Simultaneous Policy Iteration for Decentralized Control of Multi-Machine Power Systems

This brief considers a decentralized control problem of interconnected multi-machine power systems with asymmetric input constraints. Initially, such an input-constrained decentralized control problem is converted into a group of unconstrained optimal control problems via preassigning modified nonquadratic cost functions for nominal subsystems. Then, under the framework of adaptive dynamic programming, a simultaneous policy iteration (PI) algorithm is proposed to solve the Hamilton-Jacobi-Bellman equations related to these optimal control problems. To implement the simultaneous PI algorithm, actor and critic neural networks (NNs) are separately constructed to approximate the control policy and the cost function. After that, the least square method and the Monte Carlo integration technique are combined to simultaneously determine the actor and critic NNs’ weights. Finally, simulations of a two-machine power system are conducted to validate the theoretical claims.