Networked predictive control of systems with data dropouts

This paper presents a practical algorithm to design networked control systems able to cope with high data dropout rates. The algorithm is intended for application in packet based networks protocols (Ethernet-like) where data packets typically content large data fields. The key concept is using such packets to transmit not only the current control signal, but predictions on a finite horizon without significantly increasing traffic load. Thus, predictive control is used together with buffered actuators and a state estimator to compensate for eventual packet dropouts. Additionally, some ideas are proposed to decrease traffic load, limiting packet size and media access frequency. Simulation results on the control of a three-tank system are given to illustrate the effectiveness of the method.