Deep Meta-Reinforcement Learning-Based Data-Driven Active Fault Tolerance Load Frequency Control for Islanded Microgrids Considering Internet of Things
Jiawen Li, Yuanyuan Cheng
Abstract
This article proposes a data-driven active fault tolerance load frequency control (DDAFT-LFC) method for island microgrids. It is demonstrated that DDAFT-LFC can effectively prevent frequency control loss caused by sudden off-grid faults affecting regulating units and emergency faults in microgrids, as well as reduce the total generation cost and improve the frequency stability of an island microgrid relying on a high renewable energy input. The method adopts an active fault tolerance strategy which can adapt to the complex stochastic environment, thus improving the robustness of the LFC strategy and achieving multiobjective optimization of dynamic performance and economic efficiency via regulation of the controller’s control strategy. In order to realize active fault-tolerance, this article proposes a deep meta-deterministic policy gradient (DMDPG) algorithm, which employs a meta-reinforcement learning method to help the agent to perform multitask collaborative learning in order to adapt to changes in microgrid environment parameters caused by changes in the unit combinations. The method is validated in an experiment of the island microgrid of the China Southern Power Grid (CSG).