BFALCON Generalization Capability Improvement Based on PCA Initialization

Fuzzy adaptive learning control network (FALCON) may lack in generalization capability for the typical initialization based on adaptive resonance theory (ART). Aiming at this problem, a new initialization of FALCON based on the principal component analysis (PCA) is proposed in this paper. After specific analysis of topological structure of FALCON and the existing initialization based on the ART, FALCON is initialized based on the statistical results of PCA. Subsequently, a set of complete computation algorithm, which can improve the generalization capability of FALCON, is provided with term neuron distribution, pretreatment and post- treatment. Then, a metropolis learning coefficient is introduced into the supervised learning algorithm of FALCON. Eventually, being applied to real problem, the FALCON initialized based on PCA is prove to have a better generalization capability than the typical neural networks.