Robust Energy Management System With Safe Reinforcement Learning Using Short-Horizon Forecasts

In this letter, we study an energy management system (EMS) with an inconsistent energy supply that aims to minimize energy costs while avoiding failing to satisfy energy demands. To this end, we propose a robust EMS algorithm based on safe reinforcement learning which can effectively exploit short-horizon forecasts on system uncertainties. We show via experimental results using real datasets that our robust EMS algorithm outperforms other state-of-the-art algorithms in terms of both robustness and cost-efficiency thanks to its capability of utilizing short-horizon forecasts.

Automated summary

In this paper, we propose a robust EMS algorithm based on safe reinforcement learning to minimize energy costs and ensure energy demands are met in the presence of inconsistent energy supply. The algorithm effectively utilizes short-horizon forecasts on system uncertainties and outperforms other state-of-the-art algorithms in terms of both robustness and cost-efficiency, as demonstrated through experiments using real datasets.