Benchmarking Reinforcement Learning Algorithms on Island Microgrid Energy Management

Siyue Zhang, Srinivasan Nandakumar, Quanbiao Pan, Ezekiel Yang, Romain Migné, Lalitha Subramanian

Abstract: Tremendous increase of electricity demand of coun-tries in Southeast Asia has been observed over the past few years and expected to sustain in the next decade. A large number of inhabited archipelagos in the Insular region eagerly await electrification. Compared to the traditional solutions relying on distributed diesel generators and grid extension, island micro-grids demonstrate advantages of cost effectiveness and reduced environmental impact. One of the key challenges in operating island microgrid is to optimize the power dispatch to achieve the lowest operation cost while ensuring the grid reliability. In this paper, we have adopted both deterministic cycle charging algorithm and state-of-the-art Deep Reinforcement Learning algorithms in an islanded microgrid model built based on data from Indonesia. Once the reinforcement learning models are implemented and tuned using the latest Stable Baselines 3 library, we benchmark the models' performance, which is measured by the total number of blackout occurred and the cumulative fuel consumption over a year. The result quantifies the improvement and highlights the efficiency of energy management optimization achieved by applying Deep Reinforcement Learning algorithms for islanded microgrid application.