Soft-Sensor Modeling of PVC Polymerizing Process Based on F-GMDH-Type Neural Network Algorithm

Table 2

Characteristic comparison between BP neural network and GMDH-type neural network.

Name	BP neural network	GMDH-type neural network
Structure of the Network	Network structure with three or more layers of network	Completely rely on the self-evolvement of network
Connection pattern of the neurons	Full connection among layers	Partial connection among layers
Self-organizing	The number of layers and the number of neurons of the network are predetermined	The number of layers and the number of neurons in the network are self-generated
Estimating method of parameters	Generated by iteration	Estimate the parameters of the nodes according to the internal criterion
Rules of using samples	All the samples are used to train the parameters, which can easily lead to the phenomenon of over fitting	The part samples are used for training the parameters and the rest are used for screening
Interpretability of Model	It is hard to understand	There is an explicit expression
Convergence performance of the model	There is a strong relationship between convergence and initial states	Determine the convergence state according to the termination criterion
Complexity of the program	The calculation time is too long, so it is suitable for using multiprocessor	It is suitable for multiprocessor or single processor