Short-term renewable energy consumption and generation forecasting: A case study of Western Australia

Abstract

Peer-to-Peer (P2P) energy trading has gained much attention recently due to the advanced development of distributed energy resources. P2P enables prosumers to trade their surplus electricity and allows consumers to purchase affordable and locally produced renewable energy. Therefore, it is significant to develop solutions that are able to forecast energy consumption and generation toward better power management, thereby making renewable energy more accessible and empowering prosumers to make an informed decision on their energy management. In this paper, several models for forecasting short-term renewable energy consumption and generating are developed and discussed. Real-time energy datasets were collected from smart meters that were installed in residential premises in Western Australia. These datasets are collected from August 2018 to Apr 2019 at fine time resolution down to 5 s and comprise energy import from the grid, energy export to the grid, energy generation from installed rooftop PV, energy consumption in households, and outdoor temperature. Several models for forecasting short-term renewable energy consumption and generating are developed and discussed. The empirical results demonstrate the superiority of the optimised deep learning-based Long Term Short Memory (LSTM) model in forecasting both energy consumption and generation and outperforms the baseline model as well as the alternative classical and machine learning methods by a substantial margin.