Data-driven Estimation And Control In Networked Control Systems

With the wide application of network, networked control systems (NCSs) has arised inmany felds. However, when the sensor, actuator and controller exchange data in NCSs, itis inevitable to encounter situations such as quantization, data dropout, transmission delay,etc., which will result in degradation of the system performance and increasing complexityin the investigation. Model-based research on NCSs has many challenges and difcultiesespecially when the model of the system is difcult or impossible to obtain. Besides, thereare plenty of data in NCSs. Hence, when the mathematical model is unknown or hardto obtain, data-based method is particularly important. Thus, data-driven research onNCSs has become a frontier and hot topic in the feld of automation. This thesis mainlyfocuses on the problem of data-driven estimation and control design for linear discrete-timesystems in the network environment, where the following network-induced factors, such asquantized data, outlier, missing data, delayed data, are considered. The main contributionsof this thesis are summarized as follows:1. The problem of data-driven fltering for linear systems with bounded noise by us-ing quantized measurements is studied. Based on the set membership fltering, anoptimal data-driven worst-case flter based on quantized measurements has been pro-posed with a certain worst-case performance defned as the worst-case norm of theestimation error over all noise signals and possible quantization intervals. Further-more, an l2-l∞almost-optimal flter design algorithm is presented. By exploitingnon-negative scalar technique, the optimization problem with infnite constraints isturned into linear programming with fnite constraints, where the computation com-plexity has been reduced. Besides, the relationship between the estimation error andthe quantizer thresholds is discussed. It has been shown that, the quantizer thresholdcan be chosen appropriately to balance the estimation precision and the communica-tion cost. The simulation examples are fnally given to show the efectiveness of ourproposed algorithm. 2. The problem of parameter estimation is investigated for a class of SISO discrete-timelinear systems in the presence of outliers. To facilitate analysis, it is assumed thatthe unknown parameters are constant and the evolution of the systems is subjec-t to some unknown but bounded noises. To solve such a problem, a recursive setmembership estimation algorithm based on outer bounding ellipsoid is proposed andemployed. Besides, by utilizing the geometric properties of the outer bounding el-lipsoid, the data validation is processed and outliers are set aside. The redundantdata have been removed from the parameter estimation by minimizing the volumeof outer bounding ellipsoid. Furthermore, a modifed set membership parameter es-timation scheme, which combines the outer bounding ellipsoid estimation algorithmand least square algorithm, is constructed based on the quantized measurements toimprove the performance of estimator. The efectiveness of the algorithms is verifedby performing some numerical simulations.3. Data-driven robust flter design for networked systems with missing data is consid-ered. For linear systems with bounded noise, a direct networked data-driven flter iscomposed of an output predictor and a direct data-driven flter is proposed. First,an on-line output predictor is designed to give an accurate estimation of output.Then, an optimal worst-case flter is designed based only on the received observationand output prediction. Since fnding an optimal data-driven flter is a very hardand sometimes unpractical, an almost-optimal flter is thus presented and the up-per bound of worst-case fltering error is obtained. Finally, the simulation examplesshow the efectiveness of our proposed algorithm, which demonstrates the designeddata-driven flter in this paper appears to be very promising.4. A data-driven networked predictive control scheme is proposed for MIMO NCSs withrandom network delays existing in both forward (from controller to actuator) channeland feedback (from sensor to controller) channel. The networked predictive controlconsists of a control prediction generator and a network delay compensator. Thecontrol prediction generator provides a set of future control predictions to make the closed-loop system achieve the desired control performance and the network delaycompensator eliminates the efects of the network transmission delay. Simulationsand experiments are carried on the ball and beam system. The results show theproposed data-driven networked control scheme is efective and superior.