Improving economic operation of a microgrid through expert behaviors and prediction intervals

The microgrid power scheduling (MPS) problem is heavily influenced by the intermittency of renewable energy sources (RES), leading to prediction uncertainties and suboptimal scheduling decisions. While deep reinforcement learning (DRL) offers potential, it faces challenges such as high dimensionality, complex constraints, and slow training. This paper proposes a novel framework that enhances economic performance and training efficiency by incorporating prediction intervals (PIs) from prediction error distributions and leveraging expert behaviors. The approach integrates PIs into the proximal policy optimization (PPO) process, enriching decision-making, and utilizes a pre-trained deep neural network (DNN) to provide prior expert knowledge. Experimental results show that the method outperforms existing approaches in reducing operational costs, improving efficiency, and mitigating prediction uncertainty. The model is robust to variations in predictor accuracy and extreme scenarios, with ablation studies confirming the effectiveness of the proposed improvements and exploring the optimal confidence levels of PIs.