A Novel Deep Learning Approach for Short and Medium-Term Electrical Load Forecasting Based on Pooling LSTM-CNN Model

Abstract

The power system is moving towards a more smart, intelligent and interactive framework. With the transition of power systems, there is also a maximum demand for renewable power generation and load forecasting. Load forecasting plays a vital and key role in the power grid planning, maintenance, and operation for electric energy customers. Accurate and timely load forecasting helps electric power suppliers to assist load scheduling and minimize the waste of electric power. Since the behavior and nature of electric load time series are non-linear because of the irregular change and an increase in the electric power demand with an increasing population, a neural network is one of the best candidates for constructing the non-linear behavior models used for forecasting. We proposed a deep learning-based approach that uses pooling long short-term memory (LSTM) based convolutional neural network to get the forecasting models for short- and medium-term electric load forecasting. Our method resolves the non-linearity and uncertainty issues by using many linear and non-linear methods to select the best features, time series models and several layers for pooling the LSTM model. The experimental results show that our method achieves more accurate results in short-term and medium-term load forecasting on metrics such as least Mean Absolute Error (MAE) and Root Mean Square Error (RMSE).