An Evaluation of Electricity Demand Forecasting Models for Smart Energy Management Systems

An Energy Management System (EMS) has been promising in recent years. The EMS can exploit the potential of electric devices by forecasting demand. Despite demand forecasting is often based on deep learning, it is required to implement on an edge device due to costly and privacy requirements. To explore a model suitable to implement on edge devices, this paper evaluates the state-of-the-art forecasting models: Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) in the accuracy, execution time, and memory usage. The results show that the GRU-based model reduces the execution time by 9.5% and the memory usage by 30.1% without loss of the accuracy.