Data-Driven Optimal Structured Control for Unknown Symmetric Systems

This paper presents a structured feedback design approach for interconnected first-order systems with symmetric couplings and partially-unknown dynamics. Optimal structured state feedback control laws are commonly designed by solving one or more Lyapunov equations. Reinforcement learning, in conjunction with a preliminary data collecting phase, solves the Lyapunov equations without knowing the state matrix of the interconnected system. To find the optimal structured feedback matrix, a novel algorithm combines the data-driven approach with a gradient-based optimization technique. An application example validates the effectiveness of the proposed design procedure.