AI News

Power system operation is a safety-critical sequential decision-making problem, but many existing reinforcement-learning environments are narrow and rely on CPU-based simulation, limiting large-scale evaluation. The paper introduces PowerZooJax, a JAX-based benchmark suite designed to keep the full training and evaluation loop on the GPU. It includes five constrained Markov decision process tasks covering generation, transmission, distribution, distributed energy resources, and data-center microgrids. The authors rewrite power flow, economic dispatch, market clearing, and device dynamics as JAX computation graphs. Experiments report substantial speedups over CPU-based simulations. The suite also supports standardized evaluation using policy returns, safety violations, and out-of-distribution stress conditions. PowerZooJax is released as an open-source benchmark for reinforcement-learning research in power-system operation.