Nonlinear Process Identification and Predictive Control by the Weighted Sum of Multi-Model Outputs

Most industrial processes are nonlinear. In such a case only a nonlinear model valid for the whole working area can ensure a good controller design. The nonlinear process is approximated by a multi-model consisting of the intelligent combination of some linear sub-models. As a very practical way the following identification strategy was used: independent model parameter estimation in the different working points and the calculation of the global valid model output as the weighted sum of the sub-models. As a weighting function the Gaussian function is used. The parameters of the Gaussian function were chosen either without or with optimization of the identification cost function. The global valid nonlinear model was used for model based predictive control. A heat exchanger example illustrates the method.