Efficient Short-Term Electricity Load Forecasting for Effective Energy Management

• A two-phased framework for short-term electricity forecasting. • A novel data preprocessing strategy is introduced to refine the raw data. • An end-to-end hybrid residual CNN with stacked LSTM model for electricity forecasting. • A detailed ablation study is conducted to ensure the effectiveness of proposed model. • The performance is compared with state-of-the-art models over several benchmark datasets. Short-term electrical energy load forecasting is one of the most significant problems associated with energy management for smart grids, which aims to optimize the operational strategies of buildings. Electricity forecasting models are considered a key aspect of the provision of better electricity management and reductions in energy consumption. This motivates the researchers to develop efficient electricity load forecasting (ELF) models, based on historical nonlinear and high volatile data, which require appropriate forecasting strategies. Therefore, in this article, we present an innovative two-phase framework for short-term ELF. The first phase is dedicated to data cleansing, in which pre-processing strategies are applied to raw data. In the second phase, a deep residual Convolutional Neural Network (CNN) is designed to extract the important features from the refined data. To the best of our knowledge, this is the first work to introduce a deep CNN architecture for the extraction of spatial features from electricity data. The output of the residual CNN network is forwarded to a stacked Long Short-Term Memory (LSTM) network to learn the temporal information of the electricity data. The proposed model is then evaluated using the Individual-Household-Electric-Power-Consumption (IHEPC) and Pennsylvania–New Jersey–Maryland (PJM) datasets. The results reveal a significant reduction in the error rate over the IHEPC dataset in terms of Mean-Absolute-Error (MAE) (15.65%), Mean-Square-Error (MSE) (8.77%), and Root-Mean-Square-Error (RMSE) (14.85%) and over the PJM dataset our method reduced RMSE up to 3.4% as compared to baseline models i.e., linear regression, LSTM, and Gated Recurrent Unit (GRU). Furthermore, we performed several experiments with CNN, LSTM, and GRU models and evaluated it with additional Coefficient of Variation of the RMSE (CV-RMSE) metrics, which proves the effectiveness of our model for short-term load forecasting.