A data-driven method for microgrid bidding optimization in electricity market

Abstract

This paper presents a deep reinforcement learning based data-driven solution to the microgrid bidding in the electricity market considering offers for the reserve market. The framework, based on the Markov decision process, models the microgrid's participation in the electricity market at different stages, including bidding, market-clearing, and reserve activation. The problem is split into two stages: day-ahead submission and real-time market period, and the proposed method mainly focus on the first stage. The state information from state-space models of distributed energy resources serves as input for the policy network. A deep deterministic policy gradient is employed to train the network and produce a deterministic bidding strategy. The second stage can then adjust this strategy based on the results from the first stage. The method is validated with real-world microgrid systems and data from the Singapore spot market.