Trading to Protect the Grid: A Reinforcement Learning-Based Bidding Strategy for Day-Ahead Flexibility Markets

With rising electricity demand and grid investments lagging behind, flexibilitymarkets offer distribution system operators (DSO) a cost-effective alternative togrid expansion. As of 2026, flexibility markets still remain mostly pilot projectsin Europe and, to the best of our knowledege, current research has not provideda complete answer to how a DSO should place bids in a day-ahead flexibilitymarket while taking into account the full complexity of such a market. We pro-pose a bidding strategy for a DSO based on reinforcement learning (RL) that aimsto reduce the aging (degradation) of grid assets due to high load. By using atwo-level Markov decision process (MDP) and simulated intra-day rollouts, ourmethod learns to be robust to forecast errors and adjusts its bids accordingly. Theproposed method is evaluated on a stress-test on a Swiss distribution transformeragainst an optimization-based baseline under perfect foresight (5% higher aging)and realistic forecasts (24% less aging). The results showcase the potential oflearning-based bidding strategies against standard optimization methods and pro-vide a framework for DSOs to robustly bid in future day-ahead flexibility markets.