State Estimation And Controller Design For Networked Systems Based On Moving Horizon Optimization Strategy

The traditional point-to-point control strategy is the one, where the field real-timedata from the controlled plant can be centrally transmitted through the cable to the centralcontrol room and then the control variable computed by the central processing unit can besent to the field. With the rapid development of communication, computer and controltechnology, the traditional point-to-point control structure can not meet the performancerequirements of control systems. Meanwhile, a fundamental change of the traditionalpoint-to-point control structure has taken place. Specifically, control systems are made upof sensors, controllers, actuators and other components each of which can communicatewith others by the shared network, and then Networked Control Systems (NCSs)generated. NCSs have now been one of the hot topics in the control theory research.Though there have been lots of useful results on this issue, lots of problems remain to beunsolved and bring forward new challenges and opportunities to the existing controlmethods. Considering the difference between the moving horizon optimization and thefeedback strategy on the state estimation and the controller design, the estimation andcontrol problem of NCSs with packet dropouts, data quantization and communicationconstraints can be solved based on the performance index of the moving horizon window,which can obtain a good performance. Therefore, based on the moving horizonoptimization strategy, the state estimation and controller design problem of NCSs arerespectively investigated in this thesis. Furthermore, the relations among systemparameters and performances, and parameters characterizing the packet dropouts, dataquantization and communication constraints are established respectively. The maincontents are as follows.1) Based on the moving horizon optimization strategy, a moving horizon estimation(MHE) method fully utilizing the available input and output information of the system dynamics in sliding window is proposed for NCSs to overcome the influence of the datapacket dropouts which can be modeled by a stochastic variable satisfying the Bernoullibinary distribution. This estimation method can make good use of the additionalinformation about state, control input and output variables shown in the form of inequalityconstraints to enhance the accuracy and rationality of the state estimation. Compared withthe existing other state estimation methods, a distinct character of the proposed estimatoris that when the current measurement is lost during transmission, a batch of the mostrecent measurements instead of the previous value or the current value directly set to zerowill be used to the design of the proposed estimator, which can improve the estimationprecision. Moreover, a sufficient condition in the form of the inequality is presented toguarantee the convergence of the estimation performance. Finally, the efficiency of theproposed MHE can be illustrated by some simulation examples.2) Based on the moving horizon optimization strategy, a new networked predictivecontrol method is proposed for networked control systems under the network environmentwhere the number of the consecutive data packet dropouts is bounded arbitrarily.Considering the fact that the data in network can be packed into one packet and then canbe transmitted at each instant, the model of the NCSs with bounded packet dropouts isbuilt. Based on the proposed model, the networked predictive controller can be designedto stabilize the system, and a good control performance can also be obtained. Differentfrom other control methods, the proposed method has the unique merits which can predictthe future control action of the system according to the future dynamics of the controlledsystem and can accomplish the network transmission of the data packet including apredictive control sequence, but not only a current predictive control action. When thecontrol packet is lost, the corresponding predictive control can be picked up from thepredictive control sequence conserved in the buffer, and then acts on the controlled system,which can overcome the effect of the packet dropouts on the controlled system. Finally, aninverted pendulum system can be used to verify the superiority of the proposed method.3) Based on the moving horizon optimization strategy, a new robust model predictivecontrol method is proposed to deal with the control problem of NCSs with dataquantization through the communication network between controller and actuator. Firstly,considering the effect of the data quantization in the controller-actuator channel and applying the logarithmic quantizers to describe this kind of quantization, the model of theNCSs with control input quantization is established based on the sector bound approach.Secondly, relying on the proposed model, the stability of the networked control systemswith quantization is studied which can be converted into the robust control problem of thelinear uncertain systems, and then the robust predictive controller to asymptoticallystabilize the system is designed. Furthermore, the stability conditions of NCSs shown interms of linear matrix inequalities (LMIs) are obtained. On the basis of guaranteeing thestability of NCSs and obtaining a certain control performance, the coarsest quantizationdensity can be also derived by using a cone complementary linearization method. Finally,two simulation examples are given to show the validity of the proposed method.4) Based on the moving horizon optimization strategy, a novel dynamical schedulingapproach has been firstly proposed to handle the communication scheduling problem ofNCSs with communication constraints which means that because of the limited networkbandwidth only some sensors can be allowed to transmit the partial measured outputs tothe remote estimator through the communication network at each sample instant. By thisway, the state estimator can still have a good estimation performance in the case of thelimited network resources. Firstly, defining the communication sequence to explain thecommunication constraints, the communication constraints can be changed into anequality constraint with a logical variable taking the value of zero and one, and then alinear time-invariant system with communication constraints can be converted into a lineartime-varying system with an equality constraint, and then the moving horizon schedulingmethod is proposed based on a new quadratic performance criterion includingcommunication cost and estimation performance penalties. Consequently, by onlinesolving the optimization problem of the mixed integer quadratic programming, thecommunication scheduling sequence can be derived. Secondly, a moving horizon stateestimation method has been also suggested to estimate the unavailable states of NCSsincorporating state inequality constraints. It is further analyzed that sufficient conditionsare presented for the boundness on the square norm of the estimation error. Finally, apractical experiment on a two-tank liquid-level system is given to demonstrate theadvantages of the proposed scheduling method.