Spatial Load Forecastinh Method Using GAN In Data-scarce Scenarios

Spatial load forecasting(SLF)is the basic work of urban power grid planning and construction,considering the size and spatial distribution of future power load to guide and plan the future development of the power grid,and its prediction results will directly affect the power grid The scientificity and rationality of the power equipment configuration in the plan.Therefore,thinking about how to use more reasonable methods to improve the accuracy of space load forecasting results,how to solve problems in the face of the failure of traditional space load forecasting methods,so as to obtain reasonable prediction results,has important guidance for urban power grid planning significance.This article first analyzes and introduces the background of space load forecasting research and domestic and foreign research status in detail,summarizes and classifies the existing space load forecasting methods,and briefly analyzes the application conditions and advantages and disadvantages of each method;Reasonably analyze the spatial electrical load characteristics of historical electrical load data;establish the geographic information system(GIS)used in spatial load forecasting.So far,of the more than one hundred spatial load prediction methods that have been proposed,most of them can only get good results when applied in scenarios with sufficient historical load data,while the prediction accuracy when applied in scenarios with scarce historical load data.Will drop significantly,and even fail.This paper analyzes the data characteristics under the scenario of lack of historical load data,and divides the lack of historical load data into three scenarios.Aiming at the third type of historical load data shortage scenario,a spatial load prediction method based on GAN and RCGAN is proposed.From the time scale analysis,this spatial load forecasting method belongs to the mid-to long-term time-scale forecasting method in order to provide a more reasonable basis for grid planning.This method first establishes a power geographic information system,and generates Class I cells and Class II cells.Then it builds a data generation model based on the original GAN.Based on very limited historical load data,it generates a sufficient number of "?"Cell-like" historical load data to achieve the purpose of data enhancement.By improving the generator loss function of the original GAN model to generate a more reasonable data set,it laid the foundation for the establishment of subsequent prediction models.A spatial load forecasting model based on RCGAN is constructed.Considering the essential problem of the prediction model,the loss function of the more suitable Huber Loss improved prediction model generator is selected,and the spatial load prediction is realized by using the generated "class ? cell" historical load data and the RCGAN model with determined parameters.Finally,engineering examples and spatial error analysis show that the method is correct and effective.