Using neural networks in structure characterisation

Structure characterisation (of ill-defined systems), i.e. the selection of a class of models from a set of candidate model classes, can be viewed as a pattern recognition problem which may be handled using neural networks. In this paper, neural network based structure characterisation is considered on the basis of an illustrative example application (wastewater purification). Classification results are presented for various different neural classifiers and signal (pre)processing techniques. Also the issues of generalisation overtraining and noise sensitivity of the neural classifiers are discussed.