Noncooperative Control of Energy Communities through Learning-based Response Dynamics

With the growing availability of data, learning-based distributed energy management is emerging as a viable and efficient alternative to traditional model-based schemes. In this context, we propose a novel game-theoretic learning-based method for the distributed control of energy communities. In particular, we consider a community that includes several prosumers equipped with a renewable energy source and an energy storage system. The scheduling of energy activities of all prosumers is formulated as a noncooperative game. Nevertheless, unlike the state-of-the-art, where an optimization problem is typically defined to model the behavior of each prosumer, we approximate each prosumer response strategy using a neural network. We propose a distributed algorithm based on the well-known Banach-Picard iteration to efficiently seek for an equilibrium of the game. Lastly, the convergence and effectiveness of the proposed approach are validated through numerical simulations under different realistic scenarios.