To address the uncertainties of renewable generations within a microgrid (MG), a deep learning-based prediction approach coupled with MG energy scheduling is introduced in this paper. First, a novel forecasting model of the dung beetle optimizer (DBO)-convolutional neural network (CNN)-bidirectional gated recurrent unit (BiGRU)-attention is developed for the predictions of photovoltaic (PV) power, wind turbine (WT) power, and load demand. Then, the sardine optimization algorithm is employed to optimize the scheduling of the park MG considering demand response (DR). Finally, the role of battery energy storage systems (BESSs) in the optimal scheduling of the park MG is analyzed. Results demonstrate the DBO-CNN-BiGRU-attention model achieves superior accuracy over conventional methods, reducing PV power prediction mean absolute percentage error (MAPE) by 77.07%, WT power MAPE by 71.17%, and load MAPE by 27.25% compared to the better-performing CNN-BiGRU-attention integrated model in the recent literature. After optimization, the MG with DR achieves effective peak-valley scheduling, leading to an 8.17% cost reduction. Additionally, the MG equipped with BESS plays a supportive role in regulation, resulting in a further 1.77% cost saving. This study confirms that the proposed forecasting and scheduling framework effectively enhances the economic efficiency, reliability, and sustainability of the MG.
Hou et al. (Sun,) studied this question.