Short-term Power Load Forecasting Based On Combined Model

Accurate power load forecasting is of great significance to the safe and stable operation of the power grid.With the rapid development of social economy,the accuracy of traditional load forecasting methods can not meet the requirements of modern power system for load forecasting.With the continuous development of artificial intelligence technology and the popularity of intelligent devices in power grid,massive load data provide data support for deep learning,which makes power load forecasting based on deep learning become a new research direction.Power load is random,but through the load curve,we can see that the load change is similar.Selecting important load influence factors as characteristics can effectively improve the accuracy of load forecasting.In this paper,through the correlation analysis of load and load influencing factors,historical load value,holiday,temperature and other characteristics are selected as the input values of the load forecasting model.Aiming at the nonlinear and sequential characteristics of load,a short-term power load forecasting model based on Inception network and long short-term memory network(Inception-LSTM)is proposed.In order to verify the validity of the InceptionLSTM model,based on the measured power load data set,the model is compared with four models: GRU,LSTM,CNN-GRU and CNN-LSTM.The results show that the model has higher prediction accuracy in the whole test set.In order to reduce the loss of key information of time series and extract the features of power load data more accurately,the Inception-LSTM model is improved in this paper.A short-term load forecasting model combining attention mechanism and improved Inception-LSTM(i Inception-i LSTM-AM)is proposed.The improved Inception-LSTM can not only extract multi-scale features,reduce the number of parameters,but also screen useful information more comprehensively and accurately,and the attention mechanism can enhance the weight of key information.The results of numerical examples show that the model has high prediction accuracy in the whole test set.