Data-Assimilation-Assisted Reinforcement Learning for Power Grid Control under Load UncertaintyNew

This paper investigates how load information quality affects sequential decision-making in power grid control under transmission constraints, time-varying demand, and renewable variability. The research employs a customized transmission network simulator inspired by Grid2Op based on a DC power flow model, incorporating stochastic spatial loads, temporally correlated renewable-like generation, energy storage, and line parameters. The study explores the combination of data assimilation techniques with reinforcement learning to improve control performance under imperfect load information.