Bi-layer energy management framework based on neural networks for multi-microgrid systems
摘要
This paper presents a bi-layer framework developed using a new class of discrete-time recurrent neural networks for optimal energy management in a multi-microgrid (MMG) system. The first layer involves a neural network-based energy management system (NN-EMS) designed to perform multi-objective energy management in each microgrid within the MMG system. The optimal output solutions obtained by the NN-EMS serve as references for dispatchable distributed energy resources and the amount of power traded with the grid in each respective microgrid, aiming to minimize operating and environmental costs simultaneously. In the second layer, a central energy management system (CEMS) is introduced as a supervisory entity. The optimal output solutions obtained by the CEMS are employed as references for scheduling internal power trading between microgrids with surplus and deficit power, aiming to minimize costs associated with power trading between the MMG system and the main grid, as well as power losses during transactions. A case study under two different scenarios is conducted on a grid-connected MMG system using real-world data and the real-time simulator Opal-RT® OP5600 to verify the effectiveness of the proposed solution in comparison with the well-known non-dominated sorting genetic algorithm II (NSGA-II) technique. The results demonstrate that the proposed methodology improved overall MMG system costs by