance inspection on the retired batteries to be classified;

[0217] The retired batteries that have passed the appearance inspection are then sent to the battery parameter inspection stage, which is mainly used to detect various parameters of the retired batteries, including battery performance and reliability;

[0218] After passing the battery detection system, all detection parameters of the retired battery are collected and input into a first-layer binary classification model to make a cascade utilization judgment;

[0219] Then the detection parameters of retired batteries that meet the requirements of cascade utilization are input into the two-layer multi-classification model;

[0220] According to the classification results, the retired batteries are sorted into corresponding applicable scenarios.

[0221] S603: Check the appearance for scratches, deformation, damage and leakage, positive and negative electrodes without rust, and correct and clear markings; if the conditions are met, proceed to the next step; if not, the product will be discarded and will not be reused;

[0222] Through the battery detection system, the system is mainly used to detect various parameters of retired batteries, including battery performance and reliability;

[0223] Collect all the detection parameters of the retired battery, input the parameters into the first-layer binary classification model, and output the retired batteries that do not meet the conditions for cascade utilization, which will be directly eliminated and will not enter the next stage;

[0224] The battery performance test parameter range includes charge and discharge voltage and capacity at different temperatures, charge and discharge energy at different rates, DC internal resistance, storage performance, standard cycle life, SOC calibration, peak power, and energy efficiency;

[0225] The reliability testing parameters range includes electrical insulation, flame retardancy, thermal management, high temperature and high humidity.

[0226] S604: Input the detection parameters of the retired battery that meets the requirements of cascade utilization into a two-layer multi-classification model, and output the classification results of the retired battery, thereby obtaining the applicable scenarios for cascade utilization of the battery, and then entering the market for the second time;

[0227] After passing through the battery detection system, the detection parameters of the retired battery are collected and numbered, and the series of parameters are passed through a one-layer binary classification model to screen the battery for cascade utilization;

[0228] The detection parameters of the screened retired batteries are input into a two-layer multi-classification model to obtain the re-matching results of the retired batteries with the same number, so that they can re-enter the market.

[0229] Preferably, after the two-layer model is trained, the classification system of the present invention can be built and run, which is specifically implemented as follows:

[0230] The retired batteries to be tested are sent into the system, and the high-definition cameras are used to perform appearance inspection on the classified retired batteries.

[0231] Check the appearance for scratches, deformation, damage and leakage, positive and negative electrodes without rust, and correct and clear markings; if they meet the requirements, proceed to the next step; if not, they will be discarded and will not be reused.

[0232] The retired batteries that have passed the appearance inspection will then be sent to the battery parameter testing stage, which is mainly used to test various parameters of the retired batteries, including battery performance and reliability.

[0233] The battery performance detection parameter range includes but is not limited to the data features used to train the double-layer model.

[0234] After passing through the battery detection system, all detection parameters about the retired battery are collected and input into a one-layer binary classification model to make a cascade utilization judgment, and then determine whether the retired battery meets the cascade utilization requirements. If it meets the requirements, it will proceed to the next step, otherwise it will be directly eliminated.

[0235] The detection parameters of the retired batteries that meet the requirements of cascade utilization are then input into a two-layer multi-classification model, which can output the classification results of the retired batteries, indicating that the retired batteries can be re-applied to the classification scenario and then enter the market for the second time.

[0236] Furthermore, this embodiment also provides a classification system for re-matching applicable scenarios for retired batteries, including:

[0237] The construction module collects and organizes the data of previous scholars on the cascade utilization of retired batteries to build a new data set that meets the requirements of cascade utilization;

[0238] The training module performs a layer of SVM model training on the data set after preprocessing;

[0239] The output module exports the trained SVM model to a file and applies it to the classification system;

[0240] The matching module determines the applicable scenarios for retired batteries and establishes a data set for second-layer model training by collecting national standards on battery usage requirements for each scenario;

[0241] The processing module trains the two-layer Softmax regression model after data preprocessing.

[0242] The classification module exports the trained Softmax regression model to a file and applies it to the classification system.

[0243] This embodiment also provides a computer device, which is applicable to a classification method for re-matching applicable scenarios for retired batteries, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement a classification method for re-matching applicable scenarios for retired batteries as proposed in the above embodiment.

[0244] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0245] This embodiment also provides a storage medium having a computer program stored thereon. When the program is executed by a processor, a classification method for re-matching applicable scenarios for retired batteries is implemented as proposed in the above embodiment.

[0246] In summary, the purpose of the present invention is to solve the problem of retired batteries that meet the requirements of cascade utilization and re-match applicable scenarios to enter the market for the second time. It provides a classification method and system for re-matching applicable scenarios for retired batteries. First, the retired batteries are preliminarily screened with reference to the GB / T340152017 standard, mainly completing the detection of the appearance and residual energy of the retired batteries; at the same time, the appearance and residual energy of the batteries that pass the preliminary screening are recorded; and then sent to the trained one-layer two-classification model. In this link, the battery indicators are deeply detected to further determine whether the retired battery meets the retirement requirements; then, the characteristic parameters of the retired batteries that can be recycled are input into the trained two-layer multi-classification model, and then the market application scenarios that the retired batteries can be re-matched are determined through the output categories, and finally the batteries of the same type are reorganized and reused. The present invention reduces the safety hazards of recycled batteries, divides the scenarios of recycled retired batteries, can be directly matched, reduces the classification steps, and improves efficiency, providing a solution for the development of the retired battery detection and classification field.

[0247] Example 2

[0248] Reference Figure 2 - Figure 4 , which is the second embodiment of the present invention, provides a classification method for re-matching applicable scenarios for retired batteries. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0249] In order to verify the effectiveness of the classification method for retired battery re-matching applicable scenarios proposed by our solution, the following experiment was conducted. First, 500 retired power batteries were collected from the recycling center of a new energy vehicle manufacturer. These batteries came from different models and years of use. The experimental team conducted a preliminary screening in accordance with the GB / T34015-2017 standard and recorded the basic information, usage history and appearance of each battery.

[0250] Next, these batteries were deeply tested using professional battery testing equipment, measuring and recording key parameters of each battery, including maximum voltage, remaining capacity, internal resistance, SOC and SOH. To ensure the accuracy and consistency of the data, each test was repeated three times, and the average value was taken as the final result.

[0251] After the data collection was completed, the research team implemented the two-layer classification model of our solution using the Python programming language and the scikit-learn library. First, 80% of the data was used for training and 20% for testing. In the SVM model training, the grid search method was used to optimize the hyperparameters, and the parameter combination of C = 1.0, kernel = 'linear' was finally selected. For the Softmax regression model, the Adam optimizer was used, the learning rate was set to 0.001, and the number of training rounds was 100.

[0252] In order to compare the performance of our solution with the existing technology, the research team simultaneously implemented three common classification methods: a single SVM model, a decision tree model, and a random forest model. All models were trained and evaluated using the same training and test sets.

[0253] After the model training was completed, the research team used the remaining 100 retired batteries as a validation set to conduct practical application tests on each model. The testing process was strictly carried out in accordance with the process described in our proposal: first, an appearance inspection was carried out, then an electrical performance test was carried out, and the data was input into a layer of SVM model to determine whether the conditions for cascade utilization were met. Finally, the batteries that met the conditions were classified into specific scenarios using a two-layer Softmax regression model.

[0254] In order to comprehensively evaluate the performance of the model, the research team not only focused on the overall classification accuracy, but also calculated the classification precision, recall rate, and F1 score for each specific scenario. At the same time, the calculation time of each method was recorded to evaluate its efficiency in practical applications. In addition, the research team also invited 5 industry experts to manually review the classification results to evaluate the rationality and practicality of the model judgment.

[0255] The experimental results are shown in the following table:

[0256]

[0257] Through the analysis of the above experimental results, it can be clearly seen that our solution method is significantly superior to the existing technical methods in many key indicators, fully demonstrating its innovativeness and practical value.

[0258] First, in terms of overall classification accuracy, our method achieved 94.5%, 5.9 percentage points higher than the second-best random forest model and 9.3 percentage points higher than the commonly used single SVM model. This significant improvement means that in practical applications, this method can more accurately identify the applicable scenarios of retired batteries and significantly reduce the economic losses and safety risks caused by misclassification.

[0259] Secondly, our method also performed well in the classification performance of various specific application scenarios. Taking the energy storage scenario as an example, the precision and recall of this method reached 96.2% and 95.8% respectively, which is much higher than other methods. This means that this method can not only accurately identify batteries suitable for energy storage applications, but also find all potential energy storage batteries to the maximum extent, giving full play to the residual value of retired batteries.

[0260] It is particularly noteworthy that the classification performance of our method is very balanced in various scenarios, with an average F1 score of 0.943, which shows that the method has achieved a good balance between precision and recall. In contrast, other methods either have high precision but insufficient recall (such as a single SVM model) or high recall but low precision (such as a decision tree model). The balanced performance of our method makes it more reliable and stable in practical applications.

[0261] In terms of computational efficiency, although our method uses a two-layer model structure, its processing speed is still quite impressive. It only takes 2.8 seconds to process 100 batteries, which is 33.3% faster than the random forest model and only 0.7 seconds slower than the fastest decision tree model. Considering the significantly improved classification accuracy of this method, this extra computing time is completely acceptable.

[0262] Finally, from the perspective of expert evaluation, our method received a high score of 4.7 (out of 5), which is much higher than other methods. This result shows that this method not only performs well in data, but also has been recognized by industry experts in practical applications. Experts generally believe that the classification results of this method are more reasonable and more in line with actual engineering needs, which is of great significance for improving the economic benefits and safety of the cascade utilization of retired batteries.

[0263] In general, our solution method effectively combines the efficient binary classification capability of SVM and the flexible multi-classification characteristics of Softmax regression through an innovative two-layer model structure, while ensuring high accuracy, it also achieves fast processing and scene segmentation. This method not only improves the utilization efficiency of retired batteries, but also provides a reliable decision-making tool for battery recycling companies, which helps promote the sustainable development of the new energy vehicle industry. In the future, with the increase in data volume and further optimization of the model, this method is expected to play an important role in a wider range of application scenarios.

[0264] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. A classification method for re-matching applicable scenarios for retired batteries, characterized by: include, By collecting and arranging the data of previous scholars on the cascade utilization of retired batteries, a new dataset that meets the requirements of cascade utilization is constructed; After preprocessing the data set, a layer of SVM model training is performed; Export the trained SVM model to a file and apply it to the classification system; Determine the applicable scenarios for retired batteries, and establish a data set for second-layer model training by collecting national standards on battery usage requirements for each scenario; After data preprocessing, the dataset is trained with a two-layer Softmax regression model; Export the trained Softmax regression model to a file and use it in the classification system.

2. A classification method for re-matching applicable scenarios for retired batteries as claimed in claim 1, characterized in that: The collected data feature values ​​include basic information, usage history, performance parameters, and safety of retired batteries. The corresponding data categories are divided into those that meet the requirements for cascade utilization and those that do not meet the requirements for cascade utilization. The data set is then cleaned to process missing values ​​and outliers. The data set is preprocessed and then trained with a layer of SVM model. The specific implementation of the layer of SVM model adopts the following steps: Data preparation: Collect basic information including but not limited to retired batteries, usage history, performance parameters, and safety, and then clean the data set to handle missing values ​​and outliers; Data cleaning: Clean the data to remove invalid or abnormal data points; Feature selection: Select the most relevant features from the integrated dataset for input to the SVM model; The Z-score normalization method is used to scale the features to the same scale. For each feature, the normalized feature can be expressed as: Among them, X j ′ represents the standardized feature; X j Indicates characteristics; μ j Represents feature X j The average value, σ j Represents feature X j The standard deviation of Then, the correlation analysis between the features is performed and the correlation coefficient matrix between the features is calculated to evaluate the relationship between the many features. The correlation coefficient indicates the degree of correlation between features i and j: Among them, r ij represents the correlation coefficient, that is, the degree of correlation between features i and j; x ik 、x jk Respectively represent the values ​​of feature i and feature j in the kth sample; represents the sample mean of feature i and feature j; n represents the number of samples, that is, the total number of observations in the data set; k represents the sample index, a cyclic variable from 1 to n; According to the correlation coefficient matrix, features with high correlation are selected for retention, and a threshold is set to 0.5, that is, only features with correlation coefficients greater than 0.5 are retained. According to the results of correlation analysis, the features are ranked by importance, and then relatively important features are selected to form a dataset subset for input to the SVM binary classification model. The remaining capacity, maximum voltage, and internal resistance are selected as the main feature vectors, and the feature vector of each battery can be expressed as: X=(x1,x2,x3)=(remaining capacity, maximum voltage, internal resistance) Build the SVM model: For binary classification, that is, linear separability, the objective function of the SVM model can be expressed as: Among them, W is the weight vector, b is the bias term, C is the penalty parameter, and T represents the transpose of the W weight vector; in order to allow some feature points to violate the classification boundary, ε is added i is the slack variable; The constraints of the above objective function can be expressed as: y i (W T X i +b)≥1-ε i ,i=1,2,...,n Since the data is linearly separable, there is no need to use a complex kernel function, only a simple linear kernel function is needed, so the decision function of the binary classification model is: f(x)=sign(W T X+b) Classification Decision: After model training, for any new battery feature vector xj, the SVM binary classification decision model can be used to determine whether the battery meets the basic conditions for cascade utilization; If f(x)′>0, the battery meets the requirements of cascade utilization, and can enter the subsequent scene judgment classification; if f(x)′<0, the battery does not meet the requirements of cascade utilization, and is discarded, and the metal is recycled for reuse; In a single-layer SVM model, retired batteries whose output results do not meet the conditions for cascade utilization will be directly eliminated and will no longer enter the next stage.

3. A classification method for re-matching applicable scenarios for retired batteries as claimed in claim 2, characterized in that: The training process is to train the two-layer Softmax regression model after data preprocessing; train the two-layer Softmax regression model and continuously optimize the model accuracy through parameter tuning; The features are standardized using the Z-score standardization method to ensure that all features have the same scale. The standardized features can be expressed as: Among them, X ij ′ represents the jth feature of each i-th sample, μ j represents the average value of the jth feature, σ j represents the standard deviation of the jth feature; SoftMax regression model: Initialize the weight matrix: W∈R d×K Among them, d is the number of features and K is the number of categories; b represents the bias vector: b∈R K Forward propagation: After calculating the linear combination of the input features and model weights for each category, plus the bias vector, the expression can be: z i =W i T x+b i Among them, W is the model weight, b represents the bias vector, i represents the category, and x represents the feature; SoftMax function: Use the SoftMax function to transform the result of the linear combination into a normalized overview distribution: Where p(y=i|x) represents the probability that a sample belongs to category i given an input x, and K is the total number of categories; Loss function: Cross entropy loss function: For each sample, the loss function can be expressed as: L i =-log(p(y=y i |x i )) Total loss function: Based on the entire training set, the total loss function is the average of all cross entropy functions, that is, the average of all sample losses: Backward Propagation: Calculate gradients: Calculate the gradient of the loss function with respect to each weight wij and bias: Among them: W ij represents each individual weight; b i represents the bias gradient; δ(y k ,j) is the indicator function, and when y k =i is 1, otherwise it is 0; Parameter update: Gradient Descent: Update weights and biases using gradient descent: Where η is the learning rate.

4. A classification method for re-matching applicable scenarios for retired batteries as claimed in claim 3, characterized in that: The trained two-layer Softmax regression model can perform scenario classification on retired batteries that meet the requirements of second-life utilization, that is, re-match the applicable scenarios of retired batteries, and then re-enter the market to determine the data type of the data feature set in the data set, so as to facilitate the subsequent output of classification results.

5. A classification method for re-matching applicable scenarios for retired batteries as claimed in claim 4, characterized in that: The retired batteries to be tested are sent into the system, and the appearance of the retired batteries to be classified is inspected by a high-definition camera; The retired batteries that have passed the appearance inspection are then sent to the battery parameter inspection stage, which is mainly used to detect various parameters of the retired batteries, including battery performance and reliability; After passing the battery detection system, all detection parameters of the retired battery are collected and input into a first-layer binary classification model to make a cascade utilization judgment; Then the detection parameters of retired batteries that meet the requirements of cascade utilization are input into the two-layer multi-classification model; According to the classification results, the retired batteries are sorted into corresponding applicable scenarios.

6. A classification method for re-matching applicable scenarios for retired batteries as claimed in claim 5, characterized in that: The inspection appearance has no scratches, deformation, damage or leakage, the positive and negative electrodes have no rust, and the markings are correct and clear; if the conditions are met, the next step will be entered; if not, the product will be discarded and will not be reused; Through the battery detection system, the system is mainly used to detect various parameters of retired batteries, including battery performance and reliability; Collect all the detection parameters of the retired battery, input the parameters into the first-layer binary classification model, and output the retired batteries that do not meet the conditions for cascade utilization, which will be directly eliminated and will not enter the next stage; The battery performance test parameter range includes charge and discharge voltage and capacity at different temperatures, charge and discharge energy at different rates, DC internal resistance, storage performance, standard cycle life, SOC calibration, peak power, and energy efficiency; The reliability testing parameters range includes electrical insulation, flame retardancy, thermal management, high temperature and high humidity.

7. A classification method for re-matching applicable scenarios for retired batteries as claimed in claim 6, characterized in that: The detection parameters of the retired batteries that meet the requirements of cascade utilization are input into the two-layer multi-classification model, and the classification results of the retired batteries can be output, thereby obtaining the applicable scenarios for the cascade utilization of the batteries, and then entering the market for the second time; After passing through the battery detection system, the detection parameters of the retired battery are collected and numbered, and the series of parameters are passed through a one-layer binary classification model to screen the battery for cascade utilization; The detection parameters of the screened retired batteries are input into a two-layer multi-classification model to obtain the re-matching results of the retired batteries with the same number, so that they can re-enter the market.

8. A classification system for re-matching applicable scenarios for retired batteries, based on a classification method for re-matching applicable scenarios for retired batteries according to any one of claims 1 to 7, characterized in that: Also includes, The construction module collects and organizes the data of previous scholars on the cascade utilization of retired batteries to build a new data set that meets the requirements of cascade utilization; The training module performs a layer of SVM model training on the data set after preprocessing; The output module exports the trained SVM model to a file and applies it to the classification system; The matching module determines the applicable scenarios for retired batteries and establishes a data set for second-layer model training by collecting national standards on battery usage requirements for each scenario; The processing module trains the two-layer Softmax regression model after data preprocessing. The classification module exports the trained Softmax regression model to a file and applies it to the classification system.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of a classification method for re-matching applicable scenarios for retired batteries as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a classification method for re-matching applicable scenarios for retired batteries as described in any one of claims 1 to 7 are implemented.

Images