Research on PCA-LSTM-based Short-term Load Forecasting Method

Abstract

Accurate load forecasting not only makes the power company's power dispatch more reasonable, but also improves the economics and safety of grid operation. In order to make load forecasting more accurate, it is necessary to use a variety of factors that affect load changes, such as temperature and holidays, to build a huge data set, which will reduce the speed of calculations. In addition, deep learning has advantages over traditional prediction methods in time series prediction. Therefore, this paper adopts a load prediction method based on principal component analysis (PCA) and long short-term memory neural network (LSTM). First, build a multi-dimensional data set with a time series. Second, PCA is used to reduce the dimensionality of the data set, retaining the feature that the sum of the proportions of the k-dimensional principal components reaches or exceeds 85%. Thirdly, the LSTM was trained using the data set and a load prediction model was obtained, with a correlation coefficient of 94%. Finally, the experimental simulation is compared with the traditional method. Experimental results show that the method has higher accuracy and practicability.