Research On Applicability Of LSTM And CNN To Short Term Load Forecasting In Distribution Station Area

The short-term load forecasting in the distribution station area plays a key role in the operation monitoring and accurate management of the distribution network.It is very difficult to accurately predict the short-term load of distribution station area due to the complex power consumption environment,large difference between distribution station area,poor data quality and many factors affecting load forecasting.In recent years,long short-term memory(LSTM)has been widely used by scholars at domestic and abroad because of its nonlinear and sequential characteristics.In order to improve the accuracy of short-term load forecasting in the distribution station area,this paper investigates the research and application status of load forecasting in the distribution station area.Based on LSTM and convolutional neural network(CNN),the applicability of the distribution station area is studied based on some distribution stations in a city in Gansu Province.Firstly,the paper analyzes the importance of regional short-term load forecasting in modern smart grid construction.Then introduces the related concepts,forecasting requirements,classification and forecasting steps of short-term load forecasting,and summarizes the research status of short-term load forecasting at domestic and abroad.Then discusses the difficulties of short-term load forecasting in distribution area and the significance of research on the applicability of LSTM network.Secondly,the paper introduces the data source and data structure and analyze the difference between short-term load forecasting and system level shortterm load forecasting of distribution station area.Obtain the actual data of a city in Gansu Province from state grid as an example,the paper analyzes the load characteristics of the distribution station area in this region in detail.Thirdly,the paper describes the establishment process of short-term load forecasting model of distribution station area,and introduces the basic theories of LSTM network and CNN network.It completes data preprocessing and construction of LSTM model and CNN model and CNN-LSTM hybrid model,and then gave the evaluation indexes of model forecasting effect.Finally,the paper evaluates the prediction results of LSTM models and CNNs model and CNN-LSTM hybrid models,and then analyzes the applicability of the models through the characteristics of distribution station area and load characteristics.The results show that the short-term load forecasting effect of LSTM network is better than that of CNN network and CNN-LSTM hybrid network.The results of applicability analysis show that it is necessary to classify the short-term load forecasting in the distribution area,and propose and verifie the basis of classifying the short-term load forecasting in the distribution area(Power consumption type,station area capacity,daily average load,daily minimum load and daily load rate).then the paper obtains the classified reference values of daily average load and daily minimum load and daily load rate.