Optimal microgrid operation considering accurate battery degradation—World model-based reinforcement learning approach

Modern microgrids incorporate self-generating energy devices like Photovoltaic (PV) systems along with Energy Storage Systems (ESS). Effective scheduling of energy demand, generation, and storage in these systems is crucial for economic efficiency. However, optimizing a microgrid is challenging due to its complexity and uncertainties. Reinforcement learning (RL) has proven effective in making robust, sequential decisions under the uncertainties inherent in microgrid optimization. However, many RL methods frequently fail to take advantage of utilizing future information in decision-making. This study proposes an World Model-based RL under the World Model framework that leverages prediction models for future PV generation and power demand, enabling the RL agent to make more informed and effective decisions in PV-ESS operations. It is essential to consider battery degradation expenses to ensure cost-effective microgrid operation. Although ESS can lower electricity costs by storing and shifting energy, managing the trade-off with degradation costs is crucial in the optimization process. Our approach utilized an accurate degradation model suitable for RL, ensuring a comprehensive cost objective. Simulations conducted within a university campus setting showed notable cost reductions ranging from 0.3% to 25.3% compared to benchmark models.