Short-term Load Forecasting of Central China based on DPSO-LSTM

Accurate short-term load forecasting for a specific area plays an important role in power system operation. However, the load exhibits high volatility, making it challenging. In light of the superiority of deep learning models, this present study goes further by focusing on the optimization of the input feature structure of the well-established long short-term memory (LSTM) model for short-term electricity load. Discrete particle swarm optimization (DPSO) algorithm is proposed to determine the feature structure of the LSTM model to achieve better predictability of short-term load. The real data from the area of Central China including four provinces was used to justify the proposed DPSO-LSTM model by comparing with the state-of-art counterparts including standard LSTM and recurrent neural networks (RNN) models. The results show that the DPSO-LSTM model exhibits strong predictive power on both daily load and peak daily load data sets, demonstrating the effectiveness of DPSO-LSTM in short-term load forecasting.