Online adaptive dynamic programming control of urban open-channel flow system

Abstract

An online adaptive dynamic programming (ADP) design is proposed for the control of urban open-channel flow systems, whose topographic parameters are not assumed to be accessible. According to the Saint-Venant continuity equation, a simplified model is firstly built. Subsequently, an adaptive dynamic programming control scheme is implemented, whose purpose is to track the desired water level, as well as to decrease the control cost. The design contains two RBF neural networks (NN). One action NN is employed to generate the control signal. Another critic NN is designed to approximate the long-term cost function. The two NNs are coordinated to approach an optimal solution. Finally, the adaptive dynamic programming controller is validated in rainstorm situation in simulation environment. The results demonstrate that the designed scheme outperforms the traditional PID counterpart.