Research On Nonlinear System Control Based On Neural Network And U Model

In industrial processes, most of the control processes contain nonlinearity of certaindegree. For nonlinear systems, the traditional linear control theory has difficulty inmeeting the practical needs, so the nonlinear factors must be taken into account duringthe system design. When the nonlinear system is controlled, the main difficulty is tohandle the nonlinear term. The complex nonlinear systems can be controlled effectivelyby carrying out different control strategies to handle the nonlinear term. The maincontributions of the thesis are as follows:The stability and chaos synchronization control of time-delay cellular neuralnetwork are investigated. By constructing reasonable Lyapunov function and utilizing thelinear matrix inequality technique, the global asymptotic stability, exponential stabilityand robust stability of cellular neural networks are investigated. New delay-dependentstability criteria are derived. An exponential synchronization control algorithm isproposed for chaotic cellular neural network, and the proof is based on the Lyapunovstability theory.The control of time-delay nonlinear system with triangle structure is studied. Thebackstepping adaptive neural network controller is designed for pure-feedback andstrict-feedback nonlinear time-delay systems, the nonlinear part of system is toapproximated by neural network which can be realized through large scale integratedcircuit. The adaptive state feedback control method is proposed by using a Lyapunovfunction. The tracking error of closed-loop system is shown to be stable and uniformlybounded based on the Lyapunov theory.The control of nonlinear dynamic polynomial model is investigated. The nonlineardynamic polynomial model is transformed into the form of the U-model, which bridgeslinear control design method and nonlinear dynamic systems. So the design method ofthe linear control system can be used in controlling the nonlinear control systems easily.Based on the U-model, a generalized predictive control algorithm is proposed, and theoutput of controller can be derived through the Newton iterative method. The algorithm can match system model more precisely, compared to the other performance index basedon the linear approximation model. The predictive control based on U-model has theminimum performance index. Finally, simulation results illustrate the effectiveness andfeasibility of the algorithm.The control of nonlinear dynamic rational polynomial model is investigated. Basedon improvement of the U-model, the analytical pole placement controller is designed, thesolution of the controller output is converted into resolving a polynomial equation incurrent control term which significantly reduces the difficuly encountered in nonlinearcontrol system synthesis and computations. The method of pole placement control basedon U-model establishes a foundation for applying the other linear controller designmethod in control of nonlinear systems.