City-level prediction method and system for the quantity of retired power batteries based on multi-dimensional feature ML

Abstract

The present invention discloses a method and system for predicting the quantity of retired power batteries at the city level based on multi-dimensional features and ML. The method includes: obtaining the insurance volume of new energy vehicles and N kinds of urban characteristic values within a preset time period to construct a data set; analyzing the N kinds of urban characteristic values to determine the key urban characteristic values affecting the insurance volume of new energy vehicles and their weight values; screening the data set according to the key urban characteristic values and their weight values, and training a variety of optimized machine learning models using the screened data set; evaluating the prediction performance of each machine learning model, and selecting the best model for predicting the insurance volume of passenger cars and commercial vehicles respectively; inputting the future key urban characteristic values into the best model to predict the insurance volume of new energy vehicles; based on the predicted insurance volume of new energy vehicles, using the battery type ratio and battery characteristics of new energy vehicles, and combining with the Weibull life distribution, predicting the quantity of retired power batteries at the city level. The present invention can achieve the prediction accuracy and practicability of the quantity of retired power batteries at the city level.

Description

Technical Field

[0001] The present invention relates to the technical fields of new energy vehicle market and prediction technology of retired power batteries, and particularly to a method and system for predicting the quantity of urban retired power batteries based on multi-dimensional features ML. Background Art

[0002] With the dual pressures of global climate change and environmental protection, new energy vehicles, as a key tool for reducing greenhouse gas emissions and promoting energy transformation, have been emphasized globally. Power batteries are the core of new energy vehicles. Although when the energy of power batteries decays to less than 70%-80%, it can basically not meet the travel needs, but they still have high recycling value. Since key metals in power batteries can be recycled, not only can the material supply be supplemented, but also the environmental impact caused by resource extraction can be effectively reduced. Therefore, the recycling of power batteries has received extensive attention. Due to significant differences in industrial structure, resource endowment, economic development level, traffic demand, and policy effects among cities, predicting the quantity of retired power batteries by region can provide more refined data support.

[0003] In the process of implementing the method for predicting the insurance volume of new energy vehicles in the present invention, the inventors of this patent have at least found the following technical problems in the prior art:

[0004] (1) Single data dimension: Most of the prior art predicts the quantity of retired power batteries based on historical one-dimensional data. For example, the prior art does not make a detailed distinction between the types of new energy vehicles, especially does not distinguish between passenger vehicles and commercial vehicles, and there are great differences in the usage purposes, user groups, market demand laws, etc. between passenger vehicles and commercial vehicles; the prior art does not distinguish between battery types, and the installed capacity ratio of each battery will be continuously updated with changes in the market, technology, climate, and consumer demand. The single-dimensional data in the prior art cannot comprehensively reflect the actual retired weight and capacitance of retired power batteries, resulting in inaccurate prediction results.

[0005] (2) Failure to consider urban characteristic differences: The prior art makes predictions at the provincial or national level and does not fully consider the significant differences in economic development level, population structure, policy environment, etc. among cities. Cities are the main consumption areas of new energy vehicles, and the unique geographical location of each city determines the differences in its resource endowment and industrial layout, which in turn affect the sources and types of retired power batteries.

[0006] (3) Single model and poor applicability: The linear regression model adopted in the prior art is based on the core assumptions of data stationarity and linear relationship, and relies on historical laws to construct a prediction framework. In the scale conversion from macro to micro, the model exposes significant adaptability problems.

[0007] (4) Lack of dynamic adaptability of the model: When facing the rapid changes in the market environment, the existing prediction models are unable to adjust the model parameters and structure in a timely manner, making it difficult to accurately predict the impact of these changes on the insurance volume of new energy vehicles, and lacking dynamic adaptability.

[0008] In summary, the existing methods for predicting the quantity of retired power batteries cannot meet the actual needs. Summary of the Invention

[0009] The embodiments of the present invention provide a method and system for predicting the quantity of urban-level retired power batteries based on multi-dimensional features ML, which solves the problem that the existing methods for predicting the quantity of retired power batteries cannot meet the actual needs.

[0010] On the one hand, the embodiments of the present invention provide a method for predicting the quantity of urban-level retired power batteries based on multi-dimensional features ML, including:

[0011] Obtain the insurance volume of new energy vehicles and N kinds of urban characteristic values within a preset time period to construct a data set;

[0012] Analyze the N kinds of urban characteristic values to determine the key urban characteristic values affecting the insurance volume of new energy vehicles and their weight values;

[0013] Screen the data set according to the key urban characteristic values and their weight values, and use the screened data set to train multiple optimized machine learning models;

[0014] Evaluate the prediction performance of each machine learning model, and select the best model for predicting the insurance volume of passenger cars and commercial vehicles respectively;

[0015] Input the future key urban characteristic values into the best model to predict the insurance volume of new energy vehicles;

[0016] Based on the predicted insurance volume of new energy vehicles, use the battery type ratio and battery characteristics of new energy vehicles, combined with the Weibull life distribution, to predict the quantity of urban-level retired power batteries.

[0017] Optionally, the N kinds of urban characteristic values include the urban population, urban GDP, urbanization rate, Baidu index, and urban new energy vehicle promotion subsidy amount.

[0018] Optionally, the method for analyzing the N kinds of urban characteristic values to determine the key urban characteristic values affecting the insurance volume of new energy vehicles is specifically the clustering analysis method, and the method for determining the weight values of the key urban characteristics is specifically the principal component analysis method.

[0019] Optionally, the machine learning models include LSTM model, random forest model, XGBoost model, ANN model, and stacked regression model, and the optimization method is specifically the Bayesian optimization method or the hyperparameter tuning method.

[0020] Optionally, the evaluation metrics used to evaluate the prediction performance of each machine learning model include mean squared error and coefficient of determination.

[0021] Optionally, the key future urban feature values include future urban GDP, future urban population, future urbanization rate, future Baidu Index, and future urban new energy vehicle promotion subsidy amount.

[0022] Optionally, the battery types specifically include 24 types composed of LFP batteries, NCM111 batteries, NCM523 batteries, NCM622 batteries, NCM811 batteries, and NCA batteries under the supporting of pure electric passenger vehicles, hybrid electric passenger vehicles, pure electric commercial vehicles, and hybrid electric commercial vehicles; the battery characteristics specifically include battery life, battery weight, and battery capacity.

[0023] On the other hand, an embodiment of the present invention further provides an urban-level retired power battery quantity prediction system based on multi-dimensional feature ML, including:

[0024] An acquisition module, configured to acquire the new energy vehicle insurance volume and N urban feature values within a preset time period to construct a data set;

[0025] A determination module, configured to analyze the N urban feature values to determine the key urban feature values and their weight values that affect the new energy vehicle insurance volume;

[0026] A training module, configured to screen the data set according to the key urban feature values and their weight values, and use the screened data set to train multiple optimized machine learning models;

[0027] A selection module, configured to evaluate the prediction performance of each machine learning model, and select the best model for predicting the insurance volume of passenger vehicles and commercial vehicles respectively;

[0028] A processing module, configured to input the key future urban feature values into the best model to predict the new energy vehicle insurance volume, and then based on the predicted new energy vehicle insurance volume, use the battery type ratio and battery characteristics of new energy vehicles, combined with the Weibull life distribution, to predict the urban retired power battery quantity.

[0029] In a third aspect, an embodiment of the present invention further provides a computer device, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the steps of a method for predicting the quantity of urban retired power batteries based on multi-dimensional feature ML in the foregoing embodiments are implemented.

[0030] Fourthly, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of a method for predicting the quantity of retired power batteries at the city level based on multi-dimensional feature ML in the foregoing embodiments are implemented.

[0031] One or more technical solutions provided in the embodiments of the present invention have at least the following technical effects or advantages:

[0032] Firstly, in terms of data processing and analysis, by comprehensively considering N kinds of urban characteristic values, the limitation of traditional new energy vehicle insurance volume prediction relying only on a single or a small number of data dimensions is broken through. By determining the key urban characteristic values, the factors that play a key role in the new energy vehicle insurance volume in different cities can be accurately identified, effectively reducing data redundancy and interference from irrelevant information. At the same time, by determining the weight values of each key feature, the influence of important factors in the prediction model is further highlighted, making the model input data more targeted and representative, and laying a solid foundation for subsequent accurate prediction.

[0033] Secondly, in the model construction and training link, multiple machine learning models are adopted and optimized. After optimization, the coefficient of determination of the best model is increased by more than 0.85, and the mean square error is reduced close to 0.

[0034] Furthermore, in terms of the accuracy and practicality of prediction, model evaluation and selection of the best model are carried out for passenger vehicles and commercial vehicles respectively, fully considering the differences in urban characteristics between these two types of vehicles. The present invention has undergone multiple groups of strict cross-validation and time series analysis experiments. Stable and reliable prediction results have been obtained in different future urban key characteristic value data. This enables the prediction results to be more in line with the insurance situation of different vehicle types in the actual market, and then accurately obtain the annual demand of consumers for new energy vehicles.

[0035] Finally, based on the accurate demand for new energy passenger vehicles and commercial vehicles, the accurate prediction of the quantity of retired power batteries at the city level can be realized, effectively improving the accuracy and practicality of the prediction of the quantity of retired power batteries at the "city-provincial-regional" level, and promoting the rational allocation of resources and the healthy development of the power battery recycling industry. Description of the Drawings

[0036] Figure 1 It is a flowchart of a method for predicting the quantity of retired power batteries at the city level based on multi-dimensional feature ML in an embodiment of the present invention;

[0037] Figure 2 It is a schematic diagram of the clustering analysis result in an embodiment of the present invention;

[0038] Figure 3 It is a schematic diagram of the principal component analysis result in an embodiment of the present invention;

[0039] Figure 4 This is a performance comparison chart of the machine learning model in an embodiment of the present invention;

[0040] Figure 5 This is a chart showing the proportion of installed capacity of 24 battery types in a certain period in an embodiment of the present invention;

[0041] Figure 6 This is a distribution chart of Weibull life model parameters of 24 battery types in an embodiment of the present invention. Detailed implementation manners

[0042] The embodiment of the present invention provides a method and system for predicting the quantity of urban-level retired power batteries based on multi-dimensional feature ML, which solves the problem that the existing methods for predicting the quantity of retired power batteries cannot meet the actual needs.

[0043] The technical solution of the embodiment of the present invention is to solve the above problems, and the general idea is as follows:

[0044] First, in terms of data collection, the present invention is committed to obtaining comprehensive and multi-dimensional data. The data on the insurance volume of new energy vehicles within a preset time period is collected from multiple reliable data sources, and these data can directly reflect the usage of new energy vehicles in the actual market. At the same time, the present invention widely collects N kinds of urban characteristic values, which cover the key factors affecting the insurance volume of new energy vehicles.

[0045] Secondly, in-depth analysis is carried out on the collected data. Data mining and analysis techniques are used to process the urban characteristic values. For example, clustering analysis and other methods are adopted. By reasonably clustering cities, the differences in each characteristic of different urban clusters are mined, and then the key urban characteristic values that have a significant impact on the insurance volume of new energy vehicles are determined. This step can screen out the data most relevant to the prediction target from numerous complex data, reduce the interference of data noise and irrelevant information, and improve the data quality.

[0046] Then, the data set is further screened according to the selected key urban characteristic values, and the data valuable for predicting the insurance volume of new energy vehicles is retained, and redundant data is removed, so as to obtain a refined and highly relevant data set.

[0047] Next, the filtered dataset is used to train the machine learning model. The present invention is not limited to a single machine learning model, but comprehensively considers various different types of machine learning models, such as long short-term memory network models, random forest models, XGboost models, ANN neural network models, and stacked regression models. By inputting the filtered dataset into these different machine learning models for training respectively, the machine learning models learn the internal relationship between the key urban feature values and the number of newly registered new energy vehicles.

[0048] After the model training is completed, a comprehensive and detailed evaluation is carried out on each machine learning model. Using scientific and reasonable evaluation metrics such as mean squared error and coefficient of determination, a quantitative analysis of the prediction performance of each machine learning model is carried out, so as to accurately compare the advantages and disadvantages of each machine learning model.

[0049] Again, the best-performing model is selected from the multiple machine learning models that have been evaluated, and reasonable assumptions are made about possible future scenarios, and the future key urban feature values are set. These set future key urban feature values are input into the best model, so as to accurately predict the future number of newly registered new energy vehicles.

[0050] Finally, after obtaining the prediction results of the number of newly registered new energy passenger vehicles and commercial vehicles in each city, based on the predicted number of newly registered new energy vehicles, using the battery type ratio and battery characteristics of new energy vehicles, combined with the Weibull lifetime distribution, the prediction of the amount of retired power batteries in each city is realized.

[0051] In summary, through this systematic and comprehensive technical idea, the present invention can effectively overcome the problems existing in the prediction of the amount of retired power batteries in the prior art, such as single data dimension, single model and poor applicability, and failure to consider urban feature differences, etc., and provide reliable demand prediction data support for the development of the power battery recycling industry.

[0052] As Figure 1 shown, a method for predicting the amount of retired power batteries at the city level based on multi-dimensional feature ML provided by the present invention includes steps S1 to S6.

[0053] Step S1: Obtain the number of newly registered new energy vehicles and N kinds of urban feature values within a preset time period to construct a dataset. Among them, the N kinds of urban feature values include the urban population, urban GDP, urbanization rate, Baidu index, and urban new energy vehicle promotion subsidy amount.

[0054] In the specific implementation process, for example: obtain the insurance volume data of new energy vehicles within a preset time period from the database of relevant insurance institutions, and at the same time collect various urban characteristic values, such as obtaining the urban population, urbanization rate, and urban GDP from the urban statistical department, obtaining the attention of urban netizens to new energy vehicles from Baidu Index, and obtaining environmental protection policy documents from relevant government departments and quantifying them as the promotion subsidy amount for new energy vehicles in the city. Finally, the above data constitutes a data set.

[0055] Step S2: Analyze the N kinds of urban characteristic values to determine the key urban characteristic values affecting the insurance volume of new energy vehicles and their weight values. Among them, the method for analyzing the N kinds of urban characteristic values to determine the key urban characteristic values affecting the insurance volume of new energy vehicles is specifically the clustering analysis method, and the method for determining the weight values of the key urban characteristics is specifically the principal component analysis method.

[0056] In the specific implementation process, for example: perform clustering analysis on the collected urban characteristic values, use the K-means algorithm, and divide the cities into several groups according to the feature similarity. The objective function of the K-means algorithm is expressed as:

[0057] ,

[0058] In the formula, represents the objective function of clustering. The smaller its value, the better the clustering effect, that is, the smaller the sum of the distances from the sample points within the cluster to its cluster center;

[0059] : represents the number of clusters of clustering, that is, the number of groups into which the data is to be divided;

[0060] : is an index variable used to traverse clusters, from 1 to ;

[0061] : represents the set of sample points included in the th cluster;

[0062] : represents a sample point in the data set;

[0063] : represents the th cluster center, which is a vector with the same dimension as the sample point;

[0064] represents the sample point to the th cluster center Euclidean distance, and the distance is squared in the objective function.

[0065] As Figure 2 shown, the economic GDP and urban population in urban characteristics are divided into 7 groups, including Cluster 1 to Cluster 7.

[0066] The clustering effect is evaluated by the Calinski-Harabasz index, and the optimal number of clusters is selected. Principal component analysis is performed on each clustered city group to extract the main components and reduce the feature dimensions. The mathematical principle of principal component analysis: Let the original data matrix be X, X is the data matrix, each column is a sample, and each row is a feature. Calculate the covariance matrix: , where T represents the transpose of the matrix. The covariance matrix is a symmetric matrix, which reflects the linear relationship and variance among the variables in the original data. Calculate the eigenvalues and eigenvectors of the covariance matrix: , is the eigenvalue of is the eigenvector of the matrix corresponding to the eigenvalue. The principal component score , where is the matrix composed of eigenvectors , arranged by column, that is , is a matrix, and each column represents the score of the th principal component, that is, the projection value of the original data on the th principal component.

[0067] As Figure 3 shown, under different clusters, the correlation gaps between the insured quantities of new energy passenger vehicles and commercial vehicles and the urban characteristics are significant. For example, for the urban agglomerations in Cluster 6, the weight of the urbanization rate on the insured quantity of commercial vehicles is higher than that of passenger vehicles.

[0068] Step S3: Screen the data set according to the urban key feature values and their weight values, and use the screened data set to train multiple optimized machine learning models. Among them, the machine learning models include the LSTM model, random forest model, XGBoost model, ANN model, and stacked regression model, and the optimization method is specifically the Bayesian optimization method or hyperparameter tuning method.

[0069] In the specific implementation process, for example, in order to improve the generalization ability of the model, Bayesian optimization method was used to finely tune the hyperparameters during the training process, and a variety of advanced ML (Machine Learning) algorithms were introduced, including: LSTM (Long Short-Term Memory), random forest, XGBoost (eXtreme Gradient Boosting), ANN (Artificial Neural Network) and stacked regression, in order to better capture the changing trend of the new energy vehicle insurance volume. In addition, due to significant differences between different cities, before applying to a specific city, it is necessary to analyze the special conditions of the city and adjust the machine learning model structure or parameters accordingly to ensure that the prediction results are both widely applicable and in line with the local reality. To achieve the final training effect, the following steps S31 to S33 are also included.

[0070] Step S31: Construct an LSTM model, a random forest model, an XGBoost model, an ANN model and a stacked regression model.

[0071] LSTM model: Use the Keras library to build an LSTM network. The output calculation formula of the LSTM unit is:

[0072] ,

[0073] ,

[0074] ,

[0075] ,

[0076] ,

[0077] In the formula, is the input gate, which is a vector of values between 0 and 1; is the sigmoid activation function, which maps the input between 0 and 1; is the weight matrix of the input to the input gate; is the input data at time step is the weight matrix from the hidden state of the previous time step to the input gate; is the hidden state of the previous time step; is the bias vector of the input gate, which is used to adjust the output of the input gate; is the forget gate; represents the weight matrix of the input to the forget gate; is the hidden state at the previous time step The weight matrix to the forget gate; is the bias vector of the forget gate; is the time step The updated cell state, which is the core of the LSTM's long-term memory; Denotes element-wise multiplication; is the cell state at the previous time step; Denotes the hyperbolic tangent activation function, which maps the input between -1 and 1; is the input to The weight matrix of the cell state update part; is the weight matrix from the hidden state at the previous time step to the cell state update part; is the bias vector of the cell state update part; is the output gate at time step ; is the input The weight matrix to the output gate; is the hidden state at the previous time step The weight matrix to the output gate; is the bias vector of the output gate; The hidden state at time step , which is one of the outputs of the LSTM and will be passed to the next time step or used for subsequent tasks.

[0078] Random Forest Model: Configure a random forest classifier using the Scikit-learn library. A random forest consists of multiple decision trees, and the construction process of each tree is as follows: Randomly select a subset of samples; Randomly select a subset of features at each node; Perform the best split based on the selected subset of features; Repeat the above process until the stopping condition is met.

[0079] XGBoost Model: A distributed gradient boosting library that is an optimized implementation based on the gradient boosting framework. It constructs the final model by gradually optimizing the objective function through multiple rounds of iteration. The objective function includes the loss function and the regularization term :

[0080] ,

[0081] In the formula, represents the value of the objective function at the th round of iteration; represents the number of samples; is the sample index; is the th true label value of the sample; is the loss function, is the The predicted value of the round is the predicted value of the newly added tree.

[0082] ANN model: Define a simple fully connected neural network structure through TensorFlow or PyTorch. Assume the network has two hidden layers, and the output of each layer can be expressed as:

[0083] ,

[0084] ,

[0085] where is the weighted input of the th layer; and are the weight matrix and the bias vector respectively, which will be continuously updated according to the optimization algorithm used to minimize the loss function; is the activation function; the th layer's activation output.

[0086] Stacked regression model: Use the stacked regression model as the base estimator, and then integrate their outputs through another meta-model (such as linear regression). The prediction process of stacked regression can be formalized as:

[0087] ,

[0088] where represents the predicted value of the stacked regression model; in is the base estimator (which can be multiple base models, such as multiple different regression models), is the input data (a dataset containing multiple features). The base estimator processes and predicts the input data to obtain a set of intermediate results. These intermediate results can be regarded as a kind of feature transformation or preliminary prediction of the original data; is the meta-model, usually a simple model (such as linear regression), which takes the output of the base estimator as the input, further integrates this information, and finally outputs the predicted value .

[0089] Step S32: Divide the dataset into a training set and a test set, and use the training set to train the machine learning model.

[0090] Step S33: Optimize the machine learning model, including Bayesian optimization and hyperparameter tuning.

[0091] Bayesian Optimization: Use the Bayesian optimization method to tune the hyperparameters of the model, such as the learning rate, hidden layer size, batch size, etc. The core idea of Bayesian optimization is to approximate the objective function by constructing a probability model and use this probability model for sampling to gradually approach the optimal solution.

[0092] Hyperparameter Tuning: Further optimize the hyperparameters of the model through methods such as grid search or random search.

[0093] Step S4: Evaluate the prediction performance of each machine learning model and select the best model for predicting the insurance volume of passenger cars and commercial vehicles respectively. Among them, the evaluation metrics used to evaluate the prediction performance of each machine learning model include the mean squared error and the coefficient of determination.

[0094] In the specific implementation process, for example: through the coefficient of determination and the mean squared error MSE metrics to evaluate the prediction performance of each machine learning model, and comprehensively select the best models suitable for the insurance volume of new energy passenger cars and new energy commercial vehicles respectively.

[0095] Coefficient of determination The expression is as follows:

[0096] ,

[0097] The expression of the mean squared error MSE is as follows:

[0098] ,

[0099] In the formula, represents the number of samples, The actual observed value of the th sample, The predicted value of the th sample obtained by model prediction, is the average value of the actual observed values of all samples, is the sum of squared residuals, is the total sum of squares.

[0100] Such as Figure 4 shown, through the coefficient of determination and the mean squared error MSE results, it can be concluded that the best model for predicting the insurance volume of new energy passenger cars is the LSTM model, and the best model for predicting the insurance volume of new energy commercial vehicles is the stacked regression model.

[0101] Step S5: Input the future key feature values of the city into the best model to predict the insurance volume of new energy vehicles. Among them, the future key feature values of the city include the future city GDP, the future city population, the future urbanization rate, the future Baidu index, and the future promotion subsidy amount of new energy vehicles in the city.

[0102] In the specific implementation process, for example, it is necessary to accurately obtain and organize various key characteristic values of future cities. For the future urban GDP, it is determined by synthesizing the predictions of economic experts, regional economic development plans, and macroeconomic trend analyses. The future urban population is referenced based on census data, population migration trend studies, and population growth targets in urban planning. The future urbanization rate is estimated by combining information such as the urbanization promotion plan and land use plan of the city. The Baidu Index reflects the changing trend of the public's attention to new energy vehicles, and its future trend is predicted by continuously monitoring search engine data and analyzing social media popularity. The future promotion subsidy amount for new energy vehicles in the city is estimated according to the trend of the urban environmental protection subsidy policy.

[0103] After accurately inputting these future key characteristic parameters covering multiple aspects such as urban economy, population, policy, and attention into the best model that has been carefully trained and selected, the best model will perform efficient calculations and in-depth inferences based on its internal complex algorithm architecture and the internal logical relationship between the characteristic values of each city and the insurance volume of new energy vehicles learned during the previous training process. The best model will comprehensively consider the interaction and synergistic effects among various parameters, accurately predict and calculate the insurance volume of new energy vehicles in different cities under specific future scenarios, and output the predicted value of the insurance volume of new energy vehicles.

[0104] Step S6: Based on the predicted insurance volume of new energy vehicles, use the battery type ratio and battery characteristics of new energy vehicles, and combine with the Weibull life distribution to predict the amount of retired power batteries at the city level. Among them, the battery types specifically include 24 types composed of LFP batteries, NCM111 batteries, NCM523 batteries, NCM622 batteries, NCM811 batteries, and NCA batteries under passenger car pure electric type, passenger car hybrid electric type, commercial vehicle pure electric type, and commercial vehicle hybrid electric type; the battery characteristics specifically include battery life, battery weight, and battery capacity.

[0105] In the specific implementation process, for example, the 24 types of new energy vehicle batteries include: passenger car pure electric LFP batteries, passenger car pure electric NCM111 batteries, passenger car pure electric NCM523 batteries, passenger car pure electric NCM622 batteries, passenger car pure electric NCM811 batteries, passenger car pure electric NCA batteries, passenger car hybrid LFP batteries, passenger car hybrid NCM111 batteries, passenger car hybrid NCM523 batteries, passenger car hybrid NCM622 batteries, passenger car hybrid NCM811 batteries, passenger car hybrid NCA batteries, commercial vehicle pure electric LFP batteries, commercial vehicle pure electric NCM111 batteries, commercial vehicle pure electric NCM523 batteries, commercial vehicle pure electric NCM622 batteries, commercial vehicle pure electric NCM811 batteries, commercial vehicle pure electric NCA batteries, commercial vehicle hybrid LFP batteries, commercial vehicle hybrid NCM111 batteries, commercial vehicle hybrid NCM523 batteries, commercial vehicle hybrid NCM622 batteries, commercial vehicle hybrid NCM811 batteries, commercial vehicle hybrid NCA batteries.

[0106] The installed capacity proportion of the 24 types of new energy vehicle batteries in a certain period is as Figure 5 shown. The double-parameter Weibull distribution model closest to the actual use situation of power batteries is used for estimation. For vehicle models , the Weibull distribution function can be expressed as , where represents time, and the probability density function formula of the Weibull distribution is:

[0107] ,

[0108] where: is the shape parameter of vehicle model , which determines the shape of the distribution, affects the curvature of the function and the retirement mode. Different values can represent different failure characteristics. For example, when , it usually represents the early retirement mode, that is, the vehicle battery has a relatively high retirement probability in the initial stage of use; when , the Weibull distribution degenerates into an exponential distribution, which can be used to describe random retirement; when , it usually represents the wear retirement mode, that is, the retirement of power batteries increases with time. is the scale parameter of vehicle model , which is a proportional parameter related to the scale of the distribution and determines the range of the distribution. For example, it affects the peak position and width of the distribution. As Figure 6 shown, for each type of battery, different and values need to be estimated to reflect the life differences of various types of batteries.

[0109] The cumulative distribution function of the Weibull distribution is expressed in terms of the calculation time. The probability of power battery retirement within 1 year, cumulative distribution function The formula is: .

[0110] Based on the insurance data of new energy vehicles, The year when new energy vehicles are insured. To calculate the year, For the city In the insurance year The number of new energy vehicles insured is Year of insurance mid-size car The proportion of For car models In the risk The probability of Weibull retiring after 2018 is: For car models The weight of the power battery, For car models Power battery capacity. City In the calculation year mid-size car Weight of retired batteries The calculation formula is:

[0111] ,

[0112] City In the calculation year mid-size car Retired battery capacity The calculation formula is:

[0113] ,

[0114] Weibull retirement probability The calculation formula is:

[0115] ,

[0116] By calculating the weight and capacity of each type of retired power batteries in each city each year, we can then calculate the number of retired power batteries from passenger cars or commercial vehicles, LFP or NMC types, and the total number of retired power batteries in the city, as needed. This refined data can provide data science support for companies' deployment strategies at the "city-province-region" level.

[0117] It should be noted that the present invention can predict the amount of retired power batteries in 364 inland cities, including 293 prefecture-level cities, 4 municipalities directly under the Central Government, 7 regions, 30 autonomous prefectures, 3 leagues, and 27 directly administered counties.

[0118] In a second aspect, an embodiment of the present invention further provides a city-level retired power battery amount prediction system based on multi-dimensional feature ML, including:

[0119] An acquisition module for acquiring the insurance volume of new energy vehicles and N kinds of urban characteristic values within a preset time period to construct a data set;

[0120] A determination module for analyzing N kinds of urban characteristic values to determine the key urban characteristic values affecting the insurance volume of new energy vehicles and their weight values;

[0121] A training module for screening the data set according to the key urban characteristic values and their weight values, and using the screened data set to train multiple optimized machine learning models;

[0122] A selection module for evaluating the prediction performance of each machine learning model and selecting the best models for predicting the insurance volume of passenger vehicles and commercial vehicles respectively;

[0123] A processing module for inputting future key urban characteristic values into the best model to predict the insurance volume of new energy vehicles, and then based on the predicted insurance volume of new energy vehicles, using the battery type ratio and battery characteristics of new energy vehicles, combined with the Weibull life distribution, to predict the amount of retired power batteries in the city.

[0124] From the perspectives of the system and application, the city-level retired power battery amount prediction system based on multi-dimensional feature ML provided by the present invention realizes the efficient integration and convenient application of the technical solution. Each module in the system has a clear division of labor and works in coordination, forming a complete closed-loop process from data acquisition, feature analysis, model training to final prediction processing.

[0125] In a third aspect, an embodiment of the present invention provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of a method for predicting the amount of retired power batteries in a city based on multi-dimensional feature ML in the foregoing embodiment.

[0126] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of a method for predicting the amount of retired power batteries in a city based on multi-dimensional feature ML in the foregoing embodiment.

[0127] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A method for predicting the quantity of retired power batteries at the city level based on multi-dimensional feature ML, characterized in that, Including: Obtain the insurance volume of new energy vehicles and N kinds of urban characteristic values within a preset time period to construct a data set, where the N kinds of urban characteristic values cover the key factors affecting the insurance volume of new energy vehicles; Conduct data mining and analysis on the N kinds of urban characteristic values to determine the key urban characteristic values affecting the insurance volume of new energy vehicles and their weight values, which can screen out the data most relevant to the prediction target, reduce the interference of data noise and irrelevant information, and improve data quality; Screen the data set according to the key urban characteristic values and their weight values to retain the data valuable for predicting the insurance volume of new energy vehicles, remove redundant data, and use the screened data set to train multiple optimized machine learning models, enabling the machine learning models to learn the internal relationship between the key urban characteristic values and the insurance volume of new energy vehicles; Evaluate the prediction performance of each machine learning model, conduct quantitative analysis on the prediction performance of each machine learning model, and select the best models for predicting the insurance volume of passenger vehicles and commercial vehicles respectively from multiple machine learning models that have been evaluated; Input the future key urban characteristic values into the best model, and the best model will perform operations and inferences based on its internal algorithm architecture and the internal logical relationship between each urban characteristic value and the insurance volume of new energy vehicles learned during the previous training process to predict the insurance volume of new energy vehicles in different cities under specific future scenarios; Based on the predicted insurance volume of new energy vehicles, obtain the annual demand of consumers for new energy vehicles, and use the installed capacity ratio and battery characteristics of the battery types of new energy vehicles, combined with the probability of power battery retirement calculated by the Weibull life distribution, to predict the amount of retired power batteries at the city level, including the amount of retired power batteries from passenger vehicles or commercial vehicles, the amount of retired power batteries of LFP or NMC types, and the total amount of retired power batteries in the city; The N kinds of urban characteristic values include the urban population, urban GDP, urbanization rate, Baidu Index, and the amount of new energy vehicle promotion subsidies in the city; among them, obtain the urban population, urbanization rate, and urban GDP from the urban statistical department, obtain the attention of urban netizens to new energy vehicles from the Baidu Index, and obtain environmental protection policy documents from relevant government departments and quantify them as the amount of new energy vehicle promotion subsidies in the city; The future key urban characteristic values include future urban GDP, future urban population, future urbanization rate, future Baidu Index, and future amount of new energy vehicle promotion subsidies in the city; The battery types specifically include 24 types composed of LFP batteries, NCM111 batteries, NCM523 batteries, NCM622 batteries, NCM811 batteries, and NCA batteries under the matching of passenger vehicle pure electric type, passenger vehicle hybrid electric type, commercial vehicle pure electric type, and commercial vehicle hybrid electric type; the battery characteristics specifically include battery life, battery weight, and battery capacity; The method for analyzing the N kinds of urban characteristic values to determine the key urban characteristic values affecting the insurance volume of new energy vehicles is specifically the clustering analysis method, and the method for determining the weight values of the key urban characteristics is specifically the principal component analysis method; The machine learning models include LSTM model, random forest model, XGBoost model, ANN model and stacked regression model, and the specific tuning method is Bayesian optimization method or hyperparameter tuning method.

2. The method according to claim 1, wherein The evaluation metrics used to evaluate the prediction performance of each machine learning model include mean squared error and coefficient of determination.

3. A city-level prediction system for the quantity of retired power batteries based on multi-dimensional feature ML, characterized in that, including: An acquisition module, configured to acquire the insurance volume of new energy vehicles and N types of urban characteristic values within a preset time period to construct a data set, where the N types of urban characteristic values cover key factors affecting the insurance volume of new energy vehicles; A determination module, configured to perform data mining and analysis on the N types of urban characteristic values to determine the key urban characteristic values and their weight values that affect the insurance volume of new energy vehicles, which can screen out the data most relevant to the prediction target, reduce the interference of data noise and irrelevant information, and improve data quality; A training module, configured to screen the data set according to the key urban characteristic values and their weight values to retain the data valuable for predicting the insurance volume of new energy vehicles, remove redundant data, and use the screened data set to train multiple tuned machine learning models, enabling the machine learning models to learn the internal relationship between the key urban characteristic values and the insurance volume of new energy vehicles; A selection module, configured to evaluate the prediction performance of each machine learning model, perform quantitative analysis on the prediction performance of each machine learning model, and select the best models for predicting the insurance volume of passenger vehicles and commercial vehicles respectively from multiple evaluated machine learning models; A processing module, configured to input the future key urban characteristic values into the best model, and the best model will perform operations and inferences based on its internal algorithm architecture and the internal logical relationship between each urban characteristic value and the insurance volume of new energy vehicles learned during the previous training process to predict the insurance volume of new energy vehicles in different cities under specific future scenarios, and then obtain the annual demand of consumers for new energy vehicles based on the predicted insurance volume of new energy vehicles. Using the installed capacity ratio and battery characteristics of the battery types of new energy vehicles, combined with the probability of power battery retirement calculated by Weibull life distribution, predict the amount of retired power batteries in the city, including the amount of retired power batteries from passenger vehicles or commercial vehicles, the amount of retired power batteries of LFP or NMC types, and the total amount of retired power batteries in the city; Among them, the N types of urban characteristic values include the urban population, urban GDP, urbanization rate, Baidu Index, and the promotion subsidy amount of urban new energy vehicles; among them, the urban population, urbanization rate, and urban GDP are obtained from the urban statistical department, the attention of urban netizens to new energy vehicles is obtained from the Baidu Index, and environmental protection policy documents are obtained from relevant government departments and quantified as the promotion subsidy amount of urban new energy vehicles; The future key urban characteristic values include future urban GDP, future urban population, future urbanization rate, future Baidu Index, and future promotion subsidy amount of urban new energy vehicles; The specific battery types include 24 types composed of LFP batteries, NCM111 batteries, NCM523 batteries, NCM622 batteries, NCM811 batteries, and NCA batteries under the matching of pure-electric passenger vehicles, hybrid-electric passenger vehicles, pure-electric commercial vehicles, and hybrid-electric commercial vehicles; the specific battery characteristics include battery life, battery weight, and battery capacity. The method for analyzing the N urban characteristic values to determine the key urban characteristic values affecting the insurance volume of new energy vehicles is specifically the clustering analysis method, and the method for determining the weight values of the key urban characteristics is specifically the principal component analysis method. The machine learning models include LSTM models, random forest models, XGBoost models, ANN models, and stacked regression models, and the specific tuning method is the Bayesian optimization method or the hyperparameter tuning method.

4. A computer device, comprising a memory, a processor, and a computer program stored on the memory, characterized in that, The processor executes the computer program to implement the steps of the method described in any one of claims 1-2.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, the steps of the method described in any one of claims 1-2 are implemented.

Images