Electric vehicle charging load prediction method and system considering data cleaning

Abstract

The invention discloses an electric vehicle charging load prediction method and system considering data cleaning, and the method comprises the steps: collecting historical electric vehicle load related data and historical electric vehicle load data, and extracting electric vehicle load related features; performing normalization processing on the collected data, cleaning the normalized electric vehicle load data, and forming a training data set by using the normalized electric vehicle load related characteristics and the cleaned data; establishing a load prediction model based on a CNN-GRU convolutional neural network-gating cycle unit; training a load prediction model by using the training data set; and acquiring related characteristics of the electric vehicle load in real time, inputting the related characteristics into the trained load prediction model, and predicting the short-term charging load of the electric vehicle in real time. The electric vehicle charging load prediction method has the outstanding advantages that the electric vehicle charging load prediction result is good, the model practicability is high, and complex factors influencing electric vehicle charging load prediction are fully considered.

Description

Technical Field

[0001] The present invention belongs to the field of new energy vehicles, and in particular relates to electric vehicle charging load prediction and power grid dispatching management technology. Background Art

[0002] Electric vehicles play an important role in the transition to a low-carbon economy, and by reducing carbon emissions, they have promoted the development of new energy vehicles. However, the large-scale access of electric vehicles to the power grid will bring about load peak problems, which in turn will affect the stable operation of the distribution network and may cause safety hazards such as voltage instability and line overload. Therefore, the charging load prediction of electric vehicles has become the key to solving this problem.

[0003] In the field of electric vehicle load forecasting, there are several key issues. First, electric vehicle load is affected by many factors, including environmental factors, external factors, and infrastructure factors, which makes it difficult to determine which factors have a high correlation with the load. Secondly, there are often outliers in the load data, which have a significant impact on the forecast results. Therefore, the data must be cleaned before forecasting to remove or correct the outliers to improve the accuracy of the forecast model. In addition, since electric vehicle load is affected by many factors, the relationship between these factors is often complex and may have nonlinear characteristics. Deep learning models, especially deep neural networks, have powerful feature learning and nonlinear mapping capabilities, and can automatically extract potential patterns and relationships from complex data, and effectively identify key factors affecting the load.

[0004] CN116960958A discloses a method and device for predicting the charging load of electric vehicles in a distribution network. The method specifically includes the following steps: obtaining a preset traffic network simulation dynamic simulation model; combining the traffic network simulation dynamic simulation model, determining the travel behavior of each electric vehicle, and providing the start and end matrix of the electric vehicle; analyzing the travel cost of the electric vehicle, based on the minimum travel cost strategy, combining the start and end matrix, obtaining the driving path information of each electric vehicle; based on the driving path information of each electric vehicle, analyzing and determining the charging expectation of each electric vehicle, and combining the electric vehicle charging station information, completing the prediction of the charging load of the electric vehicle in the distribution network. The patent uses fewer types of data for training the model, and it is difficult to reflect the characteristics of vehicle charging. Therefore, it is difficult for the model to adjust the prediction results in real time after training, so that the prediction results have a large deviation.

[0005] CN118228865A discloses a method and system for predicting electric vehicle charging load, which relates to the technical field of electric power load prediction, and includes the following steps: receiving electric vehicle charging load data, preprocessing the electric vehicle charging load data, obtaining processed electric vehicle charging load data, performing variable mode decomposition on the processed electric vehicle charging load data, obtaining multiple decomposition sequences; inputting the multiple decomposition sequences into a pre-established multi-channel 1DCNN-GRU combined neural network model, and outputting the electric vehicle charging load prediction result. The data preprocessing method of the patent does not significantly improve the data quality, and the variable mode decomposition in the patent is used as part of the feature extraction, but this method is not comprehensive enough when processing the complex features in the patent, and it is easy to ignore important nonlinear features, so the prediction results of the trained model have large deviations and are not practical. Summary of the invention

[0006] In order to solve the shortcomings in the prior art such as the input data of the training model taking into account few complex factors affecting the prediction of electric vehicle charging load, large deviations in the prediction results of the electric vehicle charging load, and low practicality of the model, the present invention provides an electric vehicle charging load prediction method and system taking data cleaning into consideration.

[0007] The present invention adopts the following technical solution.

[0008] The present invention discloses, on one hand, a method for predicting charging load of an electric vehicle taking data cleaning into consideration, comprising:

[0009] Step 1: Collect historical electric vehicle load-related data and historical electric vehicle load data, and extract electric vehicle load-related features;

[0010] Step 2: normalize the collected data, clean the normalized electric vehicle load data, and use the normalized electric vehicle load related features and the cleaned data to form a training data set;

[0011] Step 3: Establish a load forecasting model based on CNN-GRU convolutional neural network-gated recurrent unit;

[0012] Step 4: Train the load forecasting model using the training data set;

[0013] Step 5: Collect the relevant features of electric vehicle load in real time and input them into the trained load forecasting model to make real-time forecasts of the short-term charging load of electric vehicles.

[0014] More preferably,

[0015] In step 1, the historical electric vehicle load-related data includes environmental factors and infrastructure factors; the environmental factors include temperature and humidity; the infrastructure factors include the capacity of the charging pile, the power of the electric meter and the capacity of the transformer;

[0016] Each item of historical electric vehicle load related data is recorded as a characteristic variable.

[0017] More preferably,

[0018] In step 1, the electric vehicle load-related features are extracted by calculating the correlation value between each feature variable and the electric vehicle load data, and selecting the feature variable corresponding to the correlation value whose absolute value is greater than the set correlation threshold as the electric vehicle load-related features; the correlation threshold is set to 0.5; the correlation value ρ of the nth feature variable and the electric vehicle load data is (n) Calculated as follows:

[0019]

[0020] Where n is an integer, and n∈[1,5]; m is the total number of sampling points of the nth feature variable; re (n) Calculated according to the following formula:

[0021]

[0022] Where i is an integer, and i∈[1,m]; It means the serial number value corresponding to the i-th sampling point after the sampling values ​​in the n-th characteristic variable are sorted from small to large and labeled; rank(y i ) represents the serial number value corresponding to the i-th sampling point after the sampling values ​​in the electric vehicle load are sorted from small to large and numbered.

[0023] More preferably,

[0024] In step 2, the cleaning method is: constructing a generator based on the autoencoder AE, the generative adversarial network GAN and Gaussian noise; constructing a discriminator based on the generative adversarial network;

[0025] The input of the generator is a normalized electric vehicle load data matrix; the output is a synthetic data matrix close to the normalized electric vehicle load data matrix;

[0026] The input of the discriminator is a mixed data matrix composed of a normalized electric vehicle load data matrix and a synthetic data matrix; the output is a discrimination matrix composed of values ​​in the range of [0,1]; wherein each value in the discrimination matrix reflects the judgment of the discriminator on the authenticity of the corresponding element in the mixed data matrix;

[0027] Perform adversarial training on the generator and discriminator;

[0028] The trained generator synthesized data is used as the cleaning result of the normalized electric vehicle load data.

[0029] More preferably,

[0030] The generator is constructed as follows:

[0031] The Gaussian noise is introduced into the normalized electric vehicle load data according to the following formula to obtain the noisy data matrix

[0032]

[0033] Where Y is the normalized electric vehicle load data matrix; N G is a Gaussian noise matrix with the same dimension as the normalized electric vehicle load data matrix;

[0034] The encoder is used to extract features from the noisy data matrix, denoted as z;

[0035] The decoder is used to reconstruct the data matrix based on the features extracted by the encoder as synthetic data

[0036] More preferably,

[0037] The overall loss function used by the generator Calculated as follows:

[0038]

[0039] Among them, l prior is the regularization loss; l rec is the reconstruction error; γ is a hyperparameter that balances the importance of the reconstruction error and the prior regularization term, and its initial value is set to 0.1.

[0040] More preferably,

[0041] The regularization loss is calculated as follows:

[0042]

[0043] Where D KL (P||Q) means calculating the KL divergence of probability distribution P relative to probability distribution Q;

[0044] is the distribution of features extracted by the encoder; is a standard normal distribution;

[0045] The reconstruction error is calculated as follows:

[0046]

[0047] Where N is the number of elements in the normalized electric vehicle load data matrix; Y i is the ith value in the normalized electric vehicle load data matrix, is the i-th value in the synthetic data matrix.

[0048] More preferably,

[0049] l pro Calculated as follows:

[0050] l pro =-E[M (G) ⊙log2D(Y (d) )];

[0051] Where A⊙B represents the dot product of matrix A and matrix B; (G) Y is the first mask matrix generated according to the mixed data matrix input to the discriminator; (d) is the mixed data matrix of the discriminator input, D(Y (d) ) is the discriminant matrix output by the discriminator; log2() refers to the logarithmic matrix obtained by solving the logarithm with base 2 for each value in the matrix in brackets; E() refers to the expectation of the elements in the matrix in brackets;

[0052] The first mask matrix is ​​calculated as follows:

[0053]

[0054] in is the i-th value in the first mask matrix; Y i (d) is the i-th value in the mixed data matrix; Y i is the ith value in the normalized electric vehicle load data matrix.

[0055] More preferably,

[0056] The identification loss function used by the discriminator Calculated as follows:

[0057]

[0058] Among them, M (D) is the second mask matrix generated according to the discriminator input data, and its calculation formula is as follows:

[0059]

[0060] in, is the i-th value in the second mask matrix.

[0061] More preferably,

[0062] In step 4, the electric vehicle load-related characteristics of the previous one or more time periods are used as inputs to the load forecasting model, and the synthetic data of the current time period is used as outputs of the load forecasting model;

[0063] The loss function L used in the load forecasting model is l Calculated as follows:

[0064]

[0065] in, is the i-th value in the synthetic data of the current period, is the ith value in the forecast data of the load forecasting model, and s is the number of values ​​in the synthetic data of the current period.

[0066] On the other hand, the present invention discloses an electric vehicle charging load prediction system that takes data cleaning into consideration by utilizing the aforementioned electric vehicle charging load prediction method, including a data acquisition feature extraction module, a data preprocessing module, a load prediction model establishment module, a load prediction model training module and an electric vehicle short-term charging load real-time prediction module.

[0067] The data collection feature extraction module collects historical electric vehicle load-related data and historical electric vehicle load data, and extracts electric vehicle load-related features; the historical electric vehicle load-related data includes environmental factors and infrastructure factors; the environmental factors include temperature and humidity; the infrastructure factors include the capacity of the charging pile, the power of the electric meter and the capacity of the transformer; and each data in the historical electric vehicle load-related data is recorded as a feature variable;

[0068] The data preprocessing module normalizes the collected data, cleans the normalized electric vehicle load data, and forms a training data set using the normalized electric vehicle load related features and the cleaned data;

[0069] The load forecasting model building module builds a load forecasting model based on a CNN-GRU convolutional neural network-gated recurrent unit;

[0070] The load forecasting model training module trains the load forecasting model using a training data set;

[0071] The electric vehicle short-term charging load real-time prediction module collects electric vehicle load-related features in real time and inputs them into a trained load prediction model to make a real-time prediction of the electric vehicle short-term charging load.

[0072] More preferably,

[0073] In the data acquisition feature extraction module, the electric vehicle load related features are extracted by calculating the correlation value between each feature variable and the electric vehicle load data, and selecting the feature variable corresponding to the correlation value whose absolute value is greater than the set correlation threshold as the electric vehicle load related features; the correlation threshold is set to 0.5; the correlation value ρ of the nth feature variable and the electric vehicle load data is (n) Calculated as follows:

[0074]

[0075] Where n is an integer, and n∈[1,5]; m is the total number of sampling points of the nth feature variable; re (n) Calculated according to the following formula:

[0076]

[0077] Where i is an integer, and i∈[1,m]; It means the serial number value corresponding to the i-th sampling point after the sampling values ​​in the n-th characteristic variable are sorted from small to large and labeled; rank(y i ) represents the serial number value corresponding to the i-th sampling point after the sampling values ​​in the electric vehicle load are sorted from small to large and numbered.

[0078] More preferably,

[0079] In the data preprocessing module, the cleaning method is as follows: construct a generator based on autoencoder AE, generative adversarial network GAN and Gaussian noise; construct a discriminator based on generative adversarial network;

[0080] The input of the generator is a normalized electric vehicle load data matrix; the output is a synthetic data matrix close to the normalized electric vehicle load data matrix;

[0081] The input of the discriminator is a mixed data matrix composed of a normalized electric vehicle load data matrix and a synthetic data matrix; the output is a discrimination matrix composed of values ​​in the range of [0,1]; wherein each value in the discrimination matrix reflects the judgment of the discriminator on the authenticity of the corresponding element in the mixed data matrix;

[0082] Perform adversarial training on the generator and discriminator;

[0083] The trained generator synthesized data is used as the cleaning result of the normalized electric vehicle load data.

[0084] More preferably,

[0085] The generator is constructed as follows:

[0086] The Gaussian noise is introduced into the normalized electric vehicle load data according to the following formula to obtain the noisy data matrix

[0087]

[0088] Where Y is the normalized electric vehicle load data matrix; N G is a Gaussian noise matrix with the same dimension as the normalized electric vehicle load data matrix;

[0089] The encoder is used to extract features from the noisy data matrix, denoted as z;

[0090] The decoder is used to reconstruct the data matrix based on the features extracted by the encoder as synthetic data

[0091] More preferably,

[0092] The overall loss function used by the generator Calculated as follows:

[0093]

[0094] Among them, l prior is the regularization loss; l rec is the reconstruction error; γ is a hyperparameter that balances the importance of the reconstruction error and the prior regularization term, and its initial value is set to 0.1;

[0095] The regularization loss is calculated as follows:

[0096]

[0097] Where D KL (P||Q) means calculating the KL divergence of probability distribution P relative to probability distribution Q;

[0098] is the distribution of features extracted by the encoder; is a standard normal distribution;

[0099] The reconstruction error is calculated as follows:

[0100]

[0101] Where N is the number of elements in the normalized electric vehicle load data matrix; Y i is the ith value in the normalized electric vehicle load data matrix, is the ith value in the synthetic data matrix.

[0102] l pro Calculated as follows:

[0103] l pro =-E[M (G) ⊙log2D(Y (d) )];

[0104] Where A⊙B represents the dot product of matrix A and matrix B; (G) Y is the first mask matrix generated according to the mixed data matrix input to the discriminator; (d) is the mixed data matrix of the discriminator input, D(Y (d) ) is the discriminant matrix output by the discriminator; log2() refers to the logarithmic matrix obtained by solving the logarithm with base 2 for each value in the matrix in brackets; E() refers to the expectation of the elements in the matrix in brackets;

[0105] The first mask matrix is ​​calculated as follows:

[0106]

[0107] in is the i-th value in the first mask matrix; Y i (d) is the i-th value in the mixed data matrix; Y i is the ith value in the normalized electric vehicle load data matrix.

[0108] More preferably,

[0109] The identification loss function used by the discriminator Calculated as follows:

[0110]

[0111] Among them, M (D) is the second mask matrix generated according to the discriminator input data, and its calculation formula is as follows:

[0112]

[0113] in, is the i-th value in the second mask matrix.

[0114] More preferably,

[0115] The identification loss function used by the discriminator Calculated as follows:

[0116]

[0117] Among them, M (D) is the second mask matrix generated according to the discriminator input data, and its calculation formula is as follows:

[0118]

[0119] in, is the i-th value in the second mask matrix.

[0120] More preferably,

[0121] In the load forecasting model training module, the electric vehicle load-related characteristics of the previous one or more time periods are used as inputs of the load forecasting model, and the synthetic data of the current time period is used as outputs of the load forecasting model;

[0122] The loss function L used in the load forecasting model is l Calculated as follows:

[0123]

[0124] in, is the i-th value in the synthetic data of the current period, is the ith value in the forecast data of the load forecasting model, and s is the number of values ​​in the synthetic data of the current period.

[0125] On the other hand, the present application discloses an electronic device, including a processor and a storage medium; the characteristics are:

[0126] The storage medium is used to store instructions;

[0127] The processor is used to operate according to the instructions to execute the above-mentioned electric vehicle charging load prediction method.

[0128] The present application also discloses a computer-readable storage medium having a computer program stored thereon, wherein the program implements the electric vehicle charging load prediction method when executed by a processor.

[0129] The beneficial effects of the present invention are as follows:

[0130] 1. A new framework is proposed, which includes three key components: feature selection, data cleaning and prediction model. The framework successfully reduces the noise level in the data and significantly improves the data quality. It has made innovative breakthroughs in the accuracy of short-term load forecasting and provides more reliable and advanced tools for the intelligent management and optimization of electric vehicle charging piles. The framework not only attaches importance to the construction of the prediction model, but also integrates the feature selection and data cleaning links in load forecasting into the whole process, thereby significantly improving the accuracy of short-term power load forecasting on the basis of high-quality data.

[0131] 2. Innovatively proposed a load data cleaning method based on autoencoder AE, generative adversarial network GAN and Gaussian noise, which can more effectively remove noise and outliers in the data and retain the key features of the data, thereby significantly improving the data quality and achieving innovation in the accuracy of load forecasting. In addition, the present invention performs feature extraction before data cleaning to ensure that the prediction model inputs an optimized feature data set.

[0132] 3. The present invention builds an electric vehicle charging load forecasting model based on a composite network structure consisting of stacked CNN convolutional layers and GRU layers. The model fully utilizes the complementary advantages of convolutional layers and GRU layers in capturing complex patterns of load fluctuations, significantly improves the accuracy of short-term load forecasting, and provides robust data support for the planning and operation of electric vehicle charging stations.

[0133] 4. The present invention also has outstanding advantages such as good electric vehicle charging load prediction results, strong model practicability, and full consideration of the complex factors affecting the electric vehicle charging load prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0134] Figure 1 It is a flow chart of a method for predicting charging load of electric vehicles considering data cleaning according to the present invention;

[0135] Figure 2 It is a comparison chart of the MAE of the load forecasting model prediction results of the present invention and the traditional single load forecasting model prediction results. DETAILED DESCRIPTION

[0136] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only part of the embodiments of the present invention, not all of them. Based on the spirit of the present invention, other embodiments obtained by ordinary technicians in this field without creative work are all within the scope of protection of the present invention.

[0137] The present invention discloses a method for predicting electric vehicle charging load considering data cleaning, see the attached Figure 1 ,include:

[0138] Step 1: Collect historical electric vehicle load-related data and historical electric vehicle load data, and extract electric vehicle load-related features;

[0139] The electric vehicle load-related data are divided into three categories: environmental factors, external factors, and infrastructure factors. Considering that the method for calculating the correlation coefficient used in the present invention is suitable for analyzing sequential variables, continuous variables, and processing nonlinear relationships and non-normal distribution data, temperature and humidity are selected as characteristic variables for analysis in environmental factors. For external factors, such as policies and regulations, market demand, etc., since they are usually difficult to quantify into specific values, they are not considered for the time being. As for infrastructure factors, the capacity of the charging pile, the power of the meter, and the capacity of the transformer are selected as characteristic variables for analysis.

[0140] Therefore, the historical electric vehicle load-related data of the present invention includes environmental factors and infrastructure factors; wherein the environmental factors include temperature and humidity; the infrastructure factors include the capacity of the charging pile, the power of the electric meter and the capacity of the transformer;

[0141] Each item of historical electric vehicle load related data is recorded as a characteristic variable.

[0142] The extracting of electric vehicle load related features is to calculate the correlation value between each feature variable and the electric vehicle load data, and select the feature variable corresponding to the correlation value whose absolute value is greater than the set correlation threshold as the electric vehicle load related feature;

[0143] To calculate the correlation value between each characteristic variable and the electric vehicle load data, each type of collected data needs to be arranged in order from small to large and numbered. For example, assuming that the total number of data points of a certain type of data is m, the smallest data point is ranked first and its number is 1, the second smallest data point is ranked second and its number is 2. If there are two third smallest data points, the number of both data points is 3, the next data point is ranked fifth and its number is 5, and so on.

[0144] After marking the serial numbers, the sum of squares of the serial number differences between each characteristic variable and the electric vehicle load variable number is calculated to obtain the degree of correlation between each type of characteristic variable and the electric vehicle load.

[0145] The correlation value ρ between the nth characteristic variable and the electric vehicle load data (n) Calculated as follows:

[0146]

[0147] Where n is an integer, and n∈[1,5], indicating that the correlation between each characteristic variable and the electric vehicle load data needs to be calculated in turn; m is the total number of sampling points of the nth characteristic variable; re (n) Calculated according to the following formula:

[0148]

[0149] Where i is an integer, and i∈[1,m]; is the i-th sample value of the n-th characteristic variable;

[0150] It represents the serial number value corresponding to the i-th sampling point after the sampling values ​​in the n-th characteristic variable are sorted from small to large and labeled; y i is the i-th sampling value of the electric vehicle load data; rank(y i ) represents the serial number value corresponding to the i-th sampling point after the sampling values ​​in the electric vehicle load are sorted from small to large and numbered.

[0151] The correlation between each characteristic variable and the electric vehicle load is between -1 and 1. A value of 1 indicates a complete positive correlation, a value of -1 indicates a complete negative correlation, and a value of 0 indicates no correlation. By calculating the correlation between each characteristic variable and the load data, the characteristic variable whose absolute value of the calculation result is greater than the set correlation threshold is selected as the electric vehicle load related feature; the set correlation threshold is 0.5.

[0152] Step 2: normalize the collected data, clean the normalized electric vehicle load data, and use the normalized electric vehicle load related features and the cleaned data to form a training data set;

[0153] The collected data are normalized using the following formula:

[0154]

[0155] Where X0′ is any item of the collected data, X' is the standardized result of the data represented by X0′, μ0 is the average value of the data represented by X0′, and σ0 is the standard deviation of the data represented by X0′.

[0156] The cleaning method is as follows:

[0157] The generator is constructed based on autoencoder AE, generative adversarial network GAN and Gaussian noise.

[0158] The generator generates synthetic data that is as close as possible to the normalized EV load data.

[0159] The construction method of the generator is as follows:

[0160] First, Gaussian noise is introduced into the cleaned data according to the following formula to obtain a noisy data matrix

[0161]

[0162] Where Y is the normalized electric vehicle load data matrix; NG is a Gaussian noise matrix with the same dimension as the normalized electric vehicle load data matrix; Indicates that the Gaussian noise matrix has a mean of 0 and a variance of σ 2 The normally distributed noise matrix of the two matrices has the same matrix dimension, that is, the number of rows and columns of the two matrices are the same; σ is selected according to historical experience or expert system.

[0163] An encoder is used to map the noisy data matrix to a low-dimensional latent space to generate a latent representation z, that is, to extract features from the noisy data matrix, denoted as z.

[0164] The decoder is used to decode the latent representation z back to the original data space to generate a time series similar to the input encoder data as synthetic data That is, based on the features extracted by the encoder, the data matrix is ​​reconstructed as synthetic data

[0165] The input of the generator is a normalized electric vehicle load data matrix; the output is a synthetic data matrix close to the normalized electric vehicle load data matrix.

[0166] The overall loss function used by the generator The generator synthesized data can be made as realistic as possible. The loss function is calculated as follows:

[0167]

[0168] Among them, l prior is the regularization loss, which is used to ensure that the distribution of the potential representation z is close to a normal distribution; l rec is the reconstruction error, which is used to measure the difference between the synthetic data and the normalized EV load data; l pro The loss is used to measure the ability of the generated data to deceive the discriminator; γ is a hyperparameter that balances the importance of reconstruction error and prior regularization term, and its initial value is set to 0.1.

[0169] The regularization loss is calculated as follows:

[0170]

[0171] Where D KL (P||Q) means calculating the KL divergence of probability distribution P relative to probability distribution Q, which is used to distinguish and compare the differences between two probability distributions; is the distribution of features extracted by the encoder;

[0172] is a standard normal distribution;

[0173] Assume the encoder output distribution is a Gaussian distribution with parameters μ and σ Then the calculation formula of KL divergence is:

[0174]

[0175] The reconstruction error is calculated as follows:

[0176]

[0177] Where N is the number of elements in the normalized electric vehicle load data matrix; Y i is the ith value in the normalized electric vehicle load data matrix, is the i-th value in the synthetic data matrix.

[0178] l pro Calculated as follows:

[0179] l pro =-E[M (G) ⊙log2D(Y (d) )];

[0180] Where A⊙B represents the dot product of matrix A and matrix B, that is, multiplying the elements of the corresponding positions of matrices A and B; (G) Y is the first mask matrix with the same dimension as the mixed data matrix generated according to the mixed data matrix of the input discriminator; (d) is the mixed data matrix of the discriminator input, D(Y (d) ) is the discriminant matrix output by the discriminator;

[0181] log2() means solving the logarithm with base 2 for each value in the matrix in brackets, and the result is recorded as a logarithmic matrix; E() means the expectation of the elements in the matrix in brackets;

[0182] The first mask matrix is ​​calculated as follows:

[0183]

[0184] in is the i-th value in the first mask matrix; Y i (d) is the i-th value in the mixed data matrix of the input discriminant matrix; Y i is the ith value in the normalized electric vehicle load data matrix.

[0185] Building a discriminator based on a generative adversarial network;

[0186] The discriminator is used to distinguish between the generator synthetic data and the normalized electric vehicle load data.

[0187] The input of the discriminator is a mixed data matrix Y consisting of a normalized electric vehicle load data matrix and a synthetic data matrix (d) ;

[0188] In a preferred embodiment of the present invention, the hybrid data matrix is ​​formed by randomly selecting elements in the synthetic data matrix and replacing the corresponding elements in the normalized electric vehicle load data matrix, that is, in are chosen randomly.

[0189] The output of the discriminator is a discriminant matrix composed of values ​​in the range of [0,1]; each value in the discriminant matrix reflects the discriminator's judgment on the authenticity of the corresponding elements in the mixed data matrix; if a value in the discriminant matrix is ​​0, it means that the discriminator believes that the corresponding data in the mixed matrix is ​​completely false, that is, the data comes from the synthetic data matrix. If a value in the discriminant matrix is ​​1, it means that the discriminator believes that the corresponding data in the mixed data matrix is ​​completely true, that is, the data comes from the normalized electric vehicle load data matrix.

[0190] The identification loss function used by the discriminator Calculated as follows:

[0191]

[0192] Among them, M (D) M is a second mask matrix with the same dimension as the mixed data matrix generated according to the mixed data matrix input by the discriminator; D ⊙log2[D(X d )⊙M D ] means that the loss of the discriminator correctly identifying the real data is calculated only when the corresponding position in M ​​is 1; (1-M D )⊙log2[1-D(X d )⊙M D ] means that only in M D The corresponding position in is 0, and the loss of the discriminator correctly identifying the generated data is calculated. The negative sign before the expectation in this formula is used to minimize the expected value of the loss function.

[0193] M (D) The calculation formula is as follows:

[0194]

[0195] in, is the i-th value in the second mask matrix.

[0196] The generator and the discriminator are trained adversarially; through adversarial training between the generator and the discriminator, the performance of both is optimized.

[0197] The trained generator synthesized data is used as the cleaning result of the normalized electric vehicle load data.

[0198] Step 3: Establish a load forecasting model based on CNN-GRU convolutional neural network-gated recurrent unit;

[0199] The load forecasting model is provided with an input layer, multiple one-dimensional convolutional layers 1D-CNN, multiple GRU layers and an output layer.

[0200] In the preferred embodiment of the present invention, the input layer of the load forecasting model reads the input data and transmits the input data to the first one-dimensional convolutional layer. The first one-dimensional convolutional layer extracts features from the input load data to capture local patterns and features in the data, and then passes the output result to the second one-dimensional convolutional layer to extract local patterns and features in the data again. Then the output data of the second convolutional layer is input to the first GRU layer, and the output data of the first GRU layer is input to the second GRU layer. The preferred embodiment of the present invention models the long-term dependency in the short-term load data by stacking two GRU layers, while effectively avoiding the gradient vanishing problem; finally, the output data of the second GRU layer is passed through the output layer to obtain the electric vehicle charging load forecast result.

[0201] Step 4: Train the load forecasting model using the training data set;

[0202] During training, the electric vehicle load-related characteristics of the previous one or more time periods are used as inputs to the load forecasting model, and the synthetic data of the current time period is used as outputs of the load forecasting model;

[0203] The loss function L used in the load forecasting model is l Calculated as follows:

[0204]

[0205] in, is the i-th value in the synthetic data of the current period, is the ith value in the forecast data of the load forecasting model, and s is the number of values ​​in the synthetic data of the current period.

[0206] Step 5: Collect the relevant features of electric vehicle load in real time and input them into the trained load forecasting model to make real-time forecasts of the short-term charging load of electric vehicles.

[0207] The present invention also discloses an electric vehicle charging load prediction system that takes data cleaning into consideration based on the above-mentioned electric vehicle charging load prediction method, including a data acquisition feature extraction module, a data preprocessing module, a load prediction model establishment module, a load prediction model training module and an electric vehicle short-term charging load real-time prediction module.

[0208] The data collection feature extraction module collects historical electric vehicle load related data and historical electric vehicle load data, and extracts electric vehicle load related features;

[0209] The historical electric vehicle load-related data include environmental factors and infrastructure factors; the environmental factors include temperature and humidity; the infrastructure factors include the capacity of the charging pile, the power of the electricity meter and the capacity of the transformer; and each data in the historical electric vehicle load-related data is recorded as a characteristic variable.

[0210] The extraction of electric vehicle load related features is to calculate the correlation value between each feature variable and the electric vehicle load data, and select the feature variable corresponding to the correlation value whose absolute value is greater than the set correlation threshold as the electric vehicle load related feature; the correlation threshold is set to 0.5; the correlation value ρ of the nth feature variable and the electric vehicle load data (n) Calculated as follows:

[0211]

[0212] Where n is an integer, and n∈[1,5]; m is the total number of sampling points of the nth feature variable; re (n) Calculated according to the following formula:

[0213]

[0214] Where i is an integer, and i∈[1,m]; It means the serial number value corresponding to the i-th sampling point after the sampling values ​​in the n-th characteristic variable are sorted from small to large and labeled; rank(y i ) represents the serial number value corresponding to the i-th sampling point after the sampling values ​​in the electric vehicle load are sorted from small to large and numbered.

[0215] The data preprocessing module normalizes the collected data, cleans the normalized electric vehicle load data, and forms a training data set using the normalized electric vehicle load related features and the cleaned data.

[0216] The cleaning method is:

[0217] Construct a generator based on autoencoder AE, generative adversarial network GAN and Gaussian noise;

[0218] The generator is constructed as follows:

[0219] The Gaussian noise is introduced into the normalized electric vehicle load data according to the following formula to obtain the noisy data matrix

[0220]

[0221] Where Y is the normalized electric vehicle load data matrix; N G is a Gaussian noise matrix with the same dimension as the normalized electric vehicle load data matrix;

[0222] The encoder is used to extract features from the noisy data matrix, denoted as z;

[0223] The decoder is used to reconstruct the data matrix based on the features extracted by the encoder as synthetic data

[0224] The overall loss function used by the generator Calculated as follows:

[0225]

[0226] Among them, l prior is the regularization loss; l rec is the reconstruction error; γ is a hyperparameter that balances the importance of the reconstruction error and the prior regularization term, and its initial value is set to 0.1;

[0227] The regularization loss is calculated as follows:

[0228]

[0229] Where D KL (P||Q) means calculating the KL divergence of probability distribution P relative to probability distribution Q;

[0230] is the distribution of features extracted by the encoder; is a standard normal distribution;

[0231] The reconstruction error is calculated as follows:

[0232]

[0233] Where N is the number of elements in the normalized electric vehicle load data matrix; Y i is the ith value in the normalized electric vehicle load data matrix, is the i-th value in the synthetic data matrix.

[0234] l pro Calculated as follows:

[0235] l pro =-E[M (G) ⊙log2D(Y (d) )];

[0236] Where A⊙B represents the dot product of matrix A and matrix B; (G)Y is the first mask matrix generated according to the mixed data matrix input to the discriminator; (d) is the mixed data matrix of the discriminator input, D(Y (d) ) is the discriminant matrix output by the discriminator; log2() refers to the logarithmic matrix obtained by solving the logarithm with base 2 for each value in the matrix in brackets; E() refers to the expectation of the elements in the matrix in brackets;

[0237] The first mask matrix is ​​calculated as follows:

[0238]

[0239] in is the i-th value in the first mask matrix; Y i (d) is the i-th value in the mixed data matrix; Y i is the ith value in the normalized electric vehicle load data matrix.

[0240] The input of the generator is a normalized electric vehicle load data matrix; the output is a synthetic data matrix close to the normalized electric vehicle load data matrix;

[0241] Building a discriminator based on a generative adversarial network;

[0242] The input of the discriminator is a mixed data matrix composed of a normalized electric vehicle load data matrix and a synthetic data matrix; the output is a discrimination matrix composed of values ​​in the range of [0,1]; wherein each value in the discrimination matrix reflects the judgment of the discriminator on the authenticity of the corresponding element in the mixed data matrix;

[0243] The identification loss function used by the discriminator Calculated as follows:

[0244]

[0245] Among them, M (D) is the second mask matrix generated according to the discriminator input data, and its calculation formula is as follows:

[0246]

[0247] in, is the i-th value in the second mask matrix.

[0248] Perform adversarial training on the generator and discriminator;

[0249] The trained generator synthesized data is used as the cleaning result of the normalized electric vehicle load data.

[0250] The load forecasting model building module builds a load forecasting model based on a CNN-GRU convolutional neural network-gated recurrent unit.

[0251] The load forecasting model training module uses a training data set to train the load forecasting model.

[0252] Using the electric vehicle load-related characteristics of the previous one or more time periods as the input of the load forecasting model, and using the synthetic data of the current time period as the output of the load forecasting model;

[0253] The loss function L used in the load forecasting model is l Calculated as follows:

[0254]

[0255] in, is the i-th value in the synthetic data of the current period, is the ith value in the forecast data of the load forecasting model, and s is the number of values ​​in the synthetic data of the current period.

[0256] The electric vehicle short-term charging load real-time prediction module collects electric vehicle load-related features in real time and inputs them into a trained load prediction model to make a real-time prediction of the electric vehicle short-term charging load.

[0257] Example

[0258] An embodiment of the present invention provides an electric vehicle load forecasting model method.

[0259] The data set used comes from 10 electric meters in an electric vehicle charging pile cluster, including electric vehicle load-related data and electric vehicle load data from January 1, 2021 to December 31, 2022; the electric vehicle load-related data includes temperature, humidity, charging pile capacity, meter power, and transformer capacity, etc. The data sampling interval is 15 minutes, with 96 sampling points per day, totaling about 360,000 training samples. All data in the electric vehicle load-related data are used as feature variables. The correlation between each feature variable and the electric vehicle load is calculated, and several features with a correlation degree higher than the set threshold are selected as electric vehicle load-related features. In this embodiment, the electric vehicle load-related features ultimately include three features: temperature, charging pile capacity, and meter power.

[0260] Then, the collected data are normalized. The normalized electric vehicle load data is processed using a data cleaning method based on AE, GAN and Gaussian noise settings. The cleaned electric vehicle load data is the generator synthetic data, and the synthetic data is used as the output data for training CNN-GRU.

[0261] Finally, the load forecasting model based on CNN-GRU convolutional neural network-gated recurrent unit proposed in the present invention sets two layers of one-dimensional convolution kernels and two layers of GRU to predict the charging load of electric vehicles. The normalized electric vehicle load-related features and generator synthetic data are used as data sets. During training, the electric vehicle load-related features are used as the input of the load forecasting model, and the generator synthetic data is used as the output of the load forecasting model; the data is divided into a training set and a test set in a ratio of 8:2. See Appendix Figure 2 ,The experimental results show that after 20 rounds of training, the proposed model significantly improves the effect of load forecasting; ,Compared with the traditional method, this method is far superior to the traditional single deep learning ,forecasting method in reducing the mean absolute error (MAE).

[0262] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0263] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples of computer-readable storage media (a non-exhaustive list) include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium is not to be interpreted as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0264] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0265] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be customized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0266] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for predicting electric vehicle charging load considering data cleaning, characterized in that: include: Step 1: Collect historical electric vehicle load-related data and historical electric vehicle load data, and extract electric vehicle load-related features; Step 2: normalize the collected data, clean the normalized electric vehicle load data, and use the normalized electric vehicle load related features and the cleaned data to form a training data set; Step 3: Establish a load forecasting model based on CNN-GRU convolutional neural network-gated recurrent unit; Step 4: Train the load forecasting model using the training data set; Step 5: Collect the relevant features of electric vehicle load in real time and input them into the trained load forecasting model to make real-time forecasts of the short-term charging load of electric vehicles.

2. The electric vehicle charging load prediction method according to claim 1, characterized in that: In step 1, the historical electric vehicle load-related data includes environmental factors and infrastructure factors; the environmental factors include temperature and humidity; the infrastructure factors include the capacity of the charging pile, the power of the electric meter and the capacity of the transformer; Each item of historical electric vehicle load related data is recorded as a characteristic variable.

3. The electric vehicle charging load prediction method according to claim 1 or 2, characterized in that: In step 1, extracting electric vehicle load related features is to calculate the correlation value between each feature variable and the electric vehicle load data, and select the feature variable corresponding to the correlation value whose absolute value is greater than the set correlation threshold as the electric vehicle load related feature; The correlation threshold is set to 0.5; the correlation value ρ of the nth characteristic variable and the electric vehicle load data (n) Calculated as follows: Where n is an integer, and n∈[1,5]; m is the total number of sampling points of the nth feature variable; re (n) Calculated according to the following formula: Where i is an integer, and i∈[1,m]; It means the serial number value corresponding to the i-th sampling point after the sampling values ​​in the n-th characteristic variable are sorted from small to large and labeled; rank(y i ) represents the serial number value corresponding to the i-th sampling point after the sampling values ​​in the electric vehicle load are sorted from small to large and numbered.

4. The electric vehicle charging load prediction method according to claim 1, characterized in that: In step 2, the cleaning method is: construct a generator based on the autoencoder AE, the generative adversarial network GAN and Gaussian noise; Building a discriminator based on a generative adversarial network; The input of the generator is the normalized electric vehicle load data matrix; The output is a synthetic data matrix that is close to the normalized electric vehicle load data matrix; The input of the discriminator is a mixed data matrix composed of a normalized electric vehicle load data matrix and a synthetic data matrix; the output is a discrimination matrix composed of values ​​in the range of [0,1]; wherein each value in the discrimination matrix reflects the judgment of the discriminator on the authenticity of the corresponding element in the mixed data matrix; Perform adversarial training on the generator and discriminator; The trained generator synthesized data is used as the cleaning result of the normalized electric vehicle load data.

5. The electric vehicle charging load prediction method according to claim 4 is characterized in that: The generator is constructed as follows: The Gaussian noise is introduced into the normalized electric vehicle load data according to the following formula to obtain the noisy data matrix Where Y is the normalized electric vehicle load data matrix; N G is a Gaussian noise matrix with the same dimension as the normalized electric vehicle load data matrix; The encoder is used to extract features from the noisy data matrix, denoted as z; The decoder is used to reconstruct the data matrix based on the features extracted by the encoder as synthetic data 6. The electric vehicle charging load prediction method according to claim 4 is characterized in that: The overall loss function used by the generator Calculated as follows: Among them, l prior is the regularization loss; l rec is the reconstruction error; γ is a hyperparameter that balances the importance of the reconstruction error and the prior regularization term, and its initial value is set to 0.

1.

7. The electric vehicle charging load prediction method according to claim 5 or 6, characterized in that: The regularization loss is calculated as follows: Where D KL (P||Q) means calculating the KL divergence of probability distribution P relative to probability distribution Q; is the distribution of features extracted by the encoder; is a standard normal distribution; The reconstruction error is calculated as follows: Where N is the number of elements in the normalized electric vehicle load data matrix; Y i is the ith value in the normalized electric vehicle load data matrix, is the ith value in the synthetic data matrix.

8. The electric vehicle charging load prediction method according to claim 4 or 6, characterized in that: l pro Calculated as follows: l pro =-E[M (G) ⊙log2D(Y (d) )]; Where A⊙B represents the dot product of matrix A and matrix B; (G) Y is the first mask matrix generated according to the mixed data matrix input to the discriminator; (d) is the mixed data matrix of the discriminator input, D(Y (d) ) is the discriminant matrix output by the discriminator; log2() refers to the logarithmic matrix obtained by solving the logarithm with base 2 for each value in the matrix in brackets; E() refers to the expectation of the elements in the matrix in brackets; The first mask matrix is ​​calculated as follows: in is the i-th value in the first mask matrix; is the i-th value in the mixed data matrix; Y i is the ith value in the normalized electric vehicle load data matrix.

9. The electric vehicle charging load prediction method according to claim 6, characterized in that: The identification loss function used by the discriminator Calculated as follows: Among them, M (D) is the second mask matrix generated according to the discriminator input data, and its calculation formula is as follows: in, is the i-th value in the second mask matrix.

10. The electric vehicle charging load prediction method according to claim 1 or 4, characterized in that: In step 4, the electric vehicle load-related characteristics of the previous one or more time periods are used as inputs to the load forecasting model, and the synthetic data of the current time period is used as outputs of the load forecasting model; The loss function L used in the load forecasting model is l Calculated as follows: in, is the i-th value in the synthetic data of the current period, is the ith value in the forecast data of the load forecasting model, and s is the number of values ​​in the synthetic data of the current period.

11. An electric vehicle charging load prediction system taking into account data cleaning using the electric vehicle charging load prediction method according to any one of claims 1 to 10, comprising a data acquisition feature extraction module, a data preprocessing module, a load prediction model establishment module, a load prediction model training module and an electric vehicle short-term charging load real-time prediction module, characterized in that: The data collection feature extraction module collects historical electric vehicle load related data and historical electric vehicle load data, and extracts electric vehicle load related features; The historical electric vehicle load-related data include environmental factors and infrastructure factors; the environmental factors include temperature and humidity; the infrastructure factors include the capacity of the charging pile, the power of the meter and the capacity of the transformer; the various data in the historical electric vehicle load-related data are recorded as characteristic variables; The data preprocessing module normalizes the collected data, cleans the normalized electric vehicle load data, and forms a training data set using the normalized electric vehicle load related features and the cleaned data; The load forecasting model building module builds a load forecasting model based on a CNN-GRU convolutional neural network-gated recurrent unit; The load forecasting model training module trains the load forecasting model using a training data set; The electric vehicle short-term charging load real-time prediction module collects electric vehicle load-related features in real time and inputs them into a trained load prediction model to make a real-time prediction of the electric vehicle short-term charging load.

12. The electric vehicle charging load prediction system according to claim 11, characterized in that: In the data acquisition feature extraction module, the electric vehicle load related features are extracted by calculating the correlation value between each feature variable and the electric vehicle load data, and selecting the feature variable corresponding to the correlation value whose absolute value is greater than the set correlation threshold as the electric vehicle load related features; The correlation threshold is set to 0.5; the correlation value ρ of the nth characteristic variable and the electric vehicle load data (n) Calculated as follows: Where n is an integer, and n∈[1,5]; m is the total number of sampling points of the nth feature variable; re (n) Calculated according to the following formula: Where i is an integer, and i∈[1,m]; It means the serial number value corresponding to the i-th sampling point after the sampling values ​​in the n-th characteristic variable are sorted from small to large and labeled; rank(y i ) represents the serial number value corresponding to the i-th sampling point after the sampling values ​​in the electric vehicle load are sorted from small to large and numbered.

13. The electric vehicle charging load prediction system according to claim 11, characterized in that: In the data preprocessing module, the cleaning method is: constructing a generator based on autoencoder AE, generative adversarial network GAN and Gaussian noise; Building a discriminator based on a generative adversarial network; The input of the generator is the normalized electric vehicle load data matrix; The output is a synthetic data matrix that is close to the normalized electric vehicle load data matrix; The input of the discriminator is a mixed data matrix composed of a normalized electric vehicle load data matrix and a synthetic data matrix; the output is a discrimination matrix composed of values ​​in the range of [0,1]; wherein each value in the discrimination matrix reflects the judgment of the discriminator on the authenticity of the corresponding element in the mixed data matrix; Perform adversarial training on the generator and discriminator; The trained generator synthesized data is used as the cleaning result of the normalized electric vehicle load data.

14. The electric vehicle charging load prediction system according to claim 13, characterized in that: The generator is constructed as follows: The Gaussian noise is introduced into the normalized electric vehicle load data according to the following formula to obtain the noisy data matrix Where Y is the normalized electric vehicle load data matrix; N G is a Gaussian noise matrix with the same dimension as the normalized electric vehicle load data matrix; The encoder is used to extract features from the noisy data matrix, denoted as z; The decoder is used to reconstruct the data matrix based on the features extracted by the encoder as synthetic data 15. The electric vehicle charging load prediction system according to claim 13 or 14, characterized in that: The overall loss function used by the generator Calculated as follows: Among them, l prior is the regularization loss; l rec is the reconstruction error; γ is a hyperparameter that balances the importance of the reconstruction error and the prior regularization term, and its initial value is set to 0.1; The regularization loss is calculated as follows: Where D KL (P||Q) means calculating the KL divergence of probability distribution P relative to probability distribution Q; is the distribution of features extracted by the encoder; is a standard normal distribution; The reconstruction error is calculated as follows: Where N is the number of elements in the normalized electric vehicle load data matrix; Y i is the ith value in the normalized electric vehicle load data matrix, is the i-th value in the synthetic data matrix; l pro Calculated as follows: l pro =-E[M (G) ⊙log2D(Y (d) )]; Where A⊙B represents the dot product of matrix A and matrix B; (G) Y is the first mask matrix generated according to the mixed data matrix input to the discriminator; (d) is the mixed data matrix of the discriminator input, D(Y (d) ) is the discriminant matrix output by the discriminator; log2() refers to the logarithmic matrix obtained by solving the logarithm with base 2 for each value in the matrix in brackets; E() refers to the expectation of the elements in the matrix in brackets; The first mask matrix is ​​calculated as follows: in is the i-th value in the first mask matrix; is the i-th value in the mixed data matrix; Y i is the ith value in the normalized electric vehicle load data matrix.

16. The electric vehicle charging load prediction system according to claim 13, characterized in that: The identification loss function used by the discriminator Calculated as follows: Among them, M (D) is the second mask matrix generated according to the discriminator input data, and its calculation formula is as follows: in, is the i-th value in the second mask matrix.

17. The electric vehicle charging load prediction system according to claim 11 or 13, characterized in that: In the load forecasting model training module, the electric vehicle load-related characteristics of the previous one or more time periods are used as inputs of the load forecasting model, and the synthetic data of the current time period is used as outputs of the load forecasting model; The loss function L used in the load forecasting model is l Calculated as follows: in, is the i-th value in the synthetic data of the current period, is the ith value in the forecast data of the load forecasting model, and s is the number of values ​​in the synthetic data of the current period.

18. An electronic device comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the electric vehicle charging load prediction method according to claims 1-10.

19. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the electric vehicle charging load prediction method described in claims 1-10 is implemented.

Images