Research On Theory And Method Of Model Predictive Control In Networked Systems

With the development of communication technology and computing speed,model predictive control algorithms are performed at the remote controller that connects with the local sensors and actuators via communication networks,forming the networked model predictive control.On the one hand,networked model predictive control inherits the advantages of conventional model predictive control,i.e.,it optimizes the control performance and is robust to uncertainties while explicitly considering the system con-straints,multiple-input-multiple-output property,and nonlinearity;on the other hand,its networked implementation overcomes the difficulty of the insufficient computing power of the local controller and exhibits advantages,such as security,reliability,flex-ibility,etc.,over the traditional point-to-point control system.Hence,networked model predictive control has received extensive attention from academic and industrial circles.Due to the unreliability of the communication network in data transmission,packet losses may occur,and limited communication resources impose the need to optimize its utilization.However,directly applying the conventional model predictive control al-gorithm cannot effectively handle these problems,which inevitably leads to the waste of communication resources,and the deterioration of the system performance or even instability.Therefore,there is an urgent need to design effective networked model pre-dictive control algorithms to address these problems.This thesis studies the discrete-time nonlinear constrained systems,considers three issues:packet losses,channel con-tention access as well as communication and computation resources limitation,and aims at proposing an effective networked model predictive control algorithm to mitigate the impact of the communication networks on the control system.The main content of this thesis mainly includes the following aspects:1.Considering the case of two-channel random packet losses,a packet-based model predictive control algorithm is proposed.By estimating the missing system state and compensating actively for the packet losses,the closed-loop random stability can be achieved.a theoretical mechanism for determining the packet length is further provided,which solves the problem that the prediction horizon cannot be determined because the number of consecutive packet losses may be unbounded.2.Considering the channel access problem for wireless cloud control systems with a special structure,a channel-aware dual scheduling strategy under the model predictive control framework is proposed.A distributed threshold strategy for the sensors and a prioritized threshold strategy for the controllers are designed.We also prove that all sensors with the strategy update mechanism will work at the Nash equilibrium point,the prioritized threshold strategy for the controllers is superior to the conventional independent identically distributed access strat-egy,which achieves the high-efficiency and energy-saving channel access for all systems.3.Considering the problem of high computational complexity when the maximum number of consecutive packet losses is large,a dual-mode adaptive horizon model predictive control algorithm under a dual-controller structure is proposed.By adaptively adjusting the prediction horizon and designing the switching mech-anism between the two controllers,the stability of the closed-loop system can be guaranteed,while the steady-state performance is improved and the computa-tional complexity is significantly reduced.4.Considering the problem of limited communication and computing resources,a dynamic event-triggered model predictive control algorithm and a time delay estimation-based self-triggered model predictive control algorithm are proposed.By the dynamic triggering condition to allow larger state prediction error and by designing the delay estimation and feedforward compensation strategy for the disturbance,then the requirements of the model predictive control on the num-ber of transmissions and the frequency of solving the optimization problems are reduced.In summary,this thesis systematically studies the challenges faced by networked model predictive control,and proposes corresponding solutions,which greatly pro-motes the development of networked model predictive control.