Consensus Transfer Q-Learning for Decentralized Generation Command Dispatch Based on Virtual Generation Tribe

This paper develops a consensus transfer Q-learning (CTQ) for decentralized generation command dispatch (GCD) of automatic generation control (AGC). A two-layer decentralized GCD based on virtual generation tribe (VGT) is adopted to resolve the course of dimension emerged in large-scale power systems. The leader of VGTs can calculate the generation command of each VGT through exchanging the Q-value matrices with its adjacent VGTs. In addition, a behavior transfer is employed into CTQ to exploit the prior knowledge of source tasks for a new optimization task according to their similarities, such that the algorithm convergence rate can be accelerated and the requirement of AGC period is satisfied. Case studies are carried out to evaluate the performance of CTQ for decentralized GCD on China southern power grid model.