Research of Networked Control Systems Based on New Smith Predictor
|School||Southwest Jiaotong University|
|Course||Proceedings of the|
|Keywords||wired ( wireless ) networked control systems multiloop networked control systems heterogeneous networked control systems hybrid networked control systems Smith predictor network delay data packet dropout stability|
In network control systems (NCS), undetermined network delays always exist in the data communication between sensor and controller, controller and actuator. The delays on size and property connect with network system structure, communication protocol and network load, and then it tend to be random, time varying and uncertain. The delay can cause significant deterioration of system performance, even make the system unstable. Therefore, it is especial important to analyze and research network delays.This dissertation proposes two novel methods of prediction compensation based on classical Smith predictor principle. Furthermore, the methods are extended through to some research fields, which include wired (wireless) networked control systems, wired (wireless) multiloop networked control systems, wired (wireless) heterogeneous networked control systems and hybrid networked control systems. The new Smith predictors are applicable to scopes that network delays are random, time varying and uncertain, and larger than one, even tens of sampling periods, simultaneity there are data packet dropouts in the loops.The characteristics of novel Smith predictors are shown as follows:(1) They realize double Smith dynamic prediction compensation controls on structure for delays of network and controlled plant. The novel Smith predictors are simple on structure, and they are easy for actualization based on intelligent nodes. Therefore, they have wide application prospects of the engineering.(2) The forward delays of network and controlled plant are removed from the closed loop and appear as gain blocks before the output, and the time-variant uncertain delay in the return path is totally eliminated from the system. Further, it can cancel effects of delays of network and controlled plant for system stability in the closed loop, and enhance control performance quality of entire system.(3) Because the delays on the return path can totally be eliminated, therefore the traffic on the return path does not need to be scheduled, and the output signal of sensor, whenever possible, can be transmitted back to remote controller node on line. On the one hand, this allows utilizing the capacity of the communication channel more effectively than static or dynamic scheduling could. On the other hand, increases system robustness when there are data packet dropouts on the return path of the NCS.(4) The novel Smith predictors are the real-time, on-line and dynamic predictors, and they do not include models of all network delays on actualization. Because information flow goes through true network delays of control process, therefore network delays do not need to be measured, identified or estimated on line. As a consequence, it reduces requirement of clock synchronization. Furthermore, it avoids estimate errors which are brought due to inaccurate model, and avoids node memory resources to be wasted when network delays are identified. Simultaneity, it avoids compensation errors, which are brought by network delays owing to vacancy-sampling and multi-sampling.(5) The novel Smith predictors can be implemented both controller and controlled plant sides of the network, or solely can also be installed at the controlled plant side of the network. The tradition scheme, which is only implemented in controller side of the network, is changed. Thus, it offers a flexible choice for implementing novel Smith predictors in engineering.(6) The results of simulation show that systems, based on novel Smith predictors, have stronger robustness, desirable dynamic performance and anti-jamming ability, when network delays are random, time-variant and uncertain, and possibly large compared to one, even tens sampling periods, and models of predictor and true controlled plant are unmatched, or model parameters have bigger errors, at the same time there are data packet dropouts in the loops. Therefore, the new control schemes of compensation are effective.