Method and device for processing charging load

By meshing the historical charging data and prediction of Bayesian regularized neural network model, the problem of low charging load accuracy caused by limited number of charging facilities and unreasonable layout is solved, and the spatial and temporal distribution prediction of electric vehicle charging load is realized and refine planning.

CN112613637BActive Publication Date: 2024-12-03STATE GRID BEIJING ELECTRIC POWER CO +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202011377051.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-11-30
Publication Date
2024-12-03
Estimated Expiration
2040-11-30

AI Technical Summary

Technical Problem

The number of existing charging facilities is limited and the layout is unreasonable, resulting in low accuracy in determining the charging load of electric vehicles and the actual charging needs cannot be correctly reflected.

Method used

By obtaining historical charging data, dividing it into multiple grids, and using Bayesian regularization algorithm to train the backpropagation neural network model, predicting the charging load of each grid in the preset time period, thereby realizing the charging load of electric vehicles. spatiotemporal distribution.

Benefits of technology

It improves the prediction accuracy of charging load, solves the problem of low accuracy in determining charging load, and provides a more refined charging demand analysis and facility planning basis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112613637B_ABST
    Figure CN112613637B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and device for processing charging load. The method comprises: obtaining historical charging data; dividing the historical charging data to obtain historical data corresponding to multiple grids; using a load prediction model to predict the historical data corresponding to each grid to obtain the charging load of each grid within a preset time period. The present invention solves the technical problem of low accuracy in determining charging load in related technologies.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of new energy vehicles, and in particular to a method and device for processing charging load. Background Art

[0002] The charging behavior of electric vehicle users is a key factor influencing the charging load of electric vehicles. Different charging behaviors indicate different driving characteristics of the vehicles, which in turn lead to different charging load requirements. However, due to the limited number of existing charging facilities and their irrational layout, historical charging data cannot accurately reflect the actual charging demand of electric vehicles, resulting in low accuracy in determining charging load.

[0003] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0004] The embodiments of the present invention provide a method and device for processing a charging load, so as to at least solve the technical problem of low accuracy in determining the charging load in the related art.

[0005] According to one aspect of an embodiment of the present invention, a method for processing charging load is provided, comprising: obtaining historical charging data; dividing the historical charging data to obtain historical data corresponding to multiple grids; and using a load prediction model to predict the historical data corresponding to each grid to obtain the charging load of each grid within a preset time period.

[0006] Optionally, dividing the historical charging data to obtain historical data corresponding to multiple grids includes: obtaining a target area corresponding to the historical charging data; dividing the target area according to longitude and latitude information to obtain multiple grids; and allocating the historical charging data to the multiple grids to obtain historical data corresponding to each grid.

[0007] Optionally, the load forecasting model is obtained by training a back propagation neural network model using a Bayesian regularization algorithm.

[0008] Optionally, the method also includes: obtaining prediction indicators corresponding to the charging load, wherein the prediction indicators are used to characterize factors affecting the charging load; constructing a back-propagation neural network model; generating training samples using the prediction indicators; and training the back-propagation neural network model using a Bayesian regularization algorithm and the training samples to obtain a load prediction model.

[0009] Optionally, obtaining the prediction indicator corresponding to the charging load includes: obtaining multiple target data corresponding to the charging load, wherein the multiple target data include: operating mode, number of charging facilities, number of users, regional functions, traffic flow and number of parking spaces; using variance analysis to process the multiple target data to determine the importance of each target data; and determining the prediction indicator based on the importance of each target data.

[0010] Optionally, the back propagation neural network model includes: an input layer, a hidden layer and an output layer, wherein the number of nodes in the input layer is determined based on the number of prediction indicators, the number of nodes in the output layer is 1, and the number of nodes in the hidden layer is determined based on the number of nodes in the input layer and the number of nodes in the output layer using a trial and error method.

[0011] Optionally, generating training samples using the prediction indicators includes: normalizing the prediction indicators to obtain normalized data; and generating training samples based on the normalized data.

[0012] Optionally, the back propagation neural network model is trained using a Bayesian regularization algorithm and a prediction index to obtain a load forecasting model, including: inputting training samples into the back propagation neural network model to obtain prediction results of the training samples; obtaining a mean square error based on the original results of the training samples and the prediction results of the training samples; obtaining the mean sum of squares of the network parameters of the back propagation neural network model; obtaining a total error of the back propagation neural network model based on the mean square error and the mean sum of squares; and updating the regularization parameters of the back propagation neural network model based on the total error to obtain a load forecasting model.

[0013] Optionally, the regularization parameters of the back-propagation neural network model are updated based on the total error to obtain a load forecasting model, including: judging whether the total error is greater than a preset error and whether the number of training times reaches a preset number; if the total error is less than the preset error, or the number of training times reaches a preset number, then determining that the load forecasting model training is completed; if the total error is greater than the preset error and the number of training times does not reach a preset number, then updating the regularization parameters based on a preset learning rate.

[0014] Optionally, updating the regularization parameter based on a preset learning rate includes: obtaining the Jacobian matrix of the total error; solving the Jacobian matrix using the Gauss-Newton approximation method to obtain the number of effective parameters; and obtaining the updated regularization parameter based on the mean square error, the mean of the sum of squares, the number of effective parameters and the preset learning rate.

[0015] According to another aspect of an embodiment of the present invention, a charging load processing device is also provided, including: an acquisition module for acquiring historical charging data; a division module for dividing the historical charging data to obtain historical data corresponding to multiple grids; and a prediction module for using a load prediction model to predict the historical data corresponding to each grid to obtain the charging load of each grid within a preset time period.

[0016] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the above-mentioned charging load processing method.

[0017] According to another aspect of an embodiment of the present invention, a processor is further provided. The processor is configured to run a program, wherein the program executes the above-mentioned method for processing charging load when the program is run.

[0018] In an embodiment of the present invention, after acquiring historical charging data, the historical charging data can be divided to obtain historical data corresponding to multiple grids. The historical data corresponding to each grid can then be predicted using a load forecasting model to obtain the charging load for each grid within a preset time period, thereby achieving the purpose of predicting the charging load of the electric vehicle. It is readily apparent that because the historical charging data can be divided and the charging load within the preset time period can be predicted, the spatiotemporal distribution of the electric vehicle charging load can be determined, achieving the technical effect of improving prediction accuracy, thereby resolving the technical problem of low accuracy in determining charging load in related technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0020] Figure 1 is a flow chart of a method for processing a charging load according to an embodiment of the present invention;

[0021] Figure 2 is a schematic diagram of a charging load prediction value and a historical actual value according to an embodiment of the present invention;

[0022] Figure 3 is a schematic diagram of 24-hour charging load distribution without a charging facility grid according to an embodiment of the present invention;

[0023] Figure 4 is a heat map of charging load peaks of each grid within a day according to an embodiment of the present invention;

[0024] Figure 5is a heat map of the total charging capacity of electric vehicles in each grid in one day according to an embodiment of the present invention;

[0025] Figure 6 It is a heat map of the total amount of electric vehicle charging in each grid in one day based on the existing technology;

[0026] Figure 7 2 is a schematic diagram of a charging load processing device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. 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.

[0028] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0029] Example 1

[0030] According to an embodiment of the present invention, a method for processing a charging load is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0031] Figure 1 FIG. 1 is a flow chart of a method for processing a charging load according to an embodiment of the present invention. Figure 1 As shown, the method includes the following steps:

[0032] Step S102: Acquire historical charging data.

[0033] The historical charging data in the above steps can include charging pile asset information, charging transaction records, number of occupied parking spaces, regional functional proportions, traffic flow, and other data within a prediction area. The prediction area can be a street or a region, for example, Haidian District in Beijing.

[0034] Step S104 : dividing the historical charging data to obtain historical data corresponding to a plurality of grids.

[0035] In an optional embodiment, the historical charging data in an area may be divided according to the regional division method to obtain the historical charging data corresponding to each grid, that is, the historical data mentioned above.

[0036] Step S106: Use the load prediction model to predict the historical data corresponding to each grid to obtain the charging load of each grid within a preset time period.

[0037] The above-mentioned preset time period can be a time period determined according to actual prediction needs, for example, it can be one day, one week, etc. In the embodiment of the present invention, 24 hours a day is taken as an example.

[0038] In an optional embodiment, the processing flow of charging load prediction for electric vehicles is as follows: historical charging data can be divided as basic units for spatial prediction of charging load of electric vehicles; then the existing historical charging data is processed using a pre-trained load prediction model, and the historical charging data is used as input data. The output result of the model is the spatiotemporal distribution of the charging load of the electric vehicle.

[0039] Optionally, the load forecasting model can be obtained by training a back propagation neural network model using a Bayesian regularization algorithm.

[0040] The back propagation (BP) neural network is a global approximation network that achieves a nonlinear mapping from input space to output space through the composite mapping of several simple nonlinear processing units. However, traditional BP algorithms suffer from slow convergence and are prone to falling into local minima. Their generalization capabilities are also limited, affecting prediction accuracy. To address this issue, in embodiments of the present invention, a Bayesian regularization algorithm can be used to improve the generalization capabilities of the BP neural network model.

[0041] Optionally, the above-mentioned back propagation neural network model includes: an input layer, a hidden layer and an output layer, wherein the number of nodes in the input layer is determined based on the number of prediction indicators, the number of nodes in the output layer is 1, and the number of nodes in the hidden layer is determined based on the number of nodes in the input layer and the number of nodes in the output layer using a trial and error method.

[0042] The accuracy and training efficiency of the network model depend largely on the network structure and control parameters. The network structure can include network parameters, the number of neurons in each layer, etc.; the control parameters can refer to input variables, allowable error, learning rate, and initial weight values.

[0043] In the embodiment of the present invention, a three-layer BP neural network can be used to establish a prediction model, namely, an input layer, a hidden layer and an output layer. The function of each layer is the same as the existing one and will not be described in detail here.

[0044] The number of nodes in the input and output layers can be determined by actual processing requirements and is the number of variables actually input and output by the model. In this embodiment of the present invention, the number of input layer nodes is the number of input data. The output variable of this model is the electric vehicle charge level, so the number of output layer nodes is 1.

[0045] The number of nodes in the hidden layer is related to the complexity of the problem to be solved and the number of input and output variables, and has a significant impact on the performance of the network. If the number of hidden layer nodes is too small, the neural network will have difficulty handling complex problems and the modeling will be insufficient. If the number of hidden layer nodes is too large, not only may the neural network structure become complex, iteration time is long, and computational complexity is high, but it may also provide excess degrees of freedom to adapt to noise, causing the network to be overtrained and the error not necessarily to be optimal. In this embodiment of the present invention, the number of nodes in the hidden layer can be calculated according to the formula:

[0046]

[0047] Where J represents the number of hidden layer nodes, I and O represent the number of input and output layer nodes, respectively, and a is a constant between [1, 10]. The number of hidden layer nodes can be determined by trial and error. That is, multiple BP neural networks with the same conditions except the number of hidden layer nodes are constructed, and the same samples are used for training. The optimal number of hidden layer nodes is determined by comparing the training time and prediction accuracy of each network.

[0048] In an optional embodiment, since the charging capacity of grids without charging facilities is added to the original historical charging capacity, the historical charging data should be distributed to all grids in proportion to obtain the charging capacity of each grid in a day, that is:

[0049]

[0050]

[0051] Among them, m and k represent the number of grids with charging facilities and the number of grids without charging facilities in the target area, respectively. i represents the charge capacity of grid i in one day predicted by the neural network, C totalIndicates the total charging capacity of all grids in the area in one day, that is, the historical total charging capacity of grids containing charging facilities. represents the charge amount of grid i in one day.

[0052] Assume that changes in the number of electric vehicles are approximately represented by changes in the number of new energy vehicles, and that changes in the number of electric vehicles in the forecast region are approximately represented by changes in the number of electric vehicles nationwide. Assume that factors influencing the spatial distribution of electric vehicle charging capacity in the forecast region, such as regional functions, population size, and traffic flow, remain largely unchanged. Based on the previous assumptions, if the current number of new energy vehicles is c million and the number in the forecast year is f million, the electric vehicle charging capacity for each grid in the forecast year can be expressed as:

[0053]

[0054] in, It represents the predicted value of electric vehicle charging capacity when the number of new energy vehicles is f million. It represents the charging volume of electric vehicles when the number of new energy vehicles is c million, that is, the historical charging volume.

[0055] Through the above-described embodiments of the present invention, after acquiring historical charging data, the historical charging data can be divided to obtain historical data corresponding to multiple grids. The historical data corresponding to each grid can then be predicted using a load forecasting model to obtain the charging load of each grid within a preset time period, thereby achieving the purpose of predicting the charging load of the electric vehicle. It is easy to note that because the historical charging data can be divided and the charging load within the preset time period can be predicted, the spatiotemporal distribution of the electric vehicle charging load can be achieved, achieving the technical effect of improving prediction accuracy, thereby solving the technical problem of low accuracy in determining charging load in related technologies.

[0056] Optionally, in the above embodiment of the present invention, dividing the historical charging data to obtain historical data corresponding to multiple grids includes: obtaining a target area corresponding to the historical charging data; dividing the target area according to longitude and latitude information to obtain multiple grids; and allocating the historical charging data to multiple grids to obtain historical data corresponding to each grid.

[0057] In an optional embodiment, the target area can be divided into grids of equal size with a certain degree of accuracy, providing a basis for subsequent spatial prediction of electric vehicle charging demand and refined charging facility planning. For example, taking Haidian District as an example to verify and analyze the model algorithm, considering that traffic flow data and population data are calculated based on 1000m*1000m grids, and that grids with a side length of 1000m have a moderate area and are within the acceptable walking range for users, Haidian District can be divided into 1000m*1000m grids based on longitude and latitude coordinates.

[0058] It should be noted that the regional grid division of charging network planning can refer to the grid division method of parking lot planning, which will not be elaborated here.

[0059] Optionally, in the above embodiment of the present invention, the method also includes: obtaining prediction indicators corresponding to the charging load, wherein the prediction indicators are used to characterize factors affecting the charging load; constructing a back-propagation neural network model; generating training samples using the prediction indicators; and training the back-propagation neural network model using a Bayesian regularization algorithm and the training samples to obtain a load prediction model.

[0060] In an optional embodiment, MATLAB can be used to generate, train, and test a BP neural network model, and the trained load forecasting model can be used to predict the charging load of electric vehicles. The specific implementation process is as follows: factors that have a significant impact on the charging amount of electric vehicles can be selected as prediction indicators; the feedforwardnet function is used to set the number of hidden layer nodes and the training function to generate a BP neural network model; the train function is used to train the network to obtain the relationship between various prediction indicators and charging load; finally, the various prediction indicators of the area to be predicted are input into the trained network to output the corresponding 24-hour charging load prediction value.

[0061] Optionally, in the above embodiment of the present invention, obtaining the prediction index corresponding to the charging load includes: obtaining multiple target data corresponding to the charging load, wherein the multiple target data include: operating mode, number of charging facilities, number of users, regional functions, traffic flow and number of parking spaces; using variance analysis to process the multiple target data to determine the importance of each target data; and determining the prediction index based on the importance of each target data.

[0062] The above-mentioned ANOVA, also known as "analysis of variance," can be used to test the significance of the difference between two or more sample means. One-way ANOVA can be used to study whether different levels of a control variable have a significant impact on the observed variable.

[0063] In an optional embodiment, factors that may affect the charging demand of electric vehicles (specifically electric vehicles charged at the company's charging stations) can be collected: operating model (parking fees, electricity prices, etc.), the number of charging facilities operated by operators other than the power grid company, population size, regional functions (shopping malls, residential areas, work areas, etc.), traffic flow, parking lot space, etc. (i.e., the target data mentioned above). Variance analysis is used to identify the influencing characteristics of electric vehicle charging capacity and each factor, explore and extract their impact, and evaluate the importance of each influencing factor, so that factors with a significant impact on electric vehicle charging capacity can be used as predictive indicators.

[0064] Optionally, in the above embodiment of the present invention, generating training samples using prediction indicators includes: normalizing the prediction indicators to obtain normalized data; and generating training samples based on the normalized data.

[0065] In order to facilitate calculation, eliminate dimensionality effects, and improve prediction accuracy, it is necessary to preprocess the input data. Therefore, in practical applications, it is very necessary to normalize the input and output variables.

[0066] In an optional embodiment, the following data normalization formula may be used:

[0067]

[0068] Among them, P n is the normalized data, P is the original input data, and P min and P max are the minimum and maximum values ​​in the original input data set, respectively. After this normalization process, the data values ​​are transformed to between [-1, 1], which is conducive to the training of the neural network.

[0069] Optionally, in the above embodiment of the present invention, the back propagation neural network model is trained using the Bayesian regularization algorithm and prediction indicators to obtain a load forecasting model, including: inputting the training sample into the back propagation neural network model to obtain the prediction result of the training sample; obtaining the mean square error based on the original result of the training sample and the prediction result of the training sample; obtaining the mean sum of squares of the network parameters of the back propagation neural network model; obtaining the total error of the back propagation neural network model based on the mean square error and the mean sum of squares; and updating the regularization parameters of the back propagation neural network model based on the total error to obtain the load forecasting model.

[0070] The Bayesian regularization algorithm improves the generalization ability of the neural network by modifying its training performance function. Generally, the training performance function of the neural network uses the mean square error (mse), that is:

[0071]

[0072] Where n is the total number of samples, t i is the expected output, a i is the actual output of the network.

[0073] In the Bayesian regularization algorithm, the network performance function is changed to the following form:

[0074] msereg=α.msw+β.mse,

[0075]

[0076] Among them, msereg is the improved performance function, msw is the mean of the sum of squares of all network weights, α and β are regularization parameters, whose sizes have a significant impact on the network training effect, m is the total number of network weights, ω i is the network weight.

[0077] In an optional embodiment, the training steps of the above-mentioned Bayesian regularized BP neural network are as follows: determine the network structure, initialize the hyperparameters α and β, generally set to α=0 and β=1, and assign initial values ​​to the network parameters according to the prior distribution; use the BP algorithm to train the network to minimize the total error msereg; during the network training process, adaptively adjust the size of the regularization parameters α and β to achieve the optimal value.

[0078] The Bayesian method for regularizing neural networks is an iterative process. In each iterative process, the total error function changes with the changes in the hyperparameters α and β, the minimum point also changes, and the parameters of the network are constantly corrected. Ultimately, the total error function does not change significantly during the iterative process, which is also called convergence.

[0079] Optionally, in the above embodiment of the present invention, the regularization parameter of the back-propagation neural network model is updated based on the total error to obtain the load forecasting model, including: judging whether the total error is greater than the preset error and whether the number of training times reaches the preset number; if the total error is less than the preset error, or the number of training times reaches the preset number, it is determined that the load forecasting model training is completed; if the total error is greater than the preset error and the number of training times does not reach the preset number, the regularization parameter is updated based on the preset learning rate.

[0080] The above-mentioned preset error can be the minimum allowable error of the output, which can obtain better learning accuracy, but there is excessive fitting of the samples and the network's generalization ability is poor. In the above embodiment of the present invention, the preset error can be 0.001.

[0081] The above-mentioned preset number of times can be the maximum training number of the BP neural network model, which can prevent the network from falling into a local minimum during a certain training and running indefinitely without reaching the minimum error. In the above embodiment of the present invention, the preset number of times can be 1000.

[0082] In an optional embodiment, when the total error msereg is less than a preset error, training is determined to be complete and a load forecasting model is obtained. When the total error msereg is greater than the preset error, training is determined to be incomplete and further training is required. To prevent the network from running indefinitely, the total error msereg can also be determined while determining whether the number of training times has reached a preset number. If so, training is terminated regardless of the size of the total error msereg.

[0083] Optionally, in the above embodiment of the present invention, updating the regularization parameter based on the preset learning rate includes: obtaining the Jacobian matrix of the total error; solving the Jacobian matrix using the Gauss-Newton approximation method to obtain the number of valid parameters; and obtaining the updated regularization parameter based on the mean square error, the mean of the sum of squares, the number of valid parameters and the preset learning rate.

[0084] For the regularization parameter, if α is much larger than β, the training emphasizes the generalization performance of the network, but it is easy to cause underfitting. If α is much smaller than β, the training algorithm tends to reduce the error corresponding to the learning set network, which can easily cause the network to overfit. Conventional regularization methods usually find it difficult to determine the size of the regularization parameters α and β. However, the Bayesian theory can adaptively adjust the size of the regularization parameters during the network training process and make it optimal. The regularization parameter prediction formula of this algorithm is:

[0085]

[0086] Among them, γ is called the number of effective parameters:

[0087] γ=m-2αtr(H) -1 ,

[0088] Where H is the Hessian matrix of msereg:

[0089]

[0090] When performing optimization, it is necessary to calculate the Hessian matrix of msereg at its minimum point, which is computationally intensive. In order to increase the computational speed, the Gauss-Newton approximation method can be used to further simplify the Hessian matrix, yielding:

[0091] H≈2αI m +2βJ T J,

[0092] where J is the Jacobian matrix of the mse.

[0093] The learning rate reflects the change in weights during a network training cycle and is a crucial factor in optimization calculations. Each specific network has an appropriate learning rate. Choosing a suitable rate not only directly impacts convergence speed and training time, but also significantly improves control accuracy. The learning rate typically ranges from 0.01 to 0.1. In the embodiments of the present invention, a preset learning rate of 0.01 is used as an example, resulting in rapid convergence and short training times.

[0094] In an optional embodiment, the Gauss-Newton approximation method can be used to calculate the Hessian matrix and the number of effective parameters γ, and the above formula can be used in combination with a preset learning rate to calculate new estimates of the hyperparameters α and β; and updated through multiple trainings until the training is completed.

[0095] The following example analysis uses Beijing's Haidian District as a prediction area. First, the district is divided into 1000m x 1000m grids based on longitude and latitude. Of these, 52 grids contain charging facilities and 403 do not. Based on this grid division, data such as charging pile asset information, charging transaction records, number of occupied parking spaces, area function ratio, and traffic flow are imported for each grid.

[0096] (1) Charging transaction records

[0097] A charging transaction record data warehouse was built, and the asset information of charging piles in Beijing and a total of 2.5 million charging records from April to September 2018 were imported and saved. The charging pile asset information and charging records were connected according to the charging pile operation number to determine the spatial location of the charging transaction records. The dirty data was filtered out during data preprocessing, and the valid data scale in Haidian District was obtained after filtering was 420,000. The correct data was then formatted. Next, in the data analysis and statistics stage, the data was counted according to the time dimension of 24 hours a day, and according to the longitude and latitude grid in the spatial dimension. The data of the 183 days after the statistics were averaged to obtain the typical daily charging information statistical curve under different temporal and spatial distributions.

[0098] (2) Regional functions

[0099] A regional functional data warehouse was built to import and save 145,000 map POI data from Haidian District, Beijing. During data preprocessing, dirty data was filtered out, reducing the data size to 130,000 items. The correct data was then formatted. Next, during data analysis and statistics, statistics were performed spatially according to the longitude and latitude grids. The proportion of POI points representing different industries in each grid was then calculated, thereby obtaining the proportion of different functional areas in each grid.

[0100] (3) Traffic flow

[0101] A traffic flow data warehouse was built, and 1.68 million traffic flow data items covering nine days, from September 11 to 17, May 1, and October 1, 2017, were imported and stored in Beijing. During data preprocessing, dirty data was filtered out, resulting in 290,000 items of valid data in Haidian District. The correct data was then formatted. Next, in the data analysis and statistics phase, the data was statistically analyzed temporally based on typical weekdays, weekends, and holidays, and spatially based on latitude and longitude grids. A weighted average was then calculated based on the proportion of weekdays, weekends, and holidays in a year, yielding the typical daily traffic flow data for each grid.

[0102] The results of the correlation analysis of factors affecting electric vehicle charging demand are shown in Table 1. According to the analysis results, three factors that have a significant impact on electric vehicle charging demand, namely regional functions, the number of occupied parking spaces in the parking lot, and traffic flow, are used as predictive indicators of the electric vehicle charging load prediction model.

[0103] Table 1

[0104]

[0105] Based on the above analysis, this model ultimately extracted two predictive indicators: regional function and traffic flow. Regional functions are categorized into 19 types, including residential, commercial, and office areas. The proportion of each type within the grid serves as the model input, resulting in a total of 20 input layer nodes. The model's output is the 24-hour charging load of electric vehicles, resulting in a total of 24 output layer nodes. The number of hidden layer nodes was determined to be 12 through trial and error.

[0106] A BP neural network was constructed using MATLAB. The activation function between the input layer and the hidden layer was set to tansig, and the training function was the trainbr function of the Bayesian regularization algorithm. The learning rate was 0.01, the maximum number of training cycles was 1000, the training error was 0.01, and all other parameters were left at their default values. The prediction indicators for the grids containing charging facilities in the region and the historical average 24-hour charging load were used as the input and output data for the neural network training. The network was trained using the train function to determine the relationship between the prediction indicators and the charging load. Finally, the prediction indicators for the grids without charging facilities were input into the trained neural network to output the 24-hour charging load forecast for the grids without charging facilities.

[0107] The comparison between the charging load prediction value obtained by the BP neural network model and the historical real value is as follows: Figure 2 As shown in the figure, it can be seen that the error between the predicted value and the true value of the charging load is small, which proves that the BP neural network model can accurately predict the 24-hour distribution of electric vehicle charging load.

[0108] The 24-hour distribution of electric vehicle charging load in each grid can be obtained through the neural network prediction model, as shown in Table 2. Based on the 24-hour distribution of charging load, the peak charging load and total charging power of each grid in a day can be obtained, as shown in Tables 3 and 4. Figure 3 The 24-hour charging load distribution of three grids without charging facilities obtained by the prediction model is shown. The peak charging load of each grid in a day is as follows Figure 4 As shown, the total charging capacity in one day is Figure 5 shown.

[0109] Table 2

[0110]

[0111] Table 3

[0112]

[0113] Table 4

[0114]

[0115] The total charging capacity of electric vehicles in each grid in one day is obtained based on the historical charging data of the original charging facilities, as shown in Table 5. The heat map of the total charging capacity is shown in Figure 6 shown.

[0116] Table 5

[0117]

[0118] from Figure 6 As can be seen from the figure, the number of existing charging facilities is limited and only exists in a few grids. The historical charging data cannot accurately reflect the spatial distribution of electric vehicle charging demand. However, the electric vehicle charging load prediction method proposed in the above embodiment of the present invention can distribute the charging load of existing charging facilities to all grids, and obtain the 24-hour distribution of the charging load of each grid in a day and the total charging capacity in a day. Figure 5 and Figure 6 It can be found that Figure 5 The magnitude of the predicted value of mid-charge power is significantly smaller than Figure 6 The actual historical value of the mid-charge capacity is due to Figure 6 The historical charging capacity in is distributed to all grids to ensure that the total charging capacity remains unchanged.

[0119] Through the above scheme, on the basis of planning area grid division, the historical charging data of the grid containing charging facilities is reasonably distributed to each grid through the Bayesian regularized BP neural network algorithm, and an electric vehicle charging load prediction model is established with the grid as the basic unit, including the prediction of the number of electric vehicle charging times and the calculation of the average charging load of one charge, so as to solve the problem that the historical charging data cannot correctly reflect the actual charging demand of electric vehicles due to the limited number and unreasonable layout of existing charging facilities.

[0120] Example 2

[0121] According to an embodiment of the present invention, a charging load processing device is provided, which can execute the charging load processing method in the above embodiment. The specific implementation scheme and preferred application scenario are the same as those in the above embodiment and will not be repeated here.

[0122] Figure 7 is a schematic diagram of a charging load processing device according to an embodiment of the present invention, such as Figure 7 As shown, the device includes:

[0123] An acquisition module 72 is used to acquire historical charging data;

[0124] A division module 74 is used to divide the historical charging data to obtain historical data corresponding to multiple grids;

[0125] The prediction module 76 is used to use the load prediction model to predict the historical data corresponding to each grid to obtain the charging load of each grid within a preset time period.

[0126] Example 3

[0127] According to an embodiment of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium includes a stored program. When the program is executed, the device where the computer-readable storage medium is located is controlled to execute the above-mentioned charging load processing method.

[0128] Example 4

[0129] According to an embodiment of the present invention, a processor is provided, which is used to run a program, wherein the program executes the above-mentioned charging load processing method when running.

[0130] The serial numbers of the above embodiments of the present invention are for description only and do not represent the advantages or disadvantages of the embodiments.

[0131] In the above embodiments of the present invention, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0132] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0133] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0134] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0135] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk, etc. Various media that can store program codes.

[0136] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for processing a charging load, characterized in that: include: Acquire historical charging data, wherein the historical charging data includes at least one of the following: charging pile asset information, charging transaction records, number of used parking lots in the parking lot, regional function ratio, and traffic flow in the prediction area; Dividing the historical charging data to obtain historical data corresponding to a plurality of grids; Using the load forecasting model to forecast the historical data corresponding to each grid, to obtain the charging load of each grid within a preset time period; Partitioning the historical charging data to obtain historical data corresponding to a plurality of grids includes: acquiring a target area corresponding to the historical charging data; dividing the target area according to latitude and longitude information to obtain the plurality of grids, wherein the sizes of the plurality of grids are determined based on the statistical range size of traffic flow data and population data; allocating the historical charging data to the plurality of grids to obtain historical data corresponding to each grid; Wherein, the load forecasting model is obtained by training the back propagation neural network model using the Bayesian regularization algorithm; The method further includes: obtaining a prediction index corresponding to the charging load, wherein the prediction index is used to characterize factors affecting the charging load; constructing a back propagation neural network model; generating training samples using the prediction index; training the back propagation neural network model using a Bayesian regularization algorithm and the training samples to obtain the load prediction model; Among them, the back propagation neural network model is trained using the Bayesian regularization algorithm and the prediction index to obtain the load prediction model, including: inputting the training sample into the back propagation neural network model to obtain the prediction result of the training sample; obtaining the mean square error based on the original result of the training sample and the prediction result of the training sample; obtaining the mean sum of squares of the network parameters of the back propagation neural network model; obtaining the total error of the back propagation neural network model based on the mean square error and the mean sum of squares; and updating the regularization parameter of the back propagation neural network model based on the total error to obtain the load prediction model.

2. The method according to claim 1, characterized in that Obtaining the prediction index corresponding to the charging load includes: Acquire multiple target data corresponding to the charging load, wherein the multiple target data include: operation mode, number of charging facilities, number of users, regional function, traffic flow and number of parking spaces; Processing the plurality of target data using variance analysis to determine the importance of each target data; The prediction index is determined based on the importance of each target data.

3. The method according to claim 1, characterized in that The back propagation neural network model includes: an input layer, a hidden layer and an output layer, wherein the number of nodes in the input layer is determined based on the number of prediction indicators, the number of nodes in the output layer is 1, and the number of nodes in the hidden layer is determined based on the number of nodes in the input layer and the number of nodes in the output layer using a trial and error method.

4. The method according to claim 1, characterized in that: Generating training samples using the prediction index includes: Normalizing the prediction index to obtain normalized data; The training samples are generated based on the normalized data.

5. The method according to claim 1, characterized in that The regularization parameter of the back propagation neural network model is updated based on the total error to obtain the load forecasting model, including: Determine whether the total error is greater than a preset error and whether the number of training times reaches a preset number; If the total error is less than the preset error, or the number of training times reaches the preset number of times, it is determined that the load forecasting model training is completed; If the total error is greater than the preset error and the number of training times does not reach the preset number, the regularization parameter is updated based on a preset learning rate.

6. The method according to claim 5, characterized in that Updating the regularization parameter based on a preset learning rate includes: Obtaining the Jacobian matrix of the total error; The Jacobian matrix is ​​solved by using Gauss-Newton approximation method to obtain the number of effective parameters; Based on the mean square error, the mean of the sum of squares, the number of valid parameters and the preset learning rate, an updated regularization parameter is obtained.

7. A charging load processing device, characterized in that: include: An acquisition module is used to acquire historical charging data, wherein the historical charging data includes at least one of the following: charging pile asset information, charging transaction records, number of used parking lots in the predicted area, regional function ratio and traffic flow; A division module, used for dividing the historical charging data to obtain historical data corresponding to a plurality of grids; A prediction module, used to predict the historical data corresponding to each grid using a load prediction model to obtain the charging load of each grid within a preset time period; The division module is further used to obtain a target area corresponding to the historical charging data; divide the target area according to the latitude and longitude information to obtain the multiple grids, wherein the sizes of the multiple grids are determined based on the statistical range size of the traffic flow data and the population data; distribute the historical charging data to the multiple grids to obtain the historical data corresponding to each grid; Wherein, the load forecasting model is obtained by training the back propagation neural network model using the Bayesian regularization algorithm; The device is also used to obtain a prediction index corresponding to the charging load, wherein the prediction index is used to characterize factors affecting the charging load; construct a back propagation neural network model; generate training samples using the prediction index; train the back propagation neural network model using a Bayesian regularization algorithm and the training samples to obtain the load prediction model; Among them, the device is also used to input the training samples into the back propagation neural network model to obtain the prediction results of the training samples; obtain the mean square error based on the original results of the training samples and the prediction results of the training samples; obtain the mean sum of squares of the network parameters of the back propagation neural network model; obtain the total error of the back propagation neural network model based on the mean square error and the mean sum of squares; update the regularization parameters of the back propagation neural network model based on the total error to obtain the load forecasting model.

8. A computer-readable storage medium, characterized in that:The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the charging load processing method according to any one of claims 1 to 6.

9. A processor, characterized in that: The processor is used to run a program, wherein the program, when running, executes the charging load processing method described in any one of claims 1 to 6.