Research On Data-driven Control Method Of Linear Systems Based On Subspace Technique

With the continuous development of science and technology,the scale and complex-ity of the industrial systems are also increasing.The accurate mathematical model of the system is increasingly difficult to derive based on mechanism modeling method,and it usually takes considerable time and effort.However,there is a large number of off-line and online input and output(I/O)data in actual systems,which inspired people to think about how to use data to replace the mechanism model,and design the corresponding data-driven control methods.For multiple-input-multiple-output(MIMO)linear systems,the subspace system identification method(SIM)uses the input and output data of the sys-tem to derive the state space model of the multi-variable systems,by the methods in linear algebra such as the matrix decomposition.After twenty years of development,subspace identification technique has been successfully applied to fault diagnosis/identification(F-DI)and fault-tolerant control(FTC),predictive control,pattern recognition and other fields,and achieved satisfactory results.Based on the previous results,the data-driven control methods based on subspace technique are studied for linear discrete-time systems in this dissertation.The subspace-aided methods are combined with the fault-tolerant control,the predictive control and the covert attack strategy respectively,which lead to the corresponding data-driven control methods.The designed methods achieve satisfactory control effects,and reduce the com-putation load in the subspace predictive control(SPC)algorithms.The main results of this paper are theoretically proved,and the simulation experiments are carried out for the water tank flow control system,the continuous heating system and the irrigation canal.The results show the effectiveness of the proposed methods.The full text of this dissertation is divided into eight chapters,the main contents of each chapter are given as follows:Chapters 1-2 systematically introduce and analyze the background and development of the subspace technique and its related control methods.The preliminary knowledge and research methods related to this paper are also provided.Chapter 3 designs a data-driven residual generator based on subspace identification method with unknown system parameters for the discrete-time MIMO linear systems with noise.The residual generator produces a vector residual with the same dimension of the outputs.Compared with the conventional subspace-aided scalar residual,the vector residual is more suitable for FDI and FTC of MIMO systems.The designed subspace-aided data-driven residual generator is also the basis of the following designed data-driven FTC methods.Chapter 4 designs the data-driven fault compensation methods to achieve the inte-grated data-driven FTC methods based on the designed data-driven residual generator.The fault compensation mechanism in the Youla parameterization FTC configuration is designed based on the optimization function,adaptive tuning method and the correlation-based function,respectively.With the unknown system model parameters or the fault mode,the parameters of the fault compensator are tuned based on the online system data,which can ensure the control effect of the faulty system and achieve the FTC performance.The simulation results and the comparison with the existing results show the effects and advantages of the proposed methods.Chapter 5 develops an event-triggered SPC method based on the existing open-loop SPC method.With the partially unknown system model,by the input-to-state stability(ISS)theory,an event-triggered law based on the input error and the state-based function is designed.First,the corresponding design parameters to maintain the system stabili-ty are derived by the adaptive dynamic programming(ADP)technique,then the event-triggered law is designed.The traditional receding horizon optimization in predictive control is substituted by the designed event-triggered law,which considerably reduces the data computation and transmission load of the controller.The proposed method can en-sure the system stability and the optimization quality.Finally,the effect of the proposed method is illustrated by a two-tank flow system simulation.Chapter 6 develops an event-triggered closed-loop SPC method.With the unmea-surable system state data and the unknown system model,a data-driven state observer is first developed.Then based on the input error and the observer state,the corresponding event-triggered law is developed.The data computation and transmission load have been reduced by the developed event-triggered law,while maintains the system performance.Meanwhile,a fault detection scheme is proposed based on the designed event-triggered law.Finally,the simulation results illustrate its effect.Chapter 7 designs a data-driven covert attack strategy for the closed-loop cyber-physical systems.With the unknown parameters of the plant model and the controller,with only the intercepted transmission I/O data from the cyber-space,the attacker can deviate the real output of the system to his expected reference by modifying the transmis-sion data,while cannot be detected by the attack detector incorporated in the controller.Finally,the effect of the designed data-driven covert attack strategy is illustrated by the simulation of an irrigation canal system.Chapter 8 summarizes the results of the dissertation and point out the future research topics in relevance.