Heuristic decision rules for short-term trading of renewable energy with co-located energy storage. (July 2017)

• Record Type: Journal Article
• Authors: Hassler, Michael
• Is Part Of: Computers & operations research. Volume 83(2017)
• Page Start: 199
• Page End: 213
• Publication Date: 2017-07
• Subjects: Electricity -- Renewable energy -- Energy storage -- Markov Decision Process -- Decision rule; Operations research -- Periodicals; Electronic digital computers -- Periodicals; 004.05
• DOI: 10.1016/j.cor.2016.12.027
• Languages: English
• ISSNs: 0305-0548

Abstract

Highlights: A model for trading renewable energy co-located with energy storage is presented. Easily graspable decision rules yielding good heuristic solutions are proposed. The model explicitly considers the time lag between trade and delivery of energy. Stochastic production and market prices are represented by ARIMA processes. The producer may also undertake price arbitrage by purchasing energy on the market. Abstract: In the last decade, the share of renewable energy sources in the energy mix has risen significantly in many countries, and the large-scale integration of these intermittent energy sources constitutes a major challenge to the power grid. A crucial building block of a successful transformation of today's energy systems is the use of energy storage, either co-located with renewable energy sources or on a grid-level. To this end, we present a model on the basis of a Markov Decision Process for the short-term trading of intermittent energy production co-located with energy storage. The model explicitly considers the time lag between trade and delivery of energy, which is characteristic for energy markets. Our storage representation includes asymmetrical conversion losses, asymmetrical power, and self-discharge. Stochastic production and market prices are represented by ARIMA processes, and the producer may also undertake price arbitrage by purchasing energy on the market when prices are comparatively low. Regarding the solution of our model, we develop several intuitive and easily interpretable decision rules that can be readily applied in practice. An extensive numerical study, based on real-world data, confirms the excellent performance of these rules in comparison to a sophisticated Approximate Dynamic Programming algorithm adapted from literature.