Proceedings of the PowerUp Conference

The rapid growth of distributed energy resources (DERs) is reshaping wholesale electricity markets, where aggre- gators coordinate large populations of prosumers, end users who own DERs and can both consume and produce electricity, to participate at scale. Existing models typically either optimize a single aggregator under exogenous prices or analyze a small number of strategic agents; neither captures the scale, structure, and price feedback that arise when large DER populations interact through wholesale market clearing. We address this gap with a hybrid mean field framework in which each aggregator coordinates a large prosumer population through its infinite agent limit, while a finite number of aggrega- tors participate in wholesale markets. Each aggregator maximizes the collective payoff of its prosumers, a cooperative structure rep- resented by mean field control. Across aggregators, interactions are non-cooperative; as a starting point, we model aggregators as non-strategic, price-responsive learners, while market prices remain endogenous through aggregate prosumer actions, leading to a mean field equilibrium. We establish conditions under which this equilibrium exists and is unique. To handle uncertainty in demand, renewable generation, and market prices, we design a two-phase reinforcement learning algorithm that enables aggregators to learn optimal control policies from price feedback. Numerical experiments on the Oahu power system show that the proposed framework reduces price volatility, flattens demand profiles, improves storage utilization, and lowers costs for both consumers and prosumers.