A RBF Neurocomputing Model Based on Clustering

Radial basis function neurocomputing model (RBFNM) has broad application foreground in engineering computing field; however, there is the problem of slow training/learning speed and low fitting precision in the case of large number of samples. In allusion to this case, a base on clustering radial basis function neurocomputing model (BC-RBFNM) has been proposed in this paper. Firstly, samples have been clustered and analyzed using the model; then the sub-networks have been constructed for each class according to the clustering results and the relative parameters have also been determined; finally, a BC-RBFNM has been formed by the sub-networks. Theoretical analysis and property testing have been performed on the model. The results show that the BC-RBFNM model can alter training/learning speed of network models, minimize size of network models and improve the predicting precision.