Distributed Power System Simulation and Monitoring Supported by Cloud Technology : A grid optimization scheduling method based on MADRL
Hui Luo, Jie Zhang, Yufeng Hu, Jingyue Zhang, Haiping Guo, Libin Huang · 2024
This study explores the optimization of Distributed Generation Systems (DGS) within a Multi-Agent Deep Reinforcement Learning (MADRL) framework. The focus is on transforming the overall optimization goal of the system into a global reward function and designing local rewards adapted to heterogeneous agents. Using the IEEE 33 node system as a case study, the impact of power aggregation and load management from photovoltaic systems, energy storage systems, electric vehicle charging facilities, and HVAC systems is analyzed. The results demonstrate that the proposed edge-cloud collaborative framework optimizes resource allocation and scheduling. The cloud platform is capable of integrating data from diverse sources and managing it in a unified manner, thereby coordinating the operations of multiple heterogeneous agents. The scalability and flexibility of cloud technology allow for the dynamic adjustment of computational resources according to demand, ensuring global optimization and the judicious allocation of resources.