A Lyapunov function based optimal hybrid power system controller for improved transient stability

In this paper, an intelligent power system stabilizer based on a stable and optimal hybrid learning-based adaptive control architecture is proposed which is evolved from approximate dynamic programming technique. The hybrid learning controller is based on two algorithms: (a) A reinforcement learning neural network controller implemented by adaptive critic design (ACD) and (b) A model reference adaptive controller (MRAC). The proposed method uses a novel value priority architecture for integrating these two algorithms. The designed ACD approximates nonlinear functional dynamics of the power system and interacts with the MRAC that adapts based on parametric changes. The value priority scheme developed based on a softmax network generates a hybrid control signal derived from a Lyapunov stability function that evaluates identification, performance, and stability of each controller. The overall hybrid control algorithm ensures convergence to an optimal control solution without using an explicit model of the system and at the same time ensures overall system stability. Theoretical results are validated by simulation studies for electric-generator stabilization on a 2-area 5-generator power system and as a wide-area controller on IEEE 68-bus system. (C) 2016 Elsevier B.V. All rights reserved.