Real-time power scheduling for an isolated microgrid with renewable energy and energy storage system via a supervised-learning-based strategy

In the future of decentralized energy systems, isolated microgrids integrated with renewable energy and energy storage systems (ESS) have emerged as critical solutions for areas beyond conventional grid connectivity. Optimal power scheduling is essential for the efficient operation, cost efficiency, and stability of isolated microgrids. Therefore, this study proposes a new supervised learning (SL) strategy for real-time optimal power scheduling of an isolated microgrid. The proposed approach is three-fold: First, a deterministic mixed-integer linear programming (MILP) model is established for the optimal power scheduling problem of an isolated microgrid to minimize operational costs. By harnessing historical data, this optimization model is solved by a dedicated MILP solver to obtain an expert dataset of optimal decisions in the isolated microgrids. Second, an SL strategy is deployed to learn and mimic optimal ESS charging/discharging decisions by training a dense residual neural network (ResNetD) on the obtained expert dataset. Finally, the well-trained ResNetD model is applied to provide near-optimal power scheduling decisions based on real-time information. The performance of the proposed method is validated using a comprehensive set of test scenarios and compared with the base case, myopic policy, and other well-known deep reinforcement learning. The results reveal that the SL method reduces operating costs by 5.95 % and the output of the diesel engine generator by 12.67 % compared to the base case. Moreover, the SL method provides high-quality solutions that closely approximate the ideal results with an average performance gap of 0.37 %. Therefore, the proposed method demonstrates its robust adaptability to the real-time conditions of an isolated microgrid environment.