Data-Driven Fully Distributed Load Frequency Control for an IoT-Based Interconnected Grid Considering a Performance-Based Frequency Regulation Market

To effectively prevent frequency collapse and waste of regulatory resources caused by poorly coordinated communication or failure of centralized controllers and distributors in the performance-based frequency regulation market, this paper presents a data-driven fully distributed load frequency control (DDFD-LFC) method in which each regulatory unit operates as an independent agent. Based on ubiquitous intelligence, frequency and tie-line power data are collected locally by the units to regulate the output so that LFC does not rely on communication and thus is capable of fully distributed control. To achieve effective coordination, a multiagent deep meta deterministic policy gradient (MA-DMDPG) algorithm is introduced, which combines the meta-learning method and the centralized training architecture to enable each agent to perform multitask cooperative learning. Using a Guangdong two-area LFC simulation model, this paper shows that the proposed method reduces the regulation mileage payment and area control error and therefore offers grid-side benefits.