Research On The Internal Model Control Of Nonlinear System Using Neural Networks

In this dissertation, the internal model control (IMC) of nonlinear control system is studied. The main parts are concluded as follow:(1) Nonlinear system identification using dynamical neural network (DNN) and fuzzy neural network (FNN) are summarized. The new DNN based on parallel mode is proposed. Even in the case there exists modeling error and bounded disturbance, Lyapunov theory based adaptive learning algorithms can ensures that the identification error and parameters of the DNN are stable in the sense of uniform ultimate boundedness. A novel learning algorithm is proposed for T-S FNN. Architecture of FNN and initial value of premise part parameters are obtained by fuzzy clustering algorithm firstly and then parameters of premise part and consequent part are optimized by combination of LMF and LSM algorithms.(2) A novel design procedure of IMC for affine nonlinear system modeling by dynamical neural networks is proposed. First, differential geometric feedback linearization method is applied to SISO affine nonlinear systems which their states are measurable, control law of IMC is constructed by analytical inverse model from DNN, so analysis and design method of linear IMC can extend to nonlinear system. Robustness and steady performance of closed-loop system are analyzed existing modeling error; affection of control saturation is also analyzed. Tracking error feedback loop is added to IMC in order to eliminate affection of control saturation. Second, under input-output decoupling linearization, research results are extended to MIMO affine nonlinear systems which their states are measurable. Robustness and steady performance of close loop system are analyzed existing modeling error and remaining couple action is analyzed. At last, if state of system can not be measured, a new method of identification by DNN based on slide mode state estimator is proposed to make the theory into integrity.(3) Nonlinear IMC based on FNN for open-loop stable nonlinear system is proposed. Theory analysis point out that FNN model can view as a sort of special tune varying system so inverse model can be obtained directly. By treating modeling error as structured uncertainty, robust stability analysis using method is presented and filter design is directed by ft synthesis method, so parameter of filter can tuning online to ensure the stable of close-loop system.(4) Based on inverse system method for nonlinear system, nonlinear IMC using FNN is proposed. FNN inverse model can transform the controlled nonlinear system into pseudo-linear system, so analysis and design method of linear IMC can extend to nonlinear system easily. The steady performance and robustness of closed-loop system is analyzed existing modeling error and affection of remaining couple action is analyzed especially. When modeling error does not satisfy linear growth condition, Popov stability theory is applied to analyze robustness of closed-loop system.