Short-term Load Forecasting of Long-short Term Memory Neural Network Based on Genetic Algorithm

Abstract

Accurate load forecasting is conducive to the reasonable arrangement of power grid dispatching plans. Traditional load forecasting methods cannot handle the time series and nonlinear characteristics of load well. Long-short term memory (LSTM) neural networks can record long-term and short-term information, which can effectively solve this kind of problem. But the parameters of LSTM network are difficult to determine. For this reason, this paper proposes a long-short term neural network based on genetic algorithm. The learning rate and iteration number of the LSTM network are used as chromosomes, and the genes are continuously selected, crossed, and mutated to obtain more good genes. Comparing this method with the standard LSTM network, the simulation results show that the LSTM network using genetic algorithm for parameter optimization improves the prediction accuracy of the standard LSTM network by 63%.