Dynamic state-predictive control for a remote control system with large delay fluctuation
Hiroshi Yoshida, Taichi Kumagai, Kozo Satoda · 2018
Under the concept of Internet of Things (IoT), various machines are beginning to be controlled remotely through the Internet and wireless IP networks. However, such an IP packet exchange network brings about large time delay fluctuation due to the nature of its packet queue. The delay is called dead time in control theory and makes it difficult to stabilize remote control. State-predictive control is effective for control systems containing dead time. The state-predictive control predicts the current state from the delayed states observed, and calculates a control input on the basis of the predicted state. However, the conventional state-predictive controls cannot deal with the remote control systems through IP networks that have large delay fluctuation. This is because their prediction horizon (supposed dead time) is preliminarily set to a fixed value, which makes the predicted state different from the actual one. In this paper, we propose a novel state-predictive control that adjusts the prediction horizon dynamically on the basis of a real-time estimation of the time-varying delay with a Kalman filter. We conduct a simulation of an inverted pendulum system using actual delay data obtained on a WLAN network the delay of which fluctuated up to 200 ms and show that our control method succeeds in inverting the pendulum with a probability of above 90% where a conventional state-predictive control succeeds only 60-70%. The results demonstrate that our method improves the stability and robustness of remote control through a network that has large delay fluctuation.