Suboptimal Nonlinear Model Predictive Control Based on Genetic Algorithm

This paper proposes a suboptimal nonlinear model predictive control (NMPC) algorithm based on Genetic Algorithm (GA). A nonlinear programming problem is solved online in NMPC. GA has been successfully applied to nonlinear programming problems where other decent-based methods have often failed. In this paper GA is used to optimize the control sequence per sampling time. In order to reduce the computational load on nonlinear model based predictive controllers, the key idea in this scheme is to seek a feasible descent solution rather than the optimal solution at each sampling time. The feasible solution decreases the cost function rather than minimizing the cost function. This strategy considerably reduces the online computation time at each search step, while maintaining good overall performance via iterative GA search. The low-complexity feature of the proposed algorithm makes it attractive for practical control systems with stringent requirements on fast sampling and large prediction horizon. A proof of nominal stability of the closed-loop system is also given in the paper. Computer simulations on continuous stirred tank reactor (CSTR) and experimental tests on the coupled-tank system are carried out to corroborate the effectiveness of the proposed predictive control technique.