Learning to Clear Peer-to-Peer Energy Markets
Peer-to-peer energy markets are a promising paradigm for integrating distributed energy resources. A large body of work models their clearing as distributed optimization or generalized Nash equilibrium (GNE) problems which let prosumers coordinate trades and local resources without a central operator. Yet, the computational and communication burden of distributed optimization and GNE-seeking algorithms poses a major challenge to deployment at scale. We propose a learning-to-optimize framework for decentralized markets that uses historical data to accelerate clearing algorithms without additional communication. To enforce convergence by design, we train summable augmentations to a baseline algorithm, using graph neural networks to learn permutation-equivariant updates for markets of varying size and topology. We instantiate the framework for computing a variational GNE in the market model of Le Cadre et al. (2020) with a distributed Douglas-Rachford algorithm, but the idea extends to other models and base algorithms. Numerical experiments show improved convergence and illustrate the robustness of our method to out-of-distribution instances.
Authors
- Giuseppe Belgioioso (ORCID: https://orcid.org/0000-0002-8892-1466)
- Andrea Martin
- Lisa Innocenti Uccini
Publication Details
- Journal
- arXiv (Cornell University)
- Published
- 2026-11-16
- DOI
- https://doi.org/10.48550/arxiv.2611.21179
- Primary Topic
- Smart Grid Energy Management
- Type
- preprint