چكيده به لاتين
In this thesis, an adaptive nonlinear model predictive control method is proposed to achieve offset-free behavior in the reference tracking and disturbance rejection. The uncertainties in the system parameters (i.e. internal disturbances) are considered as well as external disturbances. Since the neural networks are general approximators and can model any Lipschitz nonlinear systems using input-output data of the system, neural network model is utilized as the predictor model. Adaptive models can consider the system changes in each time step, therefore the adaptive neural network model is used and this helps to achieve offset-free tracking. To guarantee the closed-loop stability in the presence of disturbances, a constraint tightening approach based on the bounds of the disturbances and the Lipschitz constant of the adaptive model is proposed. This approach guarantees the Input-to-State Stability (ISS) of the closed-loop system in the presence of the step disturbances. To eliminate the effect of non-step disturbances such as ramp, a hybrid predictive control with feed-forward control based on disturbance observer is proposed. The important feature of this structure is that the design of predictive control for reference tracking can perform independent of the design of feed-forward control for disturbance rejection. However, in the presence of big disturbances, this structure may deteriorate the optimality of the control signal and/or not satisfy the constraints on the control signal. To solve this problem, estimation of disturbance is introduced to the neural network model as one of its inputs and the effect of disturbance is rejected in the closed-loop system via optimization problem of the predictive control. The effectiveness of the proposed methods is evaluated in the simulation and experimental studies and is compared with the recently reported methods in literature.
