Short-term Load Forecasting Based On Weighted Transfer Learning Of Holidays

With the rapid and stable development of the national economy,power system load presents a significant growth year after year.However,the traditional extensive scheduling mode of power causes the problem that the shortage of electricity when it is busy and enormous waste when it is leisure because the energy storage capacity is extremely small and costly.It becomes the common goal of the world that builds a smart grid with the characterized:interactive,self-healing,security,economic,clean,energy-saving and efficient.Load forecasting has always been the essential part of the smart grids.The results of the short-term load forecasting(STLF)are also important to the other basic functions such as energy-saving and cost-reducing and security early warning in the smart grid.However,in load forecasting,the holidays data are sparse and the regular between holidays and non-holiday is completely different,so it is difficult for the model to accurately forecast holidays.Therefore,the paper aims to increase the data of holidays of source cities to solve the problem,thus improve the forecasting performance.First,we purely transfer the holidays loads of the source cities,enrich the information of holiday and abstract new holiday features,which improve the forecasting accuracy of the holidays and avoid the influence on the forecasting performance on non-holiday.The specific work is as follows:first,we abstract a new holidays feature by analyzing the variation of holidays load.Then we calculate the incidence degree of source cities and target city.Last,we abstract the holidays information by the algorithm of source cities selection.By these steps,the problems that the holidays data is sparse can be solved.Further more,we propose a weighted transfer learning approach based on holidays,which gives different weights to different source load data according to the relevance influence level to the holidays load in target city from each training data in target city or source cities,thus reflecting the essential attribute of different load data and training a more accurate forecasting model.The specific work is as follows:the data of the source cities are the external data,which is not higher reliability than those of the target city,therefore we allocate different weights for two parts data.Weights allocation treats the target city as a reference(weights are set to 1),then compute the Pearson correlation coefficient between source cities and target city as weights of source cities.Finally,to the negative transfer phenomenon of some cities,we solve it using the improved algorithms.The algorithm can adjust the weights of diff-distribution data which reduce the negative transfer.The real data set,which includes more than a dozen cities from the Guangdong Province in China,and comparison models were used to evaluate the performance of the method.The comparative experimental results show that the prediction method proposed in the MAPE and MASE performance model are decreased 46.7%and 37.8%than the SYR model.