Wiener type SOM-and MLP-Classifiers for recognition of dynamic modes

A recognition method modified from the classical Wiener-model is introduced, which recognizes the dynamic modes of an industrial process automatically on-line on the basis of the measurement information available. The dynamical linear part is based on Laguerre systems for all inputs. The values of the Laguerre system outputs contain information about the dynamic behaviour of the input signals in the near past. The static nonlinear mapping is realized with a MLP and a Kohonen's SOM, which map the “behaviour” to a unique binary code for each functional state i.e. dynamic mode. It is applied to the B. subtilis batch fermentation producing alpha-amylase to recognize the five functional states of the process.