Nonlinear Control Based on Local Linear Neuro-Fuzzy Models
This thesis presents a model-based nonlinear control concept as a comprehensive approach. The nonlinear process is modeled by a local linear neuro-fuzzy model which decomposes the complete operating range into regimes that can locally be described by linear submodels. By blending the actions of local subcontrollers, which are designed based on the individual submodels, an overall nonlinear controller is obtained. An on-line adaptation of the process model maintains high control performance also for processes with time-variant behavior and a supervisory level ensures a reliable adaptation even in phases of weak process excitation. Furthermore, the realization of the controller in look-up tables for control hardware with limited computational resources is described. The effectiveness and real-world applicability is shown by application to vehicle velocity control and temperature control of an industrial-scale heat exchanger. The local linear approaches are intuitive and easy-to-understand by the process engineer which increases the confidence in the process model as well as in the resulting nonlinear controller.