Electric vehicle ownership prediction method and system based on multi-source information fusion

Abstract

An electric vehicle ownership prediction method and system based on multi-source information fusion comprises the following steps: S1, capturing training data from a database by using an automatic data capturing tool and an API interface, and carrying out classification marking on the training data, S2, carrying out data expansion on an original electric vehicle data training set based on a dynamic fuzzy membership driven SMOTE algorithm, and carrying out prediction on the electric vehicle ownership. S3, constructing a six-layer full-connection neural network algorithm, performing data feature extraction on the expanded electric vehicle data training set, and training a feature extraction model, and S4, classifying the electric vehicle data training set after feature extraction based on a random forest classification algorithm of fuzzy set optimization, and training a classifier model. And S5, inputting a new electric vehicle data set sample into the trained model to obtain a classification result, namely an electric vehicle inventory prediction result. The design is not only high in flexibility and adaptability, but also accurate in classification.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and system for predicting the number of electric vehicles based on multi-source information fusion. Background Art

[0002] As the global energy crisis and environmental problems intensify, electric vehicles, as a clean energy means of transportation, are gradually becoming the focus of attention of governments and the public. In order to cope with climate change and reduce carbon emissions, many countries have introduced policies to support the development of the electric vehicle industry, which has promoted the rapid growth of the electric vehicle market. The forecast of electric vehicle ownership has become an important basis for relevant policy formulation, market planning and infrastructure construction. However, there are many challenges in accurately predicting the ownership of electric vehicles.

[0003] 1. Traditional electric vehicle ownership forecasts usually rely on a single data source or a small amount of data features, such as vehicle sales data, registration data, etc., and ignore other key factors that may affect the forecast (such as economic development, policy support, charging facilities, etc.), resulting in limited accuracy and generalization of the forecast results.

[0004] 2. When processing minority class data, the SMOTE algorithm commonly used in the prior art expands samples in a relatively mechanical manner and fails to fully consider the distribution characteristics and correlation of samples. It may generate unreasonable samples and even introduce noise, affecting the model training effect.

[0005] 3. Existing neural network training methods are usually based on fixed rules and cannot be dynamically adjusted according to the distribution characteristics of the data or changes in the loss function. They are prone to gradient vanishing or exploding problems and have weak generalization capabilities.

[0006] 4. Traditional random forest algorithms or other classification methods do not take into account the uncertainty and ambiguity in the data, which leads to the model being prone to overfitting or being unable to effectively process fuzzy information when processing multi-source heterogeneous data, thus affecting the accuracy of classification. Summary of the invention

[0007] The purpose of the present invention is to overcome the problems existing in the prior art and to provide a method and system for predicting the number of electric vehicles based on multi-source information fusion to overcome the above problems.

[0008] To achieve the above objectives, the technical solution of the present invention is:

[0009] In one aspect, the present invention provides a method for predicting the number of electric vehicles based on multi-source information fusion, comprising the following steps:

[0010] S1 uses automated data capture tools and API interfaces to capture training data samples from the database, classify and annotate them, and obtain the original electric vehicle data training set;

[0011] S2 is based on the SMOTE algorithm driven by dynamic fuzzy membership, which expands the original electric vehicle data training set and trains the data expansion model to obtain the expanded electric vehicle data training set;

[0012] S3 constructs a 6-layer fully connected neural network algorithm to extract features from the expanded electric vehicle data training set, and trains the feature extraction model to obtain the electric vehicle data training set after feature extraction;

[0013] S4 classifies the electric vehicle data training set after feature extraction based on the random forest classification algorithm optimized by fuzzy sets, and trains the classifier model to obtain the classification category of electric vehicle ownership;

[0014] S5 inputs the new electric vehicle data set into the trained feature extraction model to obtain the electric vehicle data set after feature extraction, and inputs the electric vehicle data set after feature extraction into the trained classifier model to obtain the classification result, which is the electric vehicle ownership prediction result.

[0015] In S1, training data collection is performed based on multi-source heterogeneous data indicators, and the training data includes: traffic flow data, electric vehicle sales and registration data, economic development index, policy support, and charging infrastructure distribution;

[0016] The classification labels include: high growth, medium growth, and low growth.

[0017] The dynamic fuzzy membership-driven SMOTE algorithm includes:

[0018] S21 takes the original electric vehicle data training set as the set of all electric vehicle data samples X c , for X c For each electric vehicle data sample in , calculate its fuzzy membership to the minority class. The minority class refers to the class with fewer samples in the high-growth, medium-growth, and low-growth data. The calculation of the membership takes into account multiple attribute values ​​of the electric vehicle data sample and its distance relative to other electric vehicle data samples. Through fuzzy logic, each electric vehicle data sample is assigned a membership value between 0 and 1. The membership value represents the possibility that the electric vehicle data sample belongs to the minority class, which is expressed as:

[0019]

[0020] In the formula, x crepresents a single electric vehicle data sample; m c () represents the fuzzy membership function; X c represents the set of all electric vehicle data samples, k c and σ c are the parameters for adjusting the steepness and position of the curve, d c (x c ,X c ) is the electric vehicle data sample x c To X c The distance function of all electric vehicle data samples in is expressed as:

[0021]

[0022] Where ncs represents the preset electric vehicle data sample x c The number of features, represents the i-th feature of a single electric vehicle data sample; represents the i-th feature of a sample in the set of all electric vehicle data samples; is the importance weight for each dimension, expressed as:

[0023]

[0024] In the formula, is the standard deviation of the ith dimension in the electric vehicle data sample set, ∈ c is a small constant to avoid the denominator being zero;

[0025] S22 selects and generates electric vehicle data samples. According to the membership value of each electric vehicle data sample, a minority of electric vehicle data samples that need to be expanded are dynamically selected. For electric vehicle data samples with higher membership, more similar electric vehicle data samples are selected for synthesis. The generation method is to interpolate between the selected electric vehicle data sample and its neighboring electric vehicle data samples, and define a function for generating the number of electric vehicle data samples. The function for generating the number of electric vehicle data samples determines the number of new electric vehicle data samples that need to be generated for each electric vehicle data sample based on the membership value. The calculation method is expressed as:

[0026]

[0027] Where gc() is the function for generating the number of electric vehicle data samples; c It is the preset expansion factor, which determines the order of magnitude of the overall electric vehicle data sample generation; represents the gamma function; δ c (x c ) is the adjustment factor based on the sample density of electric vehicle data;

[0028] New electric vehicle data samples are generated by interpolation method. For each selected electric vehicle data sample, a neighboring electric vehicle data sample is randomly selected, and then a new electric vehicle data sample is generated, which is expressed as:

[0029]

[0030] In the formula, represents the generated new electric vehicle data sample; γ c is a random number uniformly distributed from [0, 1], used to control the ratio of interpolation; n c represents a randomly selected sample of neighboring electric vehicle data;

[0031] The adjustment factor based on the density of electric vehicle data samples makes the number of electric vehicle data samples generated inversely proportional to the density around the electric vehicle data samples, ensuring that fewer new electric vehicle data samples are generated in areas with denser electric vehicle data samples, thereby maintaining the balance of electric vehicle data sample distribution. The calculation method is expressed as:

[0032]

[0033] Where N c (x c ) is x c In X c The neighboring electric vehicle data sample set in , σ c is the width parameter of the Gaussian kernel; x c ′ is in X c Any sample in the neighboring electric vehicle data sample set; || || is the L2 norm;

[0034] S23 performs high-dimensional mapping adjustment, and adjusts the high-dimensional space mapping of the newly generated electric vehicle data samples to ensure that the electric vehicle data samples maintain the original distribution characteristics in the new feature space, and optimizes the local and global structures of the electric vehicle data samples. Specifically, the mapping function is used to adjust and refine the features, which is expressed as:

[0035]

[0036] In the formula, Θ c Represents the parameter set of the mapping function, y c is the new electric vehicle data sample after mapping; f c () is a mapping function, specifically a multi-layer neural network, used to learn the nonlinear mapping from the original feature space to the new feature space;

[0037] S24 Integration into the training set: Integrate the expanded and adjusted new electric vehicle data samples into the original training electric vehicle data set to form a new electric vehicle data set to provide data support for subsequent feature extraction and model training;

[0038] Finally, all the generated new electric vehicle data samples are integrated into the original training set, and the training electric vehicle dataset is updated, which is expressed as:

[0039]

[0040] In the formula, represents the expanded electric vehicle dataset; Y c is the set of all newly generated electric vehicle data samples, the updated training electric vehicle dataset That is: the expanded electric vehicle data training set.

[0041] The 6-layer fully connected neural network algorithm includes:

[0042] S31 generates an initial population according to the bionic algorithm initialization method in the initialization stage. Each individual represents a configuration of network weights. Specifically, the population size is N. p , initialize the weights and biases for the i-th individual, expressed as:

[0043]

[0044]

[0045] Where W pi is the weight of the neural network corresponding to the i-th individual, b pi is the bias of the neural network corresponding to the i-th individual, Represents the weight matrix of the i-th individual in the initial state; represents the bias of the i-th individual in the initial state; σ 2 represents the initialization variance; Indicates that the mean is 0 and the variance is σ 2 Normal distribution of is a normal distribution;

[0046] S32 For each individual in the population, use its corresponding neural network configuration to process the samples in the input expanded electric vehicle data training set, calculate the output of the model, and evaluate its performance according to a predetermined loss function. Specifically, for the i-th individual, use its weight and bias to calculate the loss on the training electric vehicle data set, expressed as:

[0047]

[0048] Where Lpi is the loss of the neural network corresponding to the i-th individual; mps represents the number of samples input in the current batch; l() represents the composite loss function, and fsig() represents the neural network model function; represents the electric vehicle data characteristics of the jth sample; represents the label of the jth sample; the composite loss function contains a regularization term, expressed as:

[0049]

[0050] Where MSE() is the mean square error function, λ ps is the regularization parameter, Reg(W pi ) is the regularization term, which is:

[0051]

[0052] Where W pi，k is the kth weight of the neural network corresponding to the i-th individual;

[0053] S33 selects the best performing individuals from the current population according to their fitness and retains them as candidate solutions for the next generation. Excellent individuals have a higher probability of being selected based on their fitness. The calculation method of the probability of an individual being selected is expressed as:

[0054]

[0055] Where P select (i) represents the probability of the i-th individual being selected; γ pse is the parameter that controls the selection pressure; L pk is the loss of the neural network corresponding to the kth individual;

[0056] S34 generates new individuals through crossover and mutation operations. The crossover operation allows two excellent individuals to exchange some genes to generate new offspring; the mutation operation randomly changes some genes in an individual to increase the diversity of the population. Specifically, the crossover operation randomly selects two individuals for gene exchange, which is expressed as:

[0057] W′ pi =α pcs W p1 +(1-α pcs )W p2

[0058] b′ pi =α pcs b p1 +(1-α pcs )b p2

[0059] Where W′pi is the weight of the neural network corresponding to the individual after the crossover operation, α pcs is the crossover rate, W p1 is the weight of the neural network corresponding to the first selected individual, W p2 is the weight of the neural network corresponding to the selected second individual, b′ pi is the bias of the neural network corresponding to the individual after the crossover operation, b p1 is the bias of the neural network corresponding to the first selected individual, b p2 is the bias of the neural network corresponding to the selected second individual;

[0060] The mutation operation performs a small random perturbation on the newly generated individual weights, expressed as:

[0061]

[0062]

[0063] Where W″ pi is the weight of the neural network corresponding to the individual after the mutation operation, b″ pi is the bias of the neural network corresponding to the individual after the mutation operation, τ 2 represents the variance of variation;

[0064] S35 repeats the above steps until the preset stop iteration condition is met, which means that the model training is completed and the electric vehicle data training set after feature extraction is obtained.

[0065] The fuzzy set optimized random forest classification algorithm includes:

[0066] S41 inputs the electric vehicle data training set after feature extraction into the classifier to train the classifier model, initializes multiple decision trees in the random forest, and each tree will be trained independently in the future. Set the number of trees T u and the size of the feature subset used for each tree F u In order to integrate fuzzy logic into the classifier, the fuzzy membership function is defined as It measures the importance of features for classification decisions. The output membership is used to adjust subsequent feature selection and sample split point selection, expressed as:

[0067]

[0068] In the formula, represents the i-th feature of the feature vector input to the random forest, β u is a parameter that adjusts the sensitivity of fuzzy logic, is the average value of the feature in the training set;

[0069] S42 uses fuzzy set theory to optimize data processing during the training of each decision tree. Fuzzy logic is used to select each feature and split point to deal with the inconsistency and ambiguity caused by multi-source data fusion, including the use of fuzzy membership functions to evaluate the importance of features and the contribution of data points to the classification results. The selection weight of the i-th feature is set to It is expressed as:

[0070]

[0071] In the formula, IG() is the information gain function, expressed as:

[0072]

[0073] In the formula, H(S u ) is the set S u The entropy of By feature After the segmentation, L and R represent the left and right subsets; v is an element in the set consisting of the left and right subsets;

[0074] The information gain is calculated by comparing the entropy of the parent node and the entropy of the child node after the split, expressed as:

[0075]

[0076] In the formula, p j is the proportion of samples of the jth class in the node;

[0077] For the features and split points that maximize the information gain of the node split, the entropy of the child nodes after splitting is:

[0078]

[0079] In the formula, p v is the proportion of positive classes in child node v;

[0080] Each S43 tree is trained using a randomly selected data subset and feature subset. During the tree construction process, the fuzzy set optimization standard is used to select the best split point. The split of each node is aimed at maximizing the purity improvement of subsequent nodes. The optimized feature selection weights are used when splitting nodes:

[0081]

[0082] In the formula, Split represents node splitting; tu is the splitting threshold; argmax{} represents the index function that returns the maximum value:

[0083] In the process of tree construction, nodes are split iteratively until the stopping condition is met;

[0084] A gradient optimization-based method is used to determine the optimal splitting threshold to improve the calculation accuracy of information gain and segmentation efficiency, which is expressed as:

[0085]

[0086] In the formula, ΔIG represents the change in information gain, which is solved by estimating the gradient of the information gain function;

[0087] A numerical method is used to find the optimal value of tu, expressed as:

[0088]

[0089] In the formula, It is the first-order derivative of information gain with respect to tu, indicating the rate of change of information gain at tu; is the second-order derivative of information gain with respect to tu, indicating the change in the rate of change; tu new represents the updated split threshold; tu old represents the split threshold before updating;

[0090] S44 After all decision trees are trained, they will be integrated together to form the final random forest model. The classification decision is made by voting or averaging the prediction results of all individual trees. The voting weight of each tree for a sample classification is related to its performance on the training set, expressed as:

[0091]

[0092] In the formula, C(x) is the voting result of sample x, p t (x) is the classification output of the t-th tree for sample x; mode{} is the voting strategy function of ensemble learning;

[0093] The classification category of electric vehicle ownership is obtained according to the classification decision.

[0094] In another aspect, the present invention provides an electric vehicle ownership prediction system based on multi-source information fusion, comprising: a training data capture module, a training data expansion module, a feature extraction module, a classifier module and an ownership prediction module;

[0095] The training data capture module is used to capture training data samples from the database using automated data capture tools and API interfaces, and classify and annotate them to obtain the original electric vehicle data training set;

[0096] The training data expansion module is used to expand the original electric vehicle data training set based on the SMOTE algorithm driven by dynamic fuzzy membership, and train the data expansion model to obtain the expanded electric vehicle data training set;

[0097] The feature extraction module is used to construct a 6-layer fully connected neural network, extract features from the expanded electric vehicle data training set, and train the feature extraction model to obtain the electric vehicle data training set after feature extraction;

[0098] The classifier module is used to classify the electric vehicle data training set after feature extraction using a random forest classification algorithm based on fuzzy set optimization, and to train the classifier model to obtain the classification category of the number of electric vehicles in use;

[0099] Ownership prediction module: used to input the new electric vehicle data set into the trained feature extraction model to obtain the electric vehicle data set after feature extraction, and input the electric vehicle data set after feature extraction into the trained classifier model to obtain the classification result, which is the electric vehicle ownership prediction result.

[0100] In the training data capture module, training data collection is performed based on multi-source heterogeneous data indicators, and the training data includes: traffic flow data, electric vehicle sales and registration data, economic development index, policy support and charging infrastructure distribution;

[0101] The classification labels include: high growth, medium growth, and low growth.

[0102] In the training data expansion module, S21 uses the original electric vehicle data training set as the set X of all electric vehicle data samples c , for X c For each electric vehicle data sample in , calculate its fuzzy membership to the minority class. The minority class refers to the class with fewer samples in the high-growth, medium-growth, and low-growth data. The calculation of the membership takes into account multiple attribute values ​​of the electric vehicle data sample and its distance relative to other electric vehicle data samples. Through fuzzy logic, each electric vehicle data sample is assigned a membership value between 0 and 1. The membership value represents the possibility that the electric vehicle data sample belongs to the minority class, which is expressed as:

[0103]

[0104] In the formula, x c represents a single electric vehicle data sample; m c () represents the fuzzy membership function; X c represents the set of all electric vehicle data samples, k c and σc are the parameters for adjusting the steepness and position of the curve, d c (x c , X c ) is the electric vehicle data sample x c To X c The distance function of all electric vehicle data samples in is expressed as:

[0105]

[0106] Where ncs represents the preset electric vehicle data sample x c The number of features, represents the i-th feature of a single electric vehicle data sample; represents the i-th feature of a sample in the set of all electric vehicle data samples; is the importance weight for each dimension, expressed as:

[0107]

[0108] In the formula, is the standard deviation of the ith dimension in the electric vehicle data sample set, ∈ c is a small constant to avoid the denominator being zero;

[0109] S22 selects and generates electric vehicle data samples. According to the membership value of each electric vehicle data sample, dynamically selects a minority of electric vehicle data samples that need to be expanded. For electric vehicle data samples with higher membership, more similar electric vehicle data samples are selected for synthesis. The generation method is to interpolate between the selected electric vehicle data sample and its neighboring electric vehicle data samples, and define a function for generating the number of electric vehicle data samples. The function for generating the number of electric vehicle data samples determines the number of new electric vehicle data samples that need to be generated for each electric vehicle data sample based on the membership. The calculation method is expressed as:

[0110]

[0111] In the formula, g c () is the function for generating the number of electric vehicle data samples; λ c It is the preset expansion factor, which determines the order of magnitude of the overall electric vehicle data sample generation; represents the gamma function; δ c (x c ) is the adjustment factor based on the sample density of electric vehicle data;

[0112] New electric vehicle data samples are generated by interpolation method. For each selected electric vehicle data sample, a neighboring electric vehicle data sample is randomly selected, and then a new electric vehicle data sample is generated, which is expressed as:

[0113]

[0114] In the formula, represents the generated new electric vehicle data sample; γ c is a random number uniformly distributed from [0, 1], used to control the ratio of interpolation; n c represents a randomly selected sample of neighboring electric vehicle data;

[0115] The adjustment factor based on the density of electric vehicle data samples makes the number of electric vehicle data samples generated inversely proportional to the density around the electric vehicle data samples, ensuring that fewer new electric vehicle data samples are generated in areas with denser electric vehicle data samples, thereby maintaining the balance of electric vehicle data sample distribution. The calculation method is expressed as:

[0116]

[0117] Where N c (x c ) is x c In X c The neighboring electric vehicle data sample set in , σ c is the width parameter of the Gaussian kernel; x c ′ For X c Any sample in the neighboring electric vehicle data sample set; ∥∥ is the L2 norm;

[0118] S23 performs high-dimensional mapping adjustment, and adjusts the high-dimensional space mapping of the newly generated electric vehicle data samples to ensure that the electric vehicle data samples maintain the original distribution characteristics in the new feature space, and optimizes the local and global structures of the electric vehicle data samples. Specifically, the mapping function is used to adjust and refine the features, which is expressed as:

[0119]

[0120] In the formula, Θ c Represents the parameter set of the mapping function, y c is the new electric vehicle data sample after mapping; f c () is a mapping function, specifically a multi-layer neural network, used to learn the nonlinear mapping from the original feature space to the new feature space;

[0121] S24 Integration into the training set: Integrate the expanded and adjusted new electric vehicle data samples into the original training electric vehicle data set to form a new electric vehicle data set to provide data support for subsequent feature extraction and model training;

[0122] Finally, all the generated new electric vehicle data samples are integrated into the original training set, and the training electric vehicle dataset is updated, which is expressed as:

[0123]

[0124] In the formula, represents the expanded electric vehicle dataset; Y c is the set of all newly generated electric vehicle data samples, the updated training electric vehicle dataset That is: the expanded electric vehicle data training set.

[0125] In the feature extraction module: S31 generates an initial population according to the bionic algorithm initialization method in the initialization stage, each body represents a configuration of network weights, and the population size is N p , initialize the weights and biases for the i-th individual, expressed as:

[0126]

[0127]

[0128] Where W pi is the weight of the neural network corresponding to the i-th individual, b pi is the bias of the neural network corresponding to the i-th individual, Represents the weight matrix of the i-th individual in the initial state; represents the bias of the i-th individual in the initial state; σ 2 represents the initialization variance; Indicates that the mean is 0 and the variance is σ 2 Normal distribution of is a normal distribution;

[0129] S32 For each individual in the population, use its corresponding neural network configuration to process the samples in the input expanded electric vehicle data training set, calculate the output of the model, and evaluate its performance according to a predetermined loss function. Specifically, for the i-th individual, use its weight and bias to calculate the loss on the training electric vehicle data set, expressed as:

[0130]

[0131] Where Lpi is the loss of the neural network corresponding to the i-th individual; mps represents the number of samples input in the current batch; e() represents the composite loss function, and fsig() represents the neural network model function; represents the electric vehicle data characteristics of the jth sample; represents the label of the jth sample; the composite loss function contains a regularization term, expressed as:

[0132]

[0133] Where MSE() is the mean square error function, λ ps is the regularization parameter, Reg(W pi ) is the regularization term, which is:

[0134]

[0135] Where Wpi,k is the kth weight of the neural network corresponding to the i-th individual;

[0136] S33 selects the best performing individuals from the current population according to their fitness and retains them as candidate solutions for the next generation. Excellent individuals have a higher probability of being selected based on their fitness. The calculation method of the probability of an individual being selected is expressed as:

[0137]

[0138] Where P select (i) represents the probability of the i-th individual being selected; γ pse is the parameter that controls the selection pressure; L pk is the loss of the neural network corresponding to the kth individual;

[0139] S34 generates new individuals through crossover and mutation operations. The crossover operation allows two excellent individuals to exchange some genes to generate new offspring; the mutation operation randomly changes some genes in an individual to increase the diversity of the population. Specifically, the crossover operation randomly selects two individuals for gene exchange, which is expressed as:

[0140] W′ pi =α pcs W p1 +(1-α pcs )W p2

[0141] b′ pi =α pcs b p1 +(1-α pcs )b p2

[0142] Where W′ piis the weight of the neural network corresponding to the individual after the crossover operation, α pcs is the crossover rate, W p1 is the weight of the neural network corresponding to the first selected individual, W p2 is the weight of the neural network corresponding to the selected second individual, b′ pi is the bias of the neural network corresponding to the individual after the crossover operation, b p1 is the bias of the neural network corresponding to the first selected individual, b p2 is the bias of the neural network corresponding to the selected second individual;

[0143] The mutation operation performs a small random perturbation on the newly generated individual weights, expressed as:

[0144]

[0145]

[0146] Where W″ pi is the weight of the neural network corresponding to the individual after the mutation operation, b″ pi is the bias of the neural network corresponding to the individual after the mutation operation, τ 2 represents the variance of variation;

[0147] S35 repeats the above steps until the preset stop iteration condition is met, which means that the model training is completed and the electric vehicle data training set after feature extraction is obtained.

[0148] The classifier module includes: S41 inputting the electric vehicle data training set after feature extraction into the classifier to train the classifier model, initializing multiple decision trees in the random forest, each tree will be trained independently in the subsequent training, and setting the number of trees T u and the size of the feature subset used for each tree F u In order to integrate fuzzy logic into the classifier, the fuzzy membership function is defined as It measures the importance of features for classification decisions. The output membership is used to adjust subsequent feature selection and sample split point selection, expressed as:

[0149]

[0150] In the formula, represents the i-th feature of the feature vector input to the random forest, β u is a parameter that adjusts the sensitivity of fuzzy logic, is the average value of the feature in the training set;

[0151] S42 uses fuzzy set theory to optimize data processing during the training of each decision tree. Fuzzy logic is used to select each feature and split point to deal with the inconsistency and ambiguity caused by multi-source data fusion, including the use of fuzzy membership functions to evaluate the importance of features and the contribution of data points to the classification results. The selection weight of the i-th feature is set to It is expressed as:

[0152]

[0153] In the formula, IG() is the information gain function, expressed as:

[0154]

[0155] In the formula, H(S u ) is the set S u The entropy of By feature After the segmentation, L and R represent the left and right subsets; v is an element in the set consisting of the left and right subsets;

[0156] The information gain is calculated by comparing the entropy of the parent node and the entropy of the child node after the split, expressed as:

[0157]

[0158] In the formula, p i is the proportion of samples of the jth class in the node;

[0159] For the features and split points that maximize the information gain of the node split, the entropy of the child nodes after splitting is:

[0160]

[0161] In the formula, p v is the proportion of positive classes in child node v;

[0162] Each S43 tree is trained using a randomly selected data subset and feature subset. During the tree construction process, the fuzzy set optimization standard is used to select the best split point. The split of each node is aimed at maximizing the purity improvement of subsequent nodes. The optimized feature selection weights are used when splitting nodes:

[0163]

[0164] In the formula, Split represents node splitting; tu is the splitting threshold; argmax{} represents the index function that returns the maximum value:

[0165] In the process of tree construction, nodes are split iteratively until the stopping condition is met;

[0166] A gradient optimization-based method is used to determine the optimal splitting threshold to improve the calculation accuracy of information gain and segmentation efficiency, which is expressed as:

[0167]

[0168] In the formula, ΔIG represents the change in information gain, which is solved by estimating the gradient of the information gain function;

[0169] A numerical method is used to find the optimal value of tu, expressed as:

[0170]

[0171] In the formula, It is the first-order derivative of information gain with respect to tu, indicating the rate of change of information gain at tu; is the second-order derivative of information gain with respect to tu, indicating the change in the rate of change; tu new represents the updated split threshold; tu old represents the split threshold before updating;

[0172] S44 After all decision trees are trained, they will be integrated together to form the final random forest model. The classification decision is made by voting or averaging the prediction results of all individual trees. The voting weight of each tree for a sample classification is related to its performance on the training set, expressed as:

[0173]

[0174] In the formula, C(x) is the voting result of sample x, p t (x) is the classification output of the tth tree for sample x; mode{} is the voting strategy function of ensemble learning.

[0175] Compared with the prior art, the present invention has the following beneficial effects:

[0176] 1. In the electric vehicle ownership prediction method based on multi-source information fusion of the present invention, the SMOTE algorithm driven by dynamic fuzzy membership is used to expand the electric vehicle data. By evaluating the fuzzy membership of each electric vehicle sample, the number of new electric vehicle samples generated by each minority electric vehicle sample is dynamically adjusted according to the membership, thereby improving the quality and effect of electric vehicle data expansion; through high-dimensional mapping adjustment, it is ensured that the distribution of the newly generated electric vehicle samples in the feature space maintains the original characteristics, avoiding the dimensionality explosion and data distribution imbalance problems caused by high-dimensional data expansion in the prior art.

[0177] 2. In the electric vehicle ownership prediction method based on multi-source information fusion of the present invention, a 6-layer fully connected neural network is used to extract electric vehicle data features. In the prior art, some schemes use neural networks to extract electric vehicle data features. In some neural network structures, gradient disappearance, gradient explosion or falling into the local optimal solution may be encountered, affecting the stability of training and the performance of the model. The present invention adopts a neural network algorithm based on dynamic population evolution optimization as an electric vehicle data feature extraction model. In the traditional population evolution algorithm, the evolution of all individuals is based on fixed rules. The present invention uses a self-correction mechanism to automatically adjust the evolution rules according to the characteristics of the current training electric vehicle data, and adjusts the probability of crossover and mutation according to the changing trend of the loss function, so that the algorithm can adapt to different electric vehicle data distributions more flexibly, improve the generalization ability and training efficiency of the model, and avoid falling into the local optimal solution.

[0178] 3. The electric vehicle ownership prediction method based on multi-source information fusion of the present invention adopts a random forest classification algorithm based on fuzzy set optimization, and improves the accuracy and stability of classification based on fuzzy logic and random forest algorithm. Through fuzzy set optimization, the feature selection and split point selection of each decision tree in the construction process are more accurate, and can better handle the uncertainty and ambiguity in the data.

[0179] 4. The electric vehicle ownership prediction system based on multi-source information fusion of the present invention includes a training data capture module, which is used to use an automated data capture tool and an API interface to capture training data from a database, and classify and annotate the training data to obtain an original electric vehicle data training set; a training data expansion module, which is used to perform electric vehicle data expansion on the original electric vehicle data training set based on a SMOTE algorithm driven by dynamic fuzzy membership, and train a data expansion model to obtain an expanded electric vehicle data training set; a feature extraction module, which is used to construct a 6-layer fully connected neural network, perform electric vehicle data feature extraction on the expanded electric vehicle data training set, and train a feature extraction model to obtain an electric vehicle data training set after feature extraction; a classifier module, which is used to classify the electric vehicle data training set after feature extraction based on a random forest classification algorithm optimized by fuzzy sets, and train a classifier model to obtain a classification category training result of the electric vehicle ownership; and a vehicle ownership prediction module: which is used to input a new electric vehicle data set sample into a trained feature extraction model to obtain an electric vehicle data set after feature extraction, and input the electric vehicle data set after feature extraction into a trained classifier model to obtain a classification result, which is an electric vehicle ownership prediction result. The system is used to implement the steps of the electric vehicle ownership prediction method based on multi-source information fusion as provided in any of the above technical solutions. Therefore, the decoding system also includes all the beneficial effects of the electric vehicle ownership prediction method based on multi-source information fusion as provided in any of the above technical solutions, which will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0180] Figure 1 It is a flow chart of the method of the present invention.

[0181] Figure 2 It is an algorithm flow chart of the feature extraction model of the present invention.

[0182] Figure 3 It is a system schematic diagram of the present invention. DETAILED DESCRIPTION

[0183] The present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0184] Embodiment 1:

[0185] See also Figure 1 , Figure 2 , a method for predicting the number of electric vehicles based on multi-source information fusion, comprising the following steps:

[0186] S1 uses automated data capture tools and API interfaces to capture training data from the database, classify and annotate it, and obtain the original electric vehicle data training set;

[0187] The training data is collected based on multi-source heterogeneous data indicators, including traffic flow data, electric vehicle sales and registration data, economic development index, policy support and charging infrastructure distribution;

[0188] The training data collection method uses an automated data capture tool and API interface to obtain data from government databases, traffic monitoring systems, economic research institutions, and related corporate databases;

[0189] The training data is stored in a structured database system, such as SQL or NoSQL, to ensure data integrity and query efficiency;

[0190] In one embodiment, the attributes of the data include:

[0191] Ra1 is the traffic flow index, which indicates the vehicle circulation volume in a specific area; Ra2 is the electric vehicle sales volume, which indicates the number of electric vehicles sold in the area; Ra3 is the registration data volume, which indicates the number of electric vehicles registered in the area; Ra4 is the economic development index, which integrates economic indicators such as GDP and per capita income in the region; Ra5 is the policy support index, which is compiled according to the electric vehicle support policies of local governments; Ra6 is the charging facility density, which indicates the number of charging piles per unit area; Ra7 is user preference, which is data obtained through social media and market surveys; Ra8 is the environmental protection index, which is calculated based on the environmental protection policies and implementation status in the region; Ra9 is the technology maturity, which is the development level of electric vehicles and related technologies in the region; Ra10 is the market potential index, which predicts the future growth potential of the electric vehicle market in the region.

[0192] It should be noted that this embodiment is only used to illustrate one data format and type of the present invention. In actual applications, the attributes of data are usually more than 10 attributes, and the number of attributes of data may reach dozens or even hundreds.

[0193] Furthermore, the collected data is labeled. The labeling method of the present invention is manual labeling. In one embodiment, the labeled categories include: "high growth", "medium growth" and "low growth".

[0194] S2 is based on the SMOTE algorithm driven by dynamic fuzzy membership, which expands the original electric vehicle data training set and trains the data expansion model to obtain the expanded electric vehicle data training set;

[0195] In the task of the present invention, the collection, acquisition, labeling and preprocessing of training data are time-consuming and labor-intensive, and insufficient electric vehicle samples can easily lead to poor model generalization ability and affect the accuracy of the model. The present invention adopts the SMOTE algorithm driven by dynamic fuzzy membership to expand electric vehicle data. By evaluating the fuzzy membership of each electric vehicle sample, the number of new electric vehicle samples generated by each minority electric vehicle sample is dynamically adjusted according to the membership, thereby improving the quality and effect of electric vehicle data expansion.

[0196] Specifically, the process of electric vehicle data expansion based on the SMOTE algorithm driven by dynamic fuzzy membership is expressed as follows:

[0197] S21 takes the original electric vehicle data training set as the set of all electric vehicle data samples X c , for X c For each electric vehicle data sample in , calculate its fuzzy membership to the minority class. The minority class refers to the class with fewer samples in the high-growth, medium-growth, and low-growth data. The calculation of the membership takes into account multiple attribute values ​​of the electric vehicle data sample and its distance relative to other electric vehicle data samples. Through fuzzy logic, each electric vehicle data sample is assigned a membership value between 0 and 1. The membership value represents the possibility that the electric vehicle data sample belongs to the minority class, which is expressed as:

[0198]

[0199] In the formula, x c represents a single electric vehicle data sample; m c () represents the fuzzy membership function; X c represents the set of all electric vehicle data samples, k c and σ c are the parameters for adjusting the steepness and position of the curve, d c (x c , x c ) is the electric vehicle data sample x c To X c The distance function of all electric vehicle data samples in is expressed as:

[0200]

[0201] Where ncs represents the preset electric vehicle data sample x c The number of features, represents the i-th feature of a single electric vehicle data sample; represents the i-th feature of a sample in the set of all electric vehicle data samples; is the importance weight for each dimension, expressed as:

[0202]

[0203] In the formula, is the standard deviation of the ith dimension in the electric vehicle data sample set, ∈ c is a small constant to avoid the denominator being zero; preferably, ∈ c Set to 0.001.

[0204] S22 selects and generates electric vehicle data samples. According to the membership value of each electric vehicle data sample, a minority of electric vehicle data samples that need to be expanded are dynamically selected. For electric vehicle data samples with higher membership, more similar electric vehicle data samples are selected for synthesis. The generation method is to interpolate between the selected electric vehicle data sample and its neighboring electric vehicle data samples, and define a function for generating the number of electric vehicle data samples. The function for generating the number of electric vehicle data samples determines the number of new electric vehicle data samples that need to be generated for each electric vehicle data sample based on the membership value. The calculation method is expressed as:

[0205]

[0206] In the formula, g c () is the function for generating the number of electric vehicle data samples; λ c It is the preset expansion factor, which determines the order of magnitude of the overall electric vehicle data sample generation; represents the gamma function; δ c (x c ) is an adjustment factor based on the electric vehicle data sample density; preferably, λ c Set to 2.

[0207] New electric vehicle data samples are generated by interpolation method. For each selected electric vehicle data sample, a neighboring electric vehicle data sample is randomly selected, and then a new electric vehicle data sample is generated, which is expressed as:

[0208]

[0209] In the formula, represents the generated new electric vehicle data sample; γ c is a random number uniformly distributed from [0, 1], used to control the ratio of interpolation; n c represents a randomly selected sample of neighboring electric vehicle data;

[0210] The adjustment factor based on the density of electric vehicle data samples makes the number of electric vehicle data samples generated inversely proportional to the density around the electric vehicle data samples, ensuring that fewer new electric vehicle data samples are generated in areas with denser electric vehicle data samples, thereby maintaining the balance of electric vehicle data sample distribution. The calculation method is expressed as:

[0211]

[0212] Where N c (x c ) is x c In X c The neighboring electric vehicle data sample set in , σ c is the width parameter of the Gaussian kernel; x c ′ is in X c Any sample in the neighboring electric vehicle data sample set; ∥∥ is the L2 norm;

[0213] S23 performs high-dimensional mapping adjustment, and adjusts the high-dimensional space mapping of the newly generated electric vehicle data samples to ensure that the electric vehicle data samples maintain the original distribution characteristics in the new feature space, and optimizes the local and global structures of the electric vehicle data samples. Specifically, the mapping function is used to adjust and refine the features, which is expressed as:

[0214]

[0215]

[0216] Where N c (x c ) is x c In X c The neighboring electric vehicle data sample set in , σ c is the width parameter of the Gaussian kernel; x c ′ For X c Any sample in the neighboring electric vehicle data sample set; ∥∥ is the L2 norm;

[0217] S23 performs high-dimensional mapping adjustment, and adjusts the high-dimensional space mapping of the newly generated electric vehicle data samples to ensure that the electric vehicle data samples maintain the original distribution characteristics in the new feature space, and optimizes the local and global structures of the electric vehicle data samples. Specifically, the mapping function is used to adjust and refine the features, which is expressed as:

[0218]

[0219] In the formula, represents the expanded electric vehicle dataset; Y cis the set of all newly generated electric vehicle data samples, the updated training electric vehicle dataset That is: the expanded electric vehicle data training set.

[0220] The existing SMOTE algorithm mainly handles the data imbalance problem through simple oversampling. The present invention adopts fuzzy membership to dynamically adjust the number of minority class samples generated, generates new samples based on the density, characteristics and similarity of electric vehicle samples, expands the data more specifically, and ensures the rationality and quality of the expanded data.

[0221] The present invention not only performs data expansion through SMOTE, but also performs high-dimensional mapping adjustment on the expanded data on this basis to ensure that the distribution characteristics of the newly generated samples in the new feature space remain consistent, thus solving the problem that high-dimensional expansion may cause data distribution imbalance.

[0222] S3 constructs a 6-layer fully connected neural network algorithm to extract data features from the expanded electric vehicle data training set, and trains the feature extraction model to obtain the electric vehicle data training set after feature extraction;

[0223] The expanded electric vehicle data training set is input into the feature extraction model to train the electric vehicle data feature extraction model. The present invention uses a 6-layer fully connected neural network to extract electric vehicle data features. In the prior art, some solutions use neural networks to extract electric vehicle data features. In some neural network structures, gradient vanishing, gradient explosion, or falling into local optimal solutions may be encountered, affecting the stability of training and the performance of the model. The present invention uses a neural network algorithm based on dynamic population evolution optimization as an electric vehicle data feature extraction model. In traditional population evolution algorithms, the evolution of all individuals is based on fixed rules. The present invention uses a self-correction mechanism to automatically adjust the evolution rules according to the characteristics of the current training electric vehicle data, and adjusts the probability of crossover and mutation according to the changing trend of the loss function, so that the algorithm can more flexibly adapt to different electric vehicle data distributions and improve the generalization ability and training efficiency of the model.

[0224] S31 generates an initial population according to the bionic algorithm initialization method in the initialization stage. Each individual represents a configuration of network weights. Specifically, the population size is N. p , initialize the weights and biases for the i-th individual, expressed as:

[0225]

[0226] Where W pi is the weight of the neural network corresponding to the i-th individual, b pi is the bias of the neural network corresponding to the i-th individual, Represents the weight matrix of the i-th individual in the initial state; represents the bias of the i-th individual in the initial state; σ 2 represents the initialization variance; Indicates that the mean is 0 and the variance is σ 2 Normal distribution of is a normal distribution; preferably, σ 2 Set to 0.01.

[0227] S32 For each individual in the population, use its corresponding neural network configuration to process the samples in the input expanded electric vehicle data training set, calculate the output of the model, and evaluate its performance according to a predetermined loss function. Specifically, for the i-th individual, use its weight and bias to calculate the loss on the training electric vehicle data set, expressed as:

[0228]

[0229] Where Lpi is the loss of the neural network corresponding to the i-th individual; mps represents the number of samples input in the current batch; l() represents the composite loss function, and fsig() represents the neural network model function; represents the electric vehicle data characteristics of the jth sample; represents the label of the jth sample; the composite loss function contains a regularization term, expressed as:

[0230]

[0231] Where MSE() is the mean square error function, λ ps is the regularization parameter, Reg(W pi ) is the regularization term, which is:

[0232]

[0233] Where Wpi,k is the kth weight of the neural network corresponding to the i-th individual;

[0234] S33 selects the best performing individuals from the current population according to their fitness and retains them as candidate solutions for the next generation. Excellent individuals have a higher probability of being selected based on their fitness. The calculation method of the probability of an individual being selected is expressed as:

[0235]

[0236] Where P select (i) represents the probability of the i-th individual being selected; γ pse is the parameter that controls the selection pressure; L pk is the loss of the neural network corresponding to the kth individual; preferably, γ pseSet to 2.

[0237] S34 generates new individuals through crossover and mutation operations. The crossover operation allows two excellent individuals to exchange some genes to generate new offspring; the mutation operation randomly changes some genes in an individual to increase the diversity of the population. Specifically, the crossover operation randomly selects two individuals for gene exchange, which is expressed as:

[0238] W′ pi =α pcs W p1 +(1-α pcs )W p2 ; b′ pi =α pcs b p1 +(1-α pcs )b p2 ;

[0239] Where W′ pi is the weight of the neural network corresponding to the individual after the crossover operation, α pcs is the crossover rate, W p1 is the weight of the neural network corresponding to the first selected individual, W p2 is the weight of the neural network corresponding to the selected second individual, b′ pi is the bias of the neural network corresponding to the individual after the crossover operation, b p1 is the bias of the neural network corresponding to the first selected individual, b p2 is the bias of the neural network corresponding to the selected second individual; preferably, α pcs Set to 0.3.

[0240] The mutation operation performs a small random perturbation on the newly generated individual weights, expressed as:

[0241]

[0242] Where W″ pi is the weight of the neural network corresponding to the individual after the mutation operation, b″ pi is the bias of the neural network corresponding to the individual after the mutation operation, τ 2 represents the variance of the variation; preferably, τ 2 Set to 0.04.

[0243] S35 repeats the above steps until the preset stop iteration condition is met, which means that the model training is completed and the electric vehicle data training set after feature extraction is obtained. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0244] The present invention introduces a dynamic population evolution optimization algorithm into the feature extraction model, automatically adjusts the evolution rules through a self-correction mechanism, adjusts the weight and bias of the neural network according to the changing trend of the loss function, flexibly adapts to different electric vehicle data distributions, and enhances the generalization ability and training efficiency of the model.

[0245] S4 classifies the electric vehicle data training set after feature extraction based on the random forest classification algorithm optimized by fuzzy sets, and trains the classifier model to obtain the classification category training results of the number of electric vehicles in use;

[0246] Inputting the electric vehicle data training set after feature extraction into a classifier to train a classifier model, wherein the classifier predicts the number of vehicles in use according to the electric vehicle feature extraction vector obtained after feature extraction;

[0247] The present invention adopts a random forest classification algorithm based on fuzzy set optimization, and improves the accuracy and stability of classification based on fuzzy logic and random forest algorithm. Through fuzzy set optimization, the feature selection and split point selection of each decision tree in the construction process are more accurate, and the uncertainty and ambiguity in the data can be better handled;

[0248] S41 inputs the electric vehicle data training set after feature extraction into the classifier to train the classifier model, initializes multiple decision trees in the random forest, and each tree will be trained independently in the future. Set the number of trees T u and the size of the feature subset used for each tree F u In order to integrate fuzzy logic into the classifier, the fuzzy membership function is defined as It measures the importance of features for classification decisions. The output membership is used to adjust subsequent feature selection and sample split point selection, expressed as:

[0249]

[0250] In the formula, represents the i-th feature of the feature vector input to the random forest, β u is a parameter that adjusts the sensitivity of fuzzy logic, is the average value of the feature in the training set;

[0251] S42 uses fuzzy set theory to optimize data processing during the training of each decision tree. Fuzzy logic is used to select each feature and split point to deal with the inconsistency and ambiguity caused by multi-source data fusion, including the use of fuzzy membership functions to evaluate the importance of features and the contribution of data points to the classification results. The selection weight of the i-th feature is set to It is expressed as:

[0252]

[0253] In the formula, IG() is the information gain function, expressed as:

[0254]

[0255] In the formula, H(S u ) is the set S u The entropy of By feature After the segmentation, L and R represent the left and right subsets; v is the element in the set consisting of the left and right subsets; the information gain is calculated by comparing the entropy of the parent node and the entropy of the child node after segmentation, expressed as:

[0256]

[0257] In the formula, p j is the proportion of samples of the jth class in the node;

[0258] For the features and split points that maximize the information gain of the node split, the entropy of the child nodes after splitting is:

[0259]

[0260] In the formula, p v is the proportion of positive classes in child node v;

[0261] Each S43 tree is trained using a randomly selected data subset and feature subset. During the tree construction process, the fuzzy set optimization standard is used to select the best split point. The split of each node is aimed at maximizing the purity improvement of subsequent nodes. The optimized feature selection weights are used when splitting nodes:

[0262]

[0263] In the formula, Split represents node splitting; tu is the splitting threshold; argmax{} represents the index function that returns the maximum value:

[0264] In the process of tree construction, nodes are split iteratively until the stopping condition is met;

[0265] A gradient optimization-based method is used to determine the optimal splitting threshold to improve the calculation accuracy of information gain and segmentation efficiency, which is expressed as:

[0266] In the formula, ΔIG represents the change in information gain, which is solved by estimating the gradient of the information gain function;

[0267] A numerical method is used to find the optimal value of tu, expressed as:

[0268] In the formula, It is the first-order derivative of information gain with respect to tu, indicating the rate of change of information gain at tu; is the second-order derivative of information gain with respect to tu, indicating the change in the rate of change; tu new represents the updated split threshold; tu old represents the split threshold before updating;

[0269] S44 After all decision trees are trained, they will be integrated together to form the final random forest model. The classification decision is made by voting or averaging the prediction results of all individual trees. The voting weight of each tree for a sample classification is related to its performance on the training set, expressed as:

[0270] In the formula, C(x) is the voting result of sample x, p t (x) is the classification output of the t-th tree for sample x; mode{} is the voting strategy function of ensemble learning;

[0271] The classification category of electric vehicle ownership is obtained according to the classification decision.

[0272] The present invention combines fuzzy set theory and random forest algorithm, and improves the accuracy of feature selection and split point selection through fuzzy logic, which can effectively deal with the ambiguity and uncertainty of multi-source heterogeneous data and improve the classification performance and stability of the model.

[0273] S5 inputs the new electric vehicle data set sample into the trained feature extraction model to obtain the electric vehicle data set after feature extraction, and inputs the electric vehicle data set after feature extraction into the trained classifier model to obtain the classification result, which is the electric vehicle ownership prediction result.

[0274] Embodiment 2:

[0275] An electric vehicle ownership prediction system based on multi-source information fusion, comprising: a training data capture module, a training data expansion module, a feature extraction module, a classifier module and an ownership prediction module;

[0276] The training data capture module is used to capture training data samples from the database using automated data capture tools and API interfaces, and classify and annotate them to obtain the original electric vehicle data training set;

[0277] The training data expansion module is used to expand the original electric vehicle data training set based on the SMOTE algorithm driven by dynamic fuzzy membership, and train the data expansion model to obtain the expanded electric vehicle data training set;

[0278] The feature extraction module is used to construct a 6-layer fully connected neural network, extract features from the expanded electric vehicle data training set, and train the feature extraction model to obtain the electric vehicle data training set after feature extraction;

[0279] The classifier module is used to classify the electric vehicle data training set after feature extraction using a random forest classification algorithm based on fuzzy set optimization, and train the classifier model to obtain the classification category of the number of electric vehicles in use;

[0280] Ownership prediction module: used to input the new electric vehicle data set into the trained feature extraction model to obtain the electric vehicle data set after feature extraction, and input the electric vehicle data set after feature extraction into the trained classifier model to obtain the classification result, which is the electric vehicle ownership prediction result.

[0281] The collected raw electric vehicle data is input into the trained feature extraction and feature processing, and further, the processed features are input into the classifier model for classification, thereby obtaining the classification results. In this embodiment, the classification categories include: "high growth", "medium growth" and "low growth".

[0282] In this embodiment, assuming that a complete electric vehicle training data set has been obtained after multi-source data collection and expansion, and the model training has been completed, the following describes how the model classifies new electric vehicle data to predict the growth of electric vehicle ownership.

[0283] The new EV data contains a variety of features that are in the same format as the training data;

[0284] First, the new data will pass through the trained feature extraction model, which is based on a neural network algorithm optimized by dynamic population evolution. The model extracts key features from the input data. After the model is optimized, these features will be mapped to the new feature space to maintain the overall distribution structure of the data.

[0285] Specifically, the new data will pass through a 6-layer fully connected neural network. Each layer will perform nonlinear transformation using the weights and biases learned in previous training to extract higher-dimensional and more abstract features. The goal is to convert the original multi-dimensional data into a feature vector that can better represent the complex relationship between samples so that it can be processed by subsequent classifiers.

[0286] The extracted feature vector is input into the random forest classifier based on fuzzy set optimization. The random forest classifier consists of multiple decision trees. Each tree makes classification decisions independently and finally votes to decide the classification result through ensemble learning.

[0287] In the process of building each decision tree, the fuzzy membership function is used to evaluate the importance of features. Fuzzy membership helps the model better handle uncertainty and ambiguity, making feature selection and node splitting more accurate;

[0288] In the splitting process of each node, the fuzzy membership function selects the features and splitting points that have the greatest impact on classification by optimizing the calculation of information gain;

[0289] Each tree selects weights based on the optimized features and classifies the samples independently;

[0290] Once each decision tree has independently made a classification decision, the random forest algorithm votes on the predictions of all the decision trees and outputs a category (e.g., “high growth”) if the majority of the trees predict that category.

[0291] Given new electric car data, the classifier will predict it as one of three categories:

[0292] “High growth”: indicates that the electric vehicle market in this region will have the potential for high growth in the future;

[0293] “Medium growth”: indicates that the electric vehicle market in the region will maintain moderate growth;

[0294] “Low Growth”: Indicates that the electric vehicle market in the region is growing slowly or is approaching saturation.

Claims

1. A method for predicting the number of electric vehicles based on multi-source information fusion, characterized in that: The steps include: S1 uses automated data capture tools and API interfaces to capture training data samples from the database, classify and annotate them, and obtain the original electric vehicle data training set; S2 is based on the SMOTE algorithm driven by dynamic fuzzy membership, which expands the original electric vehicle data training set and trains the data expansion model to obtain the expanded electric vehicle data training set; S3 constructs a 6-layer fully connected neural network algorithm to extract features from the expanded electric vehicle data training set, and trains the feature extraction model to obtain the electric vehicle data training set after feature extraction; S4 classifies the electric vehicle data training set after feature extraction based on the random forest classification algorithm optimized by fuzzy sets, and trains the classifier model to obtain the classification category of electric vehicle ownership; S5 inputs the new electric vehicle data set into the trained feature extraction model to obtain the electric vehicle data set after feature extraction, and inputs the electric vehicle data set after feature extraction into the trained classifier model to obtain the classification result, which is the electric vehicle ownership prediction result.

2. The method for predicting the number of electric vehicles based on multi-source information fusion according to claim 1 is characterized in that: In S1, training data collection is performed based on multi-source heterogeneous data indicators, and the training data includes: traffic flow data, electric vehicle sales and registration data, economic development index, policy support, and charging infrastructure distribution; The classification labels include: high growth, medium growth, and low growth.

3. The method for predicting the number of electric vehicles based on multi-source information fusion according to claim 1 is characterized in that: The dynamic fuzzy membership-driven SMOTE algorithm includes: S21 takes the original electric vehicle data training set as the set of all electric vehicle data samples X c , for X c For each electric vehicle data sample in , calculate its fuzzy membership to the minority class. The minority class refers to the class with fewer samples in the high-growth, medium-growth, and low-growth data. The calculation of the membership takes into account multiple attribute values ​​of the electric vehicle data sample and its distance relative to other electric vehicle data samples. Through fuzzy logic, each electric vehicle data sample is assigned a membership value between 0 and 1. The membership value represents the possibility that the electric vehicle data sample belongs to the minority class, which is expressed as: In the formula, x c represents a single electric vehicle data sample; m c () represents the fuzzy membership function; X c represents the set of all electric vehicle data samples, k c and σ c are the parameters for adjusting the steepness and position of the curve, d c (x c ,X c ) is the electric vehicle data sample x c To X c The distance function of all electric vehicle data samples in is expressed as: Where ncs represents the preset electric vehicle data sample x c The number of features, represents the i-th feature of a single electric vehicle data sample; represents the i-th feature of a sample in the set of all electric vehicle data samples; is the importance weight for each dimension, expressed as: In the formula, is the standard deviation of the ith dimension in the electric vehicle data sample set, ∈ c is a small constant to avoid the denominator being zero; S22 selects and generates electric vehicle data samples. According to the membership value of each electric vehicle data sample, a minority of electric vehicle data samples that need to be expanded are dynamically selected. For electric vehicle data samples with higher membership, more similar electric vehicle data samples are selected for synthesis. The generation method is to interpolate between the selected electric vehicle data sample and its neighboring electric vehicle data samples, and define a function for generating the number of electric vehicle data samples. The function for generating the number of electric vehicle data samples determines the number of new electric vehicle data samples that need to be generated for each electric vehicle data sample based on the membership value. The calculation method is expressed as: In the formula, g c () is the function for generating the number of electric vehicle data samples; λ c It is the preset expansion factor, which determines the order of magnitude of the overall electric vehicle data sample generation; represents the gamma function; δ c (x c ) is the adjustment factor based on the sample density of electric vehicle data; New electric vehicle data samples are generated by interpolation method. For each selected electric vehicle data sample, a neighboring electric vehicle data sample is randomly selected, and then a new electric vehicle data sample is generated, which is expressed as: In the formula, represents the generated new electric vehicle data sample; γ c is a random number uniformly distributed from [0, 1], used to control the ratio of interpolation; n c represents a randomly selected sample of neighboring electric vehicle data; The adjustment factor based on the density of electric vehicle data samples makes the number of electric vehicle data samples generated inversely proportional to the density around the electric vehicle data samples, ensuring that fewer new electric vehicle data samples are generated in areas with denser electric vehicle data samples, thereby maintaining the balance of electric vehicle data sample distribution. The calculation method is expressed as: Where N c (x c ) is x c In X c The neighboring electric vehicle data sample set in , σ c is the width parameter of the Gaussian kernel; x c ′ For X c Any sample in the neighboring electric vehicle data sample set; ∥∥ is the L2 norm; S23 performs high-dimensional mapping adjustment, and adjusts the high-dimensional space mapping of the newly generated electric vehicle data samples to ensure that the electric vehicle data samples maintain the original distribution characteristics in the new feature space, and optimizes the local and global structures of the electric vehicle data samples. Specifically, the mapping function is used to adjust and refine the features, which is expressed as: In the formula, Θ c Represents the parameter set of the mapping function, y c is the new electric vehicle data sample after mapping; f c () is a mapping function, specifically a multi-layer neural network, used to learn the nonlinear mapping from the original feature space to the new feature space; S24 Integration into the training set: Integrate the expanded and adjusted new electric vehicle data samples into the original training electric vehicle data set to form a new electric vehicle data set to provide data support for subsequent feature extraction and model training; Finally, all the generated new electric vehicle data samples are integrated into the original training set, and the training electric vehicle dataset is updated, which is expressed as: In the formula, represents the expanded electric vehicle dataset; Y c is the set of all newly generated electric vehicle data samples, the updated training electric vehicle dataset That is: the expanded electric vehicle data training set.

4. The method for predicting the number of electric vehicles based on multi-source information fusion according to claim 1 is characterized in that: The 6-layer fully connected neural network algorithm includes: S31 generates an initial population according to the bionic algorithm initialization method in the initialization stage. Each individual represents a configuration of network weights. Specifically, the population size is N. p , initialize the weights and biases for the i-th individual, expressed as: Where W pi is the weight of the neural network corresponding to the i-th individual, b pi is the bias of the neural network corresponding to the i-th individual, Represents the weight matrix of the i-th individual in the initial state; represents the bias of the i-th individual in the initial state; σ 2 represents the initialization variance; Indicates that the mean is 0 and the variance is σ 2 Normal distribution of is a normal distribution; S32 For each individual in the population, use its corresponding neural network configuration to process the samples in the input expanded electric vehicle data training set, calculate the output of the model, and evaluate its performance according to a predetermined loss function. Specifically, for the i-th individual, use its weight and bias to calculate the loss on the training electric vehicle data set, expressed as: Where, L pi is the loss of the neural network corresponding to the i-th individual; mps represents the number of samples input in the current batch; l( ) represents the composite loss function, and fsig( ) represents the neural network model function; represents the electric vehicle data characteristics of the jth sample; represents the label of the jth sample; the composite loss function contains a regularization term, expressed as: Where MSE() is the mean square error function, λ ps is the regularization parameter, Reg(W pi ) is the regularization term, which is: Where W pi,k is the kth weight of the neural network corresponding to the i-th individual; S33 selects the best performing individuals from the current population according to their fitness and retains them as candidate solutions for the next generation. Excellent individuals have a higher probability of being selected based on their fitness. The calculation method of the probability of an individual being selected is expressed as: Where P select (i) represents the probability of the i-th individual being selected; γ pse is the parameter that controls the selection pressure; L pk is the loss of the neural network corresponding to the kth individual; S34 generates new individuals through crossover and mutation operations. The crossover operation allows two excellent individuals to exchange some genes to generate new offspring; the mutation operation randomly changes some genes in an individual to increase the diversity of the population. Specifically, the crossover operation randomly selects two individuals for gene exchange, which is expressed as: W′ pi =a pcs W p1 +(1-a pcs )W p2 b′ pi =a pcs b p1 +(1-a pcs )b p2 Where W′ pi is the weight of the neural network corresponding to the individual after the crossover operation, α pcs is the crossover rate, W p1 is the weight of the neural network corresponding to the first selected individual, W p2 is the weight of the neural network corresponding to the selected second individual, b′ pi is the bias of the neural network corresponding to the individual after the crossover operation, b p1 is the bias of the neural network corresponding to the first selected individual, b p2 is the bias of the neural network corresponding to the selected second individual; The mutation operation performs a small random perturbation on the newly generated individual weights, expressed as: Where W″ pi is the weight of the neural network corresponding to the individual after the mutation operation, b″ pi is the bias of the neural network corresponding to the individual after the mutation operation, τ 2 represents the variance of variation; S35 repeats the above steps until the preset stop iteration condition is met, which means that the model training is completed and the electric vehicle data training set after feature extraction is obtained.

5. The method for predicting the number of electric vehicles based on multi-source information fusion according to claim 1 is characterized in that: The fuzzy set optimized random forest classification algorithm includes: S41 inputs the electric vehicle data training set after feature extraction into the classifier to train the classifier model, initializes multiple decision trees in the random forest, and each tree will be trained independently in the future. Set the number of trees T u and the size of the feature subset used for each tree F u In order to integrate fuzzy logic into the classifier, the fuzzy membership function is defined as It measures the importance of features for classification decisions. The output membership is used to adjust subsequent feature selection and sample split point selection, expressed as: In the formula, represents the i-th feature of the feature vector input to the random forest, β u is a parameter that adjusts the sensitivity of fuzzy logic, is the average value of the feature in the training set; S42 uses fuzzy set theory to optimize data processing during the training of each decision tree. Fuzzy logic is used to select each feature and split point to deal with the inconsistency and ambiguity caused by multi-source data fusion, including the use of fuzzy membership functions to evaluate the importance of features and the contribution of data points to the classification results. The selection weight of the i-th feature is set to It is expressed as: In the formula, IG() is the information gain function, expressed as: In the formula, H(S u ) is the set S u The entropy of By feature After the segmentation, L and R represent the left and right subsets; v is an element in the set consisting of the left and right subsets; The information gain is calculated by comparing the entropy of the parent node and the entropy of the child node after the split, expressed as: In the formula, p j is the proportion of samples of the jth class in the node; For the features and split points that maximize the information gain of the node split, the entropy of the child nodes after splitting is: In the formula, p v is the proportion of positive classes in child node v; Each S43 tree is trained using a randomly selected data subset and feature subset. During the tree construction process, the fuzzy set optimization standard is used to select the best split point. The split of each node is aimed at maximizing the purity improvement of subsequent nodes. The optimized feature selection weights are used when splitting nodes: In the formula, Split represents node splitting; tu is the splitting threshold; argmax{} represents the index function that returns the maximum value: In the process of tree construction, nodes are split iteratively until the stopping condition is met; A gradient optimization-based method is used to determine the optimal splitting threshold to improve the calculation accuracy of information gain and segmentation efficiency, which is expressed as: In the formula, ΔIG represents the change in information gain, which is solved by estimating the gradient of the information gain function; A numerical method is used to find the optimal value of tu, expressed as: In the formula, It is the first-order derivative of information gain with respect to tu, indicating the rate of change of information gain at tu; is the second-order derivative of information gain with respect to tu, indicating the change in the rate of change; tu new represents the updated split threshold; tu old represents the split threshold before updating; S44 After all decision trees are trained, they will be integrated together to form the final random forest model. The classification decision is made by voting or averaging the prediction results of all individual trees. The voting weight of each tree for a sample classification is related to its performance on the training set, expressed as: In the formula, C(x) is the voting result of sample x, p t (x) is the classification output of the t-th tree for sample x; mode{} is the voting strategy function of ensemble learning; The classification category of electric vehicle ownership is obtained according to the classification decision.

6. An electric vehicle ownership prediction system based on multi-source information fusion, characterized by: include: Training data capture module, training data expansion module, feature extraction module, classifier module and inventory prediction module; The training data capture module is used to capture training data samples from the database using automated data capture tools and API interfaces, and classify and annotate them to obtain the original electric vehicle data training set; The training data expansion module is used to expand the original electric vehicle data training set based on the SMOTE algorithm driven by dynamic fuzzy membership, and train the data expansion model to obtain the expanded electric vehicle data training set; The feature extraction module is used to construct a 6-layer fully connected neural network, extract features from the expanded electric vehicle data training set, and train the feature extraction model to obtain the electric vehicle data training set after feature extraction; The classifier module is used to classify the electric vehicle data training set after feature extraction using a random forest classification algorithm based on fuzzy set optimization, and train the classifier model to obtain the classification category of the number of electric vehicles in use; Ownership prediction module: used to input the new electric vehicle data set into the trained feature extraction model to obtain the electric vehicle data set after feature extraction, and input the electric vehicle data set after feature extraction into the trained classifier model to obtain the classification result, which is the electric vehicle ownership prediction result.

7. The electric vehicle ownership prediction system based on multi-source information fusion according to claim 6 is characterized by: In the training data capture module, training data collection is performed based on multi-source heterogeneous data indicators, and the training data includes: traffic flow data, electric vehicle sales and registration data, economic development index, policy support and charging infrastructure distribution; The classification labels include: high growth, medium growth, and low growth.

8. The method for predicting the number of electric vehicles based on multi-source information fusion according to claim 6 is characterized by: In the training data expansion module, S21 takes the original electric vehicle data training set as the set of all electric vehicle data samples X c , for X c For each electric vehicle data sample in , calculate its fuzzy membership to the minority class. The minority class refers to the class with fewer samples in the high-growth, medium-growth, and low-growth data. The calculation of the membership takes into account multiple attribute values ​​of the electric vehicle data sample and its distance relative to other electric vehicle data samples. Through fuzzy logic, each electric vehicle data sample is assigned a membership value between 0 and 1. The membership value represents the possibility that the electric vehicle data sample belongs to the minority class, which is expressed as: In the formula, x c represents a single electric vehicle data sample; m c () represents the fuzzy membership function; X c represents the set of all electric vehicle data samples, k c and σ c are the parameters for adjusting the steepness and position of the curve, d c (x c ,X c ) is the electric vehicle data sample x c To X c The distance function of all electric vehicle data samples in is expressed as: Where ncs represents the preset electric vehicle data sample x c The number of features, represents the i-th feature of a single electric vehicle data sample; represents the i-th feature of a sample in the set of all electric vehicle data samples; is the importance weight for each dimension, expressed as: In the formula, is the standard deviation of the ith dimension in the electric vehicle data sample set, ∈ c is a small constant to avoid the denominator being zero; S22 selects and generates electric vehicle data samples. According to the membership value of each electric vehicle data sample, dynamically selects a minority of electric vehicle data samples that need to be expanded. For electric vehicle data samples with higher membership, more similar electric vehicle data samples are selected for synthesis. The generation method is to interpolate between the selected electric vehicle data sample and its neighboring electric vehicle data samples, and define a function for generating the number of electric vehicle data samples. The function for generating the number of electric vehicle data samples determines the number of new electric vehicle data samples that need to be generated for each electric vehicle data sample based on the membership. The calculation method is expressed as: In the formula, g c () is the function for generating the number of electric vehicle data samples; λ c It is the preset expansion factor, which determines the order of magnitude of the overall electric vehicle data sample generation; represents the gamma function; δ c (x c ) is the adjustment factor based on the sample density of electric vehicle data; New electric vehicle data samples are generated by interpolation method. For each selected electric vehicle data sample, a neighboring electric vehicle data sample is randomly selected, and then a new electric vehicle data sample is generated, which is expressed as: In the formula, represents the generated new electric vehicle data sample; γ c is a random number uniformly distributed from [0, 1], used to control the ratio of interpolation; n c represents a randomly selected sample of neighboring electric vehicle data; The adjustment factor based on the density of electric vehicle data samples makes the number of electric vehicle data samples generated inversely proportional to the density around the electric vehicle data samples, ensuring that fewer new electric vehicle data samples are generated in areas with denser electric vehicle data samples, thereby maintaining the balance of electric vehicle data sample distribution. The calculation method is expressed as: Where N c (x c ) is x c In X c The neighboring electric vehicle data sample set in , σ c is the width parameter of the Gaussian kernel; x c ′ For X c Any sample in the neighboring electric vehicle data sample set; ∥∥ is the L2 norm; S23 performs high-dimensional mapping adjustment, and adjusts the high-dimensional space mapping of the newly generated electric vehicle data samples to ensure that the electric vehicle data samples maintain the original distribution characteristics in the new feature space, and optimizes the local and global structures of the electric vehicle data samples. Specifically, the mapping function is used to adjust and refine the features, which is expressed as: In the formula, Θ c Represents the parameter set of the mapping function, y c is the new electric vehicle data sample after mapping; f c () is a mapping function, specifically a multi-layer neural network, used to learn the nonlinear mapping from the original feature space to the new feature space; S24 Integration into the training set: Integrate the expanded and adjusted new electric vehicle data samples into the original training electric vehicle data set to form a new electric vehicle data set to provide data support for subsequent feature extraction and model training; Finally, all the generated new electric vehicle data samples are integrated into the original training set, and the training electric vehicle dataset is updated, which is expressed as: In the formula, represents the expanded electric vehicle dataset; Y c is the set of all newly generated electric vehicle data samples, the updated training electric vehicle dataset That is: the expanded electric vehicle data training set.

9. The electric vehicle ownership prediction system based on multi-source information fusion according to claim 6 is characterized by: In the feature extraction module: S31 generates an initial population according to the bionic algorithm initialization method. Each individual represents a configuration of network weights. The population size is N. p , initialize the weights and biases for the i-th individual, expressed as: Where W pi is the weight of the neural network corresponding to the i-th individual, b pi is the bias of the neural network corresponding to the i-th individual, Represents the weight matrix of the i-th individual in the initial state; represents the bias of the i-th individual in the initial state; σ 2 represents the initialization variance; Indicates that the mean is 0 and the variance is σ 2 Normal distribution of is a normal distribution; S32 For each individual in the population, use its corresponding neural network configuration to process the samples in the input expanded electric vehicle data training set, calculate the output of the model, and evaluate its performance according to a predetermined loss function. Specifically, for the i-th individual, use its weight and bias to calculate the loss on the training electric vehicle data set, expressed as: Where, L pi is the loss of the neural network corresponding to the i-th individual; mps represents the number of samples input in the current batch; l( ) represents the composite loss function, and fsig( ) represents the neural network model function; represents the electric vehicle data characteristics of the jth sample; represents the label of the jth sample; the composite loss function contains a regularization term, expressed as: Where MSE() is the mean square error function, λ ps is the regularization parameter, Reg(W pi ) is the regularization term, which is: Where W pi,k is the kth weight of the neural network corresponding to the i-th individual; S33 selects the best performing individuals from the current population according to their fitness and retains them as candidate solutions for the next generation. Excellent individuals have a higher probability of being selected based on their fitness. The calculation method of the probability of an individual being selected is expressed as: Where P select (i) represents the probability of the i-th individual being selected; γ pse is the parameter that controls the selection pressure; L pk is the loss of the neural network corresponding to the kth individual; S34 generates new individuals through crossover and mutation operations. The crossover operation allows two excellent individuals to exchange some genes to generate new offspring; the mutation operation randomly changes some genes in an individual to increase the diversity of the population. Specifically, the crossover operation randomly selects two individuals for gene exchange, which is expressed as: W′ pi =a pcs W p1 +(1-a pcs )W p2 b′ pi =a pcs b p1 +(1-a pcs )b p2 Where W′ pi is the weight of the neural network corresponding to the individual after the crossover operation, α pcs is the crossover rate, W p1 is the weight of the neural network corresponding to the first selected individual, W p2 is the weight of the neural network corresponding to the selected second individual, b′ pi is the bias of the neural network corresponding to the individual after the crossover operation, b p1 is the bias of the neural network corresponding to the first selected individual, b p2 is the bias of the neural network corresponding to the selected second individual; The mutation operation performs a small random perturbation on the newly generated individual weights, expressed as: Where W″ pi is the weight of the neural network corresponding to the individual after the mutation operation, b″ pi is the bias of the neural network corresponding to the individual after the mutation operation, τ 2 represents the variance of variation; S35 repeats the above steps until the preset stop iteration condition is met, which means that the model training is completed and the electric vehicle data training set after feature extraction is obtained.

10. The electric vehicle ownership prediction system based on multi-source information fusion according to claim 6, characterized in that: The classifier module comprises: S41 inputs the electric vehicle data training set after feature extraction into the classifier to train the classifier model, initializes multiple decision trees in the random forest, and each tree will be trained independently in the future. Set the number of trees T u and the size of the feature subset used for each tree F u In order to integrate fuzzy logic into the classifier, the fuzzy membership function is defined as It measures the importance of features for classification decisions. The output membership is used to adjust subsequent feature selection and sample split point selection, expressed as: In the formula, represents the i-th feature of the feature vector input to the random forest, β u is a parameter that adjusts the sensitivity of fuzzy logic, is the average value of the feature in the training set; S42 uses fuzzy set theory to optimize data processing during the training of each decision tree. Fuzzy logic is used to select each feature and split point to deal with the inconsistency and ambiguity caused by multi-source data fusion, including the use of fuzzy membership functions to evaluate the importance of features and the contribution of data points to the classification results. The selection weight of the i-th feature is set to It is expressed as: In the formula, IG() is the information gain function, expressed as: In the formula, H(S u ) is the set S u The entropy of By feature After the segmentation, L and R represent the left and right subsets; v is an element in the set consisting of the left and right subsets; The information gain is calculated by comparing the entropy of the parent node and the entropy of the child node after the split, expressed as: In the formula, p j is the proportion of samples of the jth class in the node; For the features and split points that maximize the information gain of the node split, the entropy of the child nodes after splitting is: In the formula, p v is the proportion of positive classes in child node v; Each S43 tree is trained using a randomly selected data subset and feature subset. During the tree construction process, the fuzzy set optimization standard is used to select the best split point. The split of each node is aimed at maximizing the purity improvement of subsequent nodes. The optimized feature selection weights are used when splitting nodes: In the formula, Split represents node splitting; tu is the splitting threshold; argmax{} represents the index function that returns the maximum value: In the process of tree construction, the nodes are split iteratively until the stopping condition is met. A gradient optimization-based method is used to determine the optimal splitting threshold to improve the calculation accuracy of information gain and segmentation efficiency, which is expressed as: In the formula, ΔIG represents the change in information gain, which is solved by estimating the gradient of the information gain function; A numerical method is used to find the optimal value of tu, expressed as: In the formula, It is the first-order derivative of information gain with respect to tu, indicating the rate of change of information gain at tu; is the second-order derivative of information gain with respect to tu, indicating the change in the rate of change; tu new represents the updated split threshold; tu old represents the split threshold before updating; S44 After all decision trees are trained, they will be integrated together to form the final random forest model. The classification decision is made by voting or averaging the prediction results of all individual trees. The voting weight of each tree for a sample classification is related to its performance on the training set, expressed as: In the formula, C(x) is the voting result of sample x, p t (x) is the classification output of the tth tree for sample x; mode{} is the voting strategy function of ensemble learning.

Images