A novel supplementary learning controller and its application in power systems

GUO Wen-tao; MEI Sheng-wei; LIU Feng

doi:10.7641/CTA.2014.41029

Abstract：“Cybernetics”by Norbert Wiener and“Engineering Cybernetics”by Tsien Hsue-shen have laid a solid foundation for classical control theory. Based on the classical control theory, modern control theory makes advance in optimizing performance and handling uncertainties. In this paper, a supplementary learning controller is proposed as a method to incorporate the classical control theory and the modern control theory, which adds a supplementary learning controller based on approximate dynamic programming on an existing classical controller. Policy iteration approximated dynamic programming algorithm and least squares method are employed as the training algorithm, which enjoys the policy-search efficiency of policy iteration and the data-utilization efficiency of least squares. Action dependent cost function is introduced to make the online learning model-free. By using such a supplementary learning controller, the prior knowledge of the existing classical controller can be fully utilized. On the other hand, the supplementary learning controller can optimize the performance of the closed-loop system. Furthermore, stability and convergence of the proposed method is proved rigorously. Simulation studies on reactive power control of doubly-fed induction generators (DFIGs) based wind farm validate the proposed supplementary learning controller.