Electricity load forecasting based on long and short-term memory neural network

Abstract

Efficient and accurate power load forecasting is extremely significant to the safety and stability of power systems and dispatching operations. In order to fully exploit the influence of multivariate data information on load forecasting accuracy, in this paper, a short-term load prediction method based on long-term short-term memory neural network is proposed. First, to reduce the influence of diverse magnitudes on the accuracy of load prediction, the relevant factors affecting the load change are normalized; then, the trained LSTM-based load prediction model is tested; finally, different methods are used to realize the short-term prediction of electric load and the predictive precision is compared. Taking the actual load data of a certain area for instance, the results suggest that the proposed method could more accurately track the future trend of electric load than other algorithms.