Electric vehicle charging data generation method and device based on attention and conditional probability distribution, and storage medium

Abstract

The invention relates to an attention and conditional probability distribution-based electric vehicle charging data generation method and device, and a storage medium, and the method comprises the following steps: constructing a feature space of an electric vehicle charging behavior, and respectively aiming at a continuous value feature, a time correlation feature and a discrete value feature; performing feature-level deep neural network enhanced conditional probability modeling to obtain a conditional probability neural network model based on feature attention and residual blocks; acquiring electric vehicle charging behavior training data, and training the conditional probability neural network model by adopting a learning rate which is reduced along with the increase of a training round; and based on the trained conditional probability neural network model, Gibbs sampling is carried out under variable constraints, and electric vehicle charging behavior data conforming to physical constraints are screened. Compared with the prior art, the method has the advantages of integrating a multi-type probability distribution model and a degree neural network to dynamically predict probability distribution parameters and the like.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method, device, and storage medium for generating electric vehicle charging data based on attention and conditional probability distribution. Background Art

[0002] Currently, traditional simulation models based on vehicle travel surveys and data-driven statistical models are the mainstream research and application in the generation and analysis of electric vehicle charging data. These models provide support for understanding and predicting electric vehicle charging behavior to a certain extent. For example, simulation models based on vehicle travel surveys can simulate electric vehicle charging needs in different scenarios by constructing travel demand models, energy consumption models, and charging selection models. Data-driven statistical models, on the other hand, model and predict charging events by learning the probability distribution of existing datasets.

[0003] However, with the rapid growth in the number of electric vehicles and the continuous expansion of charging infrastructure, these traditional models are gradually revealing some limitations.

[0004] Chinese patent publication CN114580789B provides a multi-type electric vehicle load forecasting method and system based on charging behavior. This method analyzes the probabilistic characteristics of an electric vehicle's daily mileage and initial charging time to establish a calculation model for the electric vehicle's initial state of charge and initial charging time, thereby predicting the electric vehicle's daily load. This method is based on a data-driven statistical model that learns the probability distribution of existing data. However, these models have relatively limited parameter dimensions and often focus only on basic parameters such as charging duration, start time, and charging energy. They struggle to fully capture the complexities of electric vehicle charging behavior, such as charging habits among different user groups, differences in battery capacity, and the choice of charging method (fast or slow charging). Traditional models are significantly inadequate in processing complex, high-dimensional data and capturing the complex dependencies within the data.

[0005] Therefore, in order to overcome the limitations of existing technologies and meet the needs of data analysis and prediction of large-scale electric vehicle charging systems, it is urgent to develop an electric vehicle charging data generation method based on a neural network generative model to achieve more comprehensive and accurate modeling and prediction of charging behavior, and provide stronger support for charging infrastructure planning and power system scheduling. Summary of the Invention

[0006] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide an electric vehicle charging data generation method, device, and storage medium based on attention and conditional probability distribution, so as to solve or partially solve the problems that traditional models are insufficient in processing complex high-dimensional data and capturing complex dependencies in the data when generating electric vehicle charging data.

[0007] The purpose of the present invention can be achieved by the following technical solutions: One aspect of the present invention provides a method for generating electric vehicle charging data based on attention and conditional probability distribution, comprising the following steps: A feature space for electric vehicle charging behavior is constructed. Feature-level deep neural network-enhanced conditional probability modeling is performed for continuous-valued features, time-dependent features, and discrete-valued features, resulting in a conditional probability neural network model based on feature attention and residual blocks. Obtaining electric vehicle charging behavior training data, and training the conditional probabilistic neural network model using a learning rate that decreases with increasing training rounds; Based on the trained conditional probability neural network model, Gibbs sampling is performed under variable constraints to screen electric vehicle charging behavior data that meets physical constraints.

[0008] As a preferred technical solution, the conditional probability modeling of the deep neural network enhancement of the continuous value feature is:

[0009]

[0010]

[0011]

[0012]

[0013]

[0014] in, For the condition On the premise The probability of For the modeling features, For all features except the modeled features, 、 is the weight, 、 is the bias, is the conditional probability density of the feature, is the Gamma function, 、 To limit the parameters to the preset range, is the hidden layer representation of the neural network after processing by the attention mechanism and residual connection.

[0015] As a preferred technical solution, the conditional probability modeling of the deep neural network enhancement of the time-related features is:

[0016]

[0017]

[0018]

[0019]

[0020]

[0021]

[0022] in, For the condition On the premise The probability of For the modeling features, For all features except the modeled features, 、 、 、 is the weight, 、 、 、 is the bias, is the number of mixed ingredients, For the The weight of the components, represents a normal distribution, and Respectively The mean and standard deviation of each component, 、 is the preset minimum and maximum value, is a restricted interval function, is the feature set input to the conditional encoder, including all features except the current predicted feature.

[0023] As a preferred technical solution, the conditional probability modeling of the deep neural network enhancement of the discrete value feature is:

[0024]

[0025]

[0026]

[0027] in, is the feature gating result, is the Sigmoid activation function, 、 、 is the weight, 、 、 is the bias, is the feature set input to the conditional encoder, including all features except the currently predicted features, represents element-wise multiplication, The discrete value feature is The probability of Target feature y The value is category c The corresponding logit value, is the number of categories.

[0028] As a preferred technical solution, the process of training the conditional probability neural network model includes the following steps: Perform feature extraction, normalization, and encoding on electric vehicle charging behavior training data; Initialize the parameters of the conditional probability model corresponding to each feature; In each training round, residual connections are used to improve gradient flow, attention mechanisms are used to extract features, the gradient of each variable is calculated, and the parameters of the corresponding conditional probability model are updated using gradient descent with a learning rate that decreases with the number of training rounds. After training is completed, the parameters of the conditional probability model are returned.

[0029] As a preferred technical solution, the learning rate is:

[0030] in, For the t The learning rate of the round, is the initial learning rate, is the total number of rounds, is the termination factor.

[0031] As a preferred technical solution, the process of performing Gibbs sampling under variable constraints based on the trained conditional probability neural network model to screen electric vehicle charging behavior data that meets physical constraints includes the following steps: Initialize the batch sample matrix and generate initial values; Update the sample value through Gibbs sampling iteration; Set variable-level constraints; When the number of iterations reaches the threshold, a stable distribution of samples is obtained; Filtering electric vehicle charging behavior data that complies with physical constraints; The filtered electric vehicle charging behavior data is saved as structured data for electric vehicle charging behavior analysis and prediction.

[0032] As a preferred technical solution, the continuous value features include the initial state of charge and the final state of charge, the time-related features include the start time of charging and the charging duration, and the discrete value features include the battery capacity and the user type.

[0033] Another aspect of the present invention provides an electronic device comprising: one or more processors and a memory, wherein the memory stores one or more programs, and the one or more programs include instructions for executing the aforementioned electric vehicle charging data generation method based on attention and conditional probability distribution.

[0034] Another aspect of the present invention provides a computer-readable storage medium comprising one or more programs for execution by one or more processors of an electronic device, wherein the one or more programs include instructions for executing the aforementioned method for generating electric vehicle charging data based on attention and conditional probability distribution.

[0035] Compared with the prior art, the present invention has at least one of the following beneficial effects: (1) Integrated multi-type probability distribution model: The present invention addresses the problem that the traditional model has limited parameter dimensions, making it difficult to accurately model different types of charging behavior features such as continuous bounded, continuous multi-peak, and discrete, and unable to fully capture the complexity of charging behavior. The present invention constructs a feature space for electric vehicle charging behavior, and performs feature-level deep neural network enhanced conditional probability modeling for continuous value features, time-related features, and discrete value features, respectively. A conditional probability neural network model based on feature attention and residual blocks is obtained. For electric vehicle charging behavior features with different characteristics, an adapted probability distribution model is selected and integrated into the same framework, which significantly improves the modeling accuracy and expression ability. The distribution of the generated data in each feature dimension is highly consistent with the original data, and the range characteristics of continuous bounded values, the distribution form of continuous multi-peak features, and the category ratio of discrete features can be accurately restored.

[0036] (2) Dynamic prediction of probability distribution parameters using deep neural networks: This invention addresses the problem that traditional models use fixed parameter distributions, are unable to learn nonlinear dependencies between features, and are insufficient in processing high-dimensional complex data and capturing complex dependencies between features. By combining deep neural networks with conditional probability distributions, the neural network is supplemented by residual blocks and attention mechanisms to dynamically calculate the parameters of each probability distribution, enabling the model to learn nonlinear dependencies between features, improving its expressive power and flexibility. The generated data can effectively retain the complex correlations between features in charging behavior, providing more accurate support for comprehensive modeling and prediction of charging behavior.

[0037] (3) Combining feature attention mechanism and residual block: In order to solve the problem of gradient vanishing and exploding in deep neural network training, and the inability to adaptively distinguish the importance of features, resulting in poor model stability and insufficient capture of key features, the present invention constructs a feature attention mechanism module and a residual block module. The attention mechanism enhances the ability to focus on important features, and the residual block improves the gradient flow of the deep network, thereby enhancing the stability and expression ability of model training. In particular, it performs well in processing multi-modal distribution features and maintaining the dependency relationship between features, thereby ensuring the authenticity of the generated data. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 Flowchart of a method for generating electric vehicle charging data based on attention and conditional probability distribution in an embodiment; Figure 2 Schematic diagram of the model training process in the embodiment; Figure 3 A schematic diagram of the data generation process in the embodiment; Figure 4 Schematic diagram showing the comparison of the distribution of the electric vehicle charging behavior data and the original data on the continuous features in the embodiment; Figure 5 Schematic diagram showing the comparison of the distribution of electric vehicle charging behavior data and original data on discrete features in an embodiment; Figure 6 Schematic diagram of an electronic device in an embodiment. DETAILED DESCRIPTION

[0039] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0040] Example 1 In response to the problems of the aforementioned prior art, this embodiment provides a method for generating electric vehicle charging data based on attention and conditional probability distribution. By combining multiple conditional probability distributions and deep neural network technology, it simulates and generates electric vehicle charging behavior data. Figure 1 The overall architecture of the method mainly includes three core parts: First, the conditional probability distribution is directly modeled using a neural network , and then through the conditional probability model and chain rule , implicitly expresses the joint probability distribution of electric vehicle (EV) charging data, and finally converges to a state that conforms to the joint probability distribution through Gibbs sampling and multiple iterations to generate EV charging events. The method specifically includes the following steps: Step S1: Neural network directly models conditional probability distribution .

[0041] Specifically, step S1 includes: Step S101 : construct a feature set to define the feature space of electric vehicle charging behavior, including the initial state of charge (start_soc), the end state of charge (end_soc), the start time of charging (start_hour), the duration of charging (duration), the battery capacity (battery_capacity) and the user type (veh_label).

[0042] Step S102: constructing a deep neural network enhanced conditional probability model for each feature, specifically including: (1) Construct a feature attention mechanism module, which contains the following structure:

[0043] in, is the input feature vector, representing a sample point in the feature space, Represents the Sigmoid activation function. The second term on the right side of the above formula represents the residual connection, which adds the original input directly to the output of the attention mechanism to improve information flow and gradient propagation.

[0044] Use linear layers and Sigmoid activation functions to calculate feature weights, apply weights to input features, and add residual connections to achieve adaptive feature enhancement.

[0045] The mathematical expression of the feature attention mechanism is:

[0046] in, represents the Sigmoid function, and are the weight matrix and bias term respectively, and ⊙ represents element-wise multiplication.

[0047] (2) Construct a residual block module, which contains the following structure:

[0048] A sequence of two linear layers and two normalization layers uses residual connections to improve the gradient flow performance of deep networks.

[0049] The mathematical expression of the residual block is:

[0050]

[0051]

[0052]

[0053]

[0054]

[0055] In the above formula, 、 is the weight matrix, 、 is the bias term, z Represents the saved original input, used to implement the residual connection, x Represents the input feature vector, represents the feature representation of the current processing in the residual block, Represents the layer normalization operation, which is used to stabilize deep network training. The input of each layer is normalized by subtracting the mean and dividing by the standard deviation. ReLU( ) represents the rectified linear unit activation function, which is defined as ReLU( x ) = max(0, x ), used to introduce nonlinearity.

[0056] (3) For continuous value features (initial state of charge, final state of charge), a Beta distribution model is constructed:

[0057] In the above formula, For the condition On the premise The probability of is the feature currently being modeled, For all features except the one currently being modeled, use them as conditions. and Calculated by deep neural network:

[0058]

[0059] In the above formula, 、 The model consists of two neural network functions that compute distribution parameters from conditional features. The model's input layer receives all features except the target feature. The hidden layer applies BatchNorm normalization and ReLU activation, embeds a feature attention mechanism, and embeds residual blocks to improve network expressiveness. The output layer predicts the α and β parameters of the Beta distribution, applies Softplus to ensure that the parameters are positive, and implements parameter validity checks to prevent numerical instability.

[0060] The probability density function of the Beta distribution is:

[0061] in, is the Beta function, defined as:

[0062] is the Gamma function. and Output through the network and softplus transformation:

[0063]

[0064] The softplus function is defined as:

[0065] To ensure numerical stability, the parameters are restricted:

[0066]

[0067] (4) For time-related features (charging start time, charging duration), a Gaussian mixture model (GMM) is constructed:

[0068] In the above formula, For the condition On the premise The probability of is the feature currently being modeled, For all features other than the currently modeled feature, as conditions, 、 and Calculated by deep neural network:

[0069]

[0070]

[0071] In the above formula, 、 、 It is a neural network function used to calculate weights, mean, and standard deviation from conditional features. The basic network structure of the model is similar to the Beta distribution network. A conditional encoder is added to allow other features to adjust the distribution. The output layer predicts the mean, standard deviation, and weight parameters of the mixed Gaussian respectively. Log_softmax is applied to improve numerical stability and realize a backup distribution mechanism in abnormal situations.

[0072] The probability density function of the Gaussian mixture model is:

[0073] in is the number of mixed ingredients, is the first i The weights of the Gaussian components, represents a normal distribution, and Respectively i The mean and standard deviation of each component, Indicates the mean , the variance is Normal distribution.

[0074] The network outputs these parameters which are calculated as:

[0075]

[0076]

[0077] in, The function is defined as:

[0078] Conditional information fusion uses:

[0079] in, , The feature set input to the conditional encoder includes all features except the currently predicted features. For example, when predicting a certain dimension feature, Contains the eigenvalues ​​of the other five dimensions, Conditional encoding represents the calculation process of conditional information, which aims to convert other features into conditional information that affects the current predicted features, thereby adjusting the distribution of the current predicted features.

[0080] (5) For discrete features (battery capacity, user type), build a classification distribution model:

[0081] In the above formula, is a probability vector of classification distribution, where each element represents the probability of the corresponding category, which is calculated by a deep neural network:

[0082] In the above formula, This is a neural network function used to calculate category probabilities from conditional features. In addition to the basic network structure, the classification distribution model adds a feature gating mechanism and applies a Dropout regularization layer to prevent overfitting. The output layer uses the Softmax activation function to generate the category probability distribution.

[0083] The mathematical expression of the feature gating mechanism is:

[0084]

[0085] The class probability is calculated as:

[0086]

[0087] In the above formula is the number of categories, Represents element-wise multiplication, i.e., Hadamard product, which represents the modulation of the feature by the gating mechanism. The weight vector generated by the feature gating mechanism is used to selectively focus on features. is the middle layer representation of the neural network, the features after gate processing, Represents the original category score without softmax.

[0088] Step S2, through the conditional probability model and conditional probability chain rule , implicitly represents the complete joint probability distribution of Electric Vehicle (EV) data.

[0089] This step uses a conditional probability neural network model for training and optimizes the conditional probability distribution parameters so that the model can capture the interdependencies between features. The training goal is to maximize the conditional log-likelihood:

[0090] For details, see Figure 2The training process includes data preprocessing, model initialization, application of feature attention and residual connection mechanism, and gradient descent update. In each training round, the parameters of each conditional probability model are updated in turn, and the trained model parameters are finally obtained. Step S2 includes: Step S201: data preprocessing.

[0091] Read the raw charging data and filter the six features: initial SOC, final SOC, start time, duration, battery capacity, and user label. Reset the start time to a standard reference time, such as 6:00 AM. Map the battery capacity to a category label. Create a data loader and set the batch size to 512.

[0092] Step S202: Initialize Beta model, GMM model, and Discreate model parameters .

[0093] Six conditional probability models are initialized, corresponding to the six feature dimensions. Two Beta distribution models model the initial SOC and the final SOC respectively. Two GMM models model the start time and duration respectively. Two discrete distribution models model the battery capacity and the user type respectively.

[0094] Step S203: model training.

[0095] During training, the Adam optimizer was used with an initial learning rate of 0.005. The LinearLR learning rate scheduler was applied, and the learning rate was linearly reduced from the initial value to half during training. For each batch of data, forward propagation was performed on the six models to calculate the loss, perform gradient backpropagation and parameter updates, record the training loss of each model, and implement an exception handling mechanism to ensure the robustness of the training process.

[0096] Loss function calculation: Beta distribution model:

[0097] in is the probability density of the conditional Beta distribution.

[0098] GMM model:

[0099] Discrete distribution models:

[0100] Learning rate scheduling strategy:

[0101] in, is the learning rate of the tth round, is the initial learning rate, is the total number of rounds, represents the computational expectation, is the termination factor, which is 0.5 in this embodiment.

[0102] See also Figure 2 The training process specifically includes the following steps: (1) For each training round e From 1 to E Execute (2)-(5) and the training round reaches E Then execute (6).

[0103] (2) Apply attention mechanism to extract relevant features ,in For the i The hidden representation of each feature before attention is applied, is the hidden representation after applying the attention mechanism.

[0104] (3) Applying residual connections to improve gradient flow ,in It is the hidden representation after applying the residual connection, that is, the intermediate feature vector obtained after the original input features are processed by the neural network layer, which captures the abstract features and deeper patterns of the data.

[0105] (4) Calculation variables i Gradient .

[0106] (5) Update variables using gradient descent i The weight of , return to (1).

[0107] (6) End all training and return model parameters .

[0108] In step S3, Gibbs sampling is used to converge to a state that conforms to the joint probability distribution through multiple iterations, and an electric vehicle charging event is generated.

[0109] See also Figure 3 The data generation process is based on Gibbs sampling technology, including initialization parameter setting, batch sample initialization, multi-step iterative sampling, and sample post-processing. Iterative sampling is performed through the conditional probability model to generate electric vehicle charging behavior data that meets physical constraints. Step S3 specifically includes: Step S301, initialize sample size N , sampling steps S , batch size B Probabilistic Model and its parameters

[0110] Load the six trained models, initialize the batch sample matrix, and randomly generate initial values:

[0111] Where d is the feature dimension, and in this method d=6.

[0112] Step S302: Initialize an empty sample set Store the sample.

[0113] Initialize an empty sample set .

[0114] Step S303: iterative calculation.

[0115] The iterative process specifically includes: (1) For each batch b From 1 to N / B Execute (2)-(3), and execute the batch to N / B Then it ends.

[0116] (2) Initialize batch samples with random values .

[0117] (3) For each step s From 1 to S Execute (4), and execute the steps to reach S Then execute (8).

[0118] (4) For each variable i From 1 to m Execute (5), the execution variable reaches m back, , execute step (3).

[0119] (5) Calculating conditional distribution

[0120] (6) Sampling new values ​​from the conditional distribution

[0121] (7) Apply constraints based on variable types, , return to step (4).

[0122] (8) The final sample Add to dataset , , execute step (1).

[0123] In step (7), constraints are added to specific features to ensure the physical rationality of the generated data: SOC values ​​are limited to the range [0,1]; The start time is limited to the range [0,1], corresponding to 0-24 hours; The charging duration is limited to a positive value.

[0124] Step S304: post-process the generated samples.

[0125] Post-process the generated samples to filter valid charging data, including the following constraints: The charge level is a positive value: Final SOC > Initial SOC.

[0126] Charging duration is not zero: Duration>0.

[0127] Reasonable charging power: (Final SOC - Initial SOC) * Battery capacity / Duration / 100 < 120kW, where Final SOC and Initial SOC are the states of charge before and after charging, respectively, Duration is the charging duration, and Battery capacity is the battery capacity.

[0128] Step S305: Return the processed data set .

[0129] The filtered effective charging behavior data is saved as structured data for electric vehicle charging behavior analysis and prediction.

[0130] To verify the effectiveness of this method, the original charging data in step S201 is compared with the data set in step S305. For comparison of the distribution of each feature of Figure 4 Comparison of the distribution of continuous features (initial SOC, final SOC, start time, and charging duration) between the electric vehicle charging behavior data generated by the method (slash) and the original data (blank). It can be seen that the generated data successfully retains the distribution characteristics of the original data, especially in terms of multi-peak distribution features (such as the start time of charging). Figure 5 Comparison of the distribution of discrete features (battery capacity and user label) between the generated electric vehicle charging behavior data (slash) and the original data (blank). The generated data accurately restores the distribution ratios of different battery capacity types and user types.

[0131] according to Figure 4 and Figure 5The generated electric vehicle charging behavior data (Generated data) based on the initial SOC (Initial SOC), final SOC (Final SOC), start time (Start time), duration (Duration), battery capacity (Battery capacity), and user label (Userlabel) are highly consistent with the original data (Raw data) in terms of statistical distribution, demonstrating that this method can effectively capture and restore the complex characteristics of electric vehicle charging behavior. Compared with traditional methods, this method significantly improves the expressiveness and stability of the model by introducing an attention mechanism and residual connections. It excels in handling multimodal distribution features and maintaining inter-feature dependencies.

[0132] Example 2 Based on Example 1, this embodiment provides an electronic device, including: one or more processors and a memory, wherein the memory stores one or more programs, and the one or more programs include instructions for executing the electric vehicle charging data generation method based on attention and conditional probability distribution as described in Example 1.

[0133] like Figure 6 As mentioned above, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory and a non-volatile memory, and may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 Of course, in addition to software implementation, the present invention does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0134] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0135] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory media such as modulated data signals and carrier waves.

[0136] This method integrates different types of probability distribution models (Beta distribution, Gaussian mixture model, and categorical distribution) into a single framework. It selects the most appropriate probability model for different characteristics of electric vehicle charging behavior (continuous bounded value, continuous unbounded value, and discrete value), improving modeling accuracy and expressiveness. It also combines deep neural networks with conditional probability distribution, using the neural network to dynamically predict probability distribution parameters, rather than the fixed parameter distribution used in traditional methods. This design enables the model to learn nonlinear dependencies between features, significantly improving its expressiveness and flexibility.

[0137] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A method for generating electric vehicle charging data based on attention and conditional probability distribution, characterized in that: The steps include: A feature space for electric vehicle charging behavior is constructed. Feature-level deep neural network-enhanced conditional probability modeling is performed for continuous-valued features, time-dependent features, and discrete-valued features, resulting in a conditional probability neural network model based on feature attention and residual blocks. Obtaining electric vehicle charging behavior training data, and training the conditional probabilistic neural network model using a learning rate that decreases with increasing training rounds; Based on the trained conditional probability neural network model, Gibbs sampling is performed under variable constraints to screen electric vehicle charging behavior data that meets physical constraints.

2. The method for generating electric vehicle charging data based on attention and conditional probability distribution according to claim 1, characterized in that: The conditional probability modeling of the deep neural network enhancement of continuous valued features is as follows: in, For the condition On the premise The probability of For the modeling features, For all features except the modeled features, 、 is the weight, 、 is the bias, is the conditional probability density of the feature, is the Gamma function, 、 To limit the parameters to the preset range, is the hidden layer representation of the neural network after processing by the attention mechanism and residual connection.

3. The method for generating electric vehicle charging data based on attention and conditional probability distribution according to claim 1, characterized in that: The conditional probability modeling of the deep neural network enhancement of the time-related features is: in, For the condition On the premise The probability of For the modeling features, For all features except the modeled features, 、 、 、 is the weight, 、 、 、 is the bias, is the number of mixed ingredients, For the The weight of the components, represents a normal distribution, and Respectively The mean and standard deviation of each component, 、 is the preset minimum and maximum value, is a restricted interval function, is the feature set input to the conditional encoder, including all features except the currently predicted features.

4. The method for generating electric vehicle charging data based on attention and conditional probability distribution according to claim 1, characterized in that: The conditional probability modeling of the deep neural network enhancement of discrete value features is as follows: in, is the feature gating result, is the Sigmoid activation function, 、 、 is the weight, 、 、 is the bias, is the feature set input to the conditional encoder, including all features except the currently predicted features, represents element-wise multiplication, The discrete value feature is The probability of Target feature y The value is category c The corresponding logit value, is the number of categories.

5. The method for generating electric vehicle charging data based on attention and conditional probability distribution according to claim 1, characterized in that: The process of training the conditional probability neural network model includes the following steps: Perform feature extraction, normalization, and encoding on electric vehicle charging behavior training data; Initialize the parameters of the conditional probability model corresponding to each feature; In each training round, residual connections are used to improve gradient flow, attention mechanisms are used to extract features, the gradient of each variable is calculated, and the parameters of the corresponding conditional probability model are updated using gradient descent with a learning rate that decreases with the number of training rounds. After training is completed, the parameters of the conditional probability model are returned.

6. The method for generating electric vehicle charging data based on attention and conditional probability distribution according to claim 5, characterized in that: The learning rate is: in, For the t The learning rate of the round, is the initial learning rate, is the total number of rounds, is the termination factor.

7. The method for generating electric vehicle charging data based on attention and conditional probability distribution according to claim 1, characterized in that: The process of performing Gibbs sampling under variable constraints based on the trained conditional probability neural network model to screen electric vehicle charging behavior data that meets physical constraints includes the following steps: Initialize the batch sample matrix and generate initial values; Update the sample value through Gibbs sampling iteration; Set variable-level constraints; When the number of iterations reaches the threshold, a stable distribution of samples is obtained; Filtering electric vehicle charging behavior data that complies with physical constraints; The filtered electric vehicle charging behavior data is saved as structured data for electric vehicle charging behavior analysis and prediction.

8. The method for generating electric vehicle charging data based on attention and conditional probability distribution according to claim 1, characterized in that: The continuous value features include an initial state of charge and an end state of charge, the time-related features include a charging start time and a charging duration, and the discrete value features include a battery capacity and a user type.

9. An electronic device, characterized in that: include: One or more processors and a memory, wherein the memory stores one or more programs, wherein the one or more programs include instructions for executing the electric vehicle charging data generation method based on attention and conditional probability distribution as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that It includes one or more programs for execution by one or more processors of an electronic device, wherein the one or more programs include instructions for executing the electric vehicle charging data generation method based on attention and conditional probability distribution as described in any one of claims 1-8.

Images