Bidding Strategy for Periodic Double Auctions Using Monte Carlo Tree Search

Bidding strategies for Periodic Double Auctions (PDAs) are complicated because they need to predict and plan for future auctions, which may affect the bidding strategy in the current auction. We present a general bidding strategy for PDAs based on forecasting clearing prices and using Monte Carlos Tree Search (MCTS) to plan a bidding strategy across multiple time periods. We developed a controlled simulator by isolating Power Trading Agent Competition's wholesale market to evaluate bidding strategies in a realistic PDA energy market. We show that our MCTS bidding strategy is cost effective in buying energy compared to other baseline and state-ofthe-art strategies and it's performance improves with increasing number of MCTS simulations.