Optimal Scheduling of Microgrids Using Dual Stage Attention-Based Deep Learning Forecasting Model

The increasing integration of renewable energy sources and loads into microgrid systems presents significant challenges due to their randomness and uncertainty. Accurate forecasting of these renewable energy outputs and electrical loads is crucial for effective microgrid design and optimization. This paper proposes an advanced deep learning architecture that combines Convolutional Neural Networks (CNN) and Gated Recurrent Units (GRU) to enhance the accuracy of microgrid behavior forecasting by extracting temporal features from energy datasets. Additionally, a dual-stage attention (DSA) mechanism is incorporated to optimize the forecasting process by highlighting critical features in time-series data. An optimal microgrid operation scheduling model, incorporating solar panels, wind turbines, micro gas turbines, and battery energy storage systems (BESS), is developed to minimize total costs using Mixed-Integer Linear Programming. This CNN-GRU-DSA based model allows the microgrid to flexibly switch between grid-connected and island modes based on short-term load forecasts. The proposed model also identifies the optimal capacity and location of renewable energy resources in the microgrid. It provides optimal dispatch strategies for multi-energy systems, ensuring the flexibility, reliability, and economic efficiency of the microgrid.