Properties of Energy-Price Forecasts for Scheduling

Wholesale electricity markets are becoming ubiquitous, offering consumers access to competitively-priced energy. The cost of energy is often correlated with its environmental impact; for example, environmentally sustainable forms of energy might benefit from subsidies, while the use of high-carbon sources might be discouraged through taxes or levies. Reacting to real-time electricity price fluctuations can lead to high cost savings, in particular for large energy consumers such as data centres or manufacturing plants. In this paper we focus on the challenge of day-ahead energy price prediction, using the Irish Single Electricity Market Operator (SEMO) as a case-study. We present techniques that significantly out-perform SEMO’s own prediction. We evaluate the energy savings that are possible in a production scheduling context, but show that better prediction does not necessarily yield energy-cost savings. We explore this issue further and characterize, and evaluate, important properties that an energy price predictor must have in order to give rise to significant scheduling-cost savings in practice.