Long Short-Term Memory Network and Support Vector Regression for Electrical Load Forecasting

Combination of long short-term memory (LSTM) network with support vector regression (SVR) is proposed for short-term electrical load forecasting in this paper. Three different cases are introduced and assessed for load forecasting. Although, LSTM can be lonely used for both feature extraction and forecasting, but, the features extracted by LSTM can be used as input of a SVR for forecasting. The experimental results show that using the load features and the temperature features extracted by individual LSTM networks beside the original values of load and temperature measurements of 24 recent hours provide good forecasting results.