Study On Recurrent Algorithm Of A Type Of Compound Fuzzy Neural Network And Its Application

This paper has presented a type of recurrent compound fuzzy neural network (RCFNN) that has a recurrent function network and a rule network for identifying, modeling and controlling dynamic nonlinear systems. Based on the process knowledge and general fuzzy neural network structure, the rule network partitions the process operation region into a number of local operating regions. While the local behavior of the process is approximated by the recurrent function network and by adding recurrent nodes in the second layer of the function network, RCFNN not only has the ability of dynamic mapping, but also can effectively use the process knowledge due to its compound structure. Simulations have been made with some dynamic nonlinear function models on fuzzy neural network, dynamic fuzzy neural network, compound fuzzy neural network and RCFNN. The results show that the RCFNN is better than other fuzzy neural networks in learning precision and dealing with the dynamic systems. Its network structure with internal recurrent feedback is smaller than which of external feedback network. Simulation has also been made with some noise models, the result show that the RCFNN has good resist disturb ability and by modifying the network structure, the RCFNN can dealing with colored noise system. Finally, the RCFNN is applied to the DMF recycle system and obtain satisfied result.This research shows that the RCFNN not only has the stronger ability of handling the dynamic systems, but also possesses a good prospect in the application process.