Short-term load forecasting for microgrid energy management system using hybrid SPM-LSTM

Abstract

Load forecasting in power microgrids and load management systems is still a challenge and needs an accurate method. Although in recent years, short-term load forecasting is done by statistical or learning algorithms. There are still two unsolvable challenges in the conventional data-driven-based prediction methods. The first challenge is that extracting the correlation between metrological and load data still cannot be taken full advantage of, and the other is that extracting patterns independently of the fixed pattern length is not supported. To address these challenges, a fixed SPM-LSTM approach is proposed. SPM for discrete data and LSTM for continuous data. The proposed model uses past consumption, temperature, humidity, wind speed, and weather data. A sequential pattern mining algorithm is used to extract sequential patterns that are independent of the fixed pattern length of correlated data between load and metrological data and can predict only the range of future load. Then, an LSTM network is used for exact load forecasting. It was tested using real data from the consumption and generation data retrieved from the ENTSOE project and its performance was compared with other forecasting methods. Results have shown that the proposed approach with the correlation squared (R2) of 0.951 in the best situation outperformed other methods like LSTM, LSTM-ANN, and CNN-GA. Also, the proposed method reduced training time by one-fifth against others.