Modelling and Predictive control of a Nonlinear System Using Local Model Network

In this work the optimization of the local model network structure and predictive control that utilize the local model network to predict the future response of a plant is studied. The main idea is based on development of the local linear models for the whole operating range of the controlled process. The local models are identified from measured data using clustering and local least squares method. The nonlinear plant is then approximated by a set of locally valid sub-models, which are smoothly connected using the validity function. The manipulated variable adjustments are computed through optimization at each sampling interval. The parameters of the plant at each sampling point are derived from the linearization of local model network. The proposed identification and control method is illustrated by the simulation study on the nonlinear process.