Jour Fixe - Arbitrage of Energy Storage With Reinforcement Learning in the Intraday Electricity Market

Arbitrage of Energy Storage With Reinforcement Learning in the Intraday Electricity Market (joint with Anton Motornenko)

In 2020, 45.4% of the electricity in Germany had been produced from highly volatile renewable sources, and the share is expected to increase.  Energy storage technologies could enhance the flexibility of the energy infrastructure. However, energy storage has a high cost. Within this project, an agent-based algorithm that determines the optimal buying and selling strategy of storage is developed. In our research,  we try to make use of energy price fluctuations by charging the storage at times when there is an excess of energy on the market and therefore the prices are low, and selling the energy at peak load times when the prices are high. The energy price has systematic trends which cannot be understood by conventional approaches. We propose a deep reinforcement learning approach to understand the price formation depending on time and thus generate profits using energy arbitrage.