Neurofuzzy GMDH network and its application to the soft-sensor for ethene distillation process

A neurofuzzy GMDH network for system modelling and prediction is developed. This network introduces the structure of GMDH network into its structure and its partial descriptions are represented by the radial functions network which is equivalent to fuzzy models in the first layer and by an N-Adeline network in other layers. The results of function approximation shows that the NF-GMDH network outperforms the conventional GMDH network. The final example shows that NF-GMDH can be used in soft-sensing modelling of the ethene rectification process.