Unsupervised Deep Architecture for Forecast of a Tropical Electricity Load

Research on electricity load forecasting has been well circulated in journals. However, this was not particularly well done in the tropics. After all, forecasting electricity loads has been established to vary along climatic regions owing to different weather conditions, with the consequential effect of contrasting load requirements. This characteristic change has triggered the purport of this study for a while. Since the study began, as this is only an extension of previously done works by this team, deep architectures have been found more reliable than the classical models for load forecasting. As a result, in this study, an unsupervised deep learning architecture namely Stacked Autoencoder (SAE) was built for and applied on a 3-year historic electricity consumption and meteorological data for day-ahead prediction of electricity consumption of a tropical region. Consequently, the developed unsupervised (SAE) model demonstrated good results on both validation and test data, and its prediction cost was very minimal.