Bi-LSTM-based load forecasting for power grid

Abstract

Abstract It is very important to optimize the power supply and alleviate the conflict of electric load demand, so an accurate and reasonable power load prediction technology has been the focus of research. An evolutional LSTM named Bi-LSTM processes two independent input sequences with two opposite time directions, which aims to improve the efficiency of load forecasting for the power grid. The four main processes are illustrated, which include preprocessing the historical electrical load data, setting the parameters of the model, training the model, and testing the model. Comprising with typical models of RNN, LSTM, and GRU, the performance is verified, and the comparison result shows that it is better in several typical evaluation metrics. The proposed Bi-LSTM model effectively leverages historical data to forecast the imminent state of the power grid, providing valuable insights for electrical load management.