Ultra-short term wholesale electricity price forecasting through deep learning

The increasing mix of renewable generation in the Australian National Electricity Market (NEM) electricity market presents economic challenges because generators are subject to extreme price drops in the short term dispatch cycles. The traditional forecasting models are inadequate in predicting such extreme price events due to the complexity in how numerous fac-tors impacting the short-term price. This paper's aim is to answer the degree of how deep learning can contribute to forecasting prices thresholds in a volatile real-time electricity price market. Additional research in this area is required post NEM transition to a unified 5 minute dispatch and bidding cycle. Effective short term forecasting can enable the transition from manual to automatic bidding fully optimising the revenue potential for generators. The algorithms presented in this paper present an ultra-short horizon prediction forecasting of 15 minutes into the future. The Queensland wholesale electricity market price is used to present the research findings. Our results show deep learning models can achieve significant performance gains.