Abstract: An intelligent ident ification combined structure model is proposed using self-or ganizing map (SOM) andsupport vector machines ( SVM) . This model can improve the self-organizing map algorithm using Voronoi vector toreduce space occupation and improve convergence, and develop pr obability intelligent identification training samplesset. Due to the complexity of the summary statistics, the aut hors select kurtosis, skewness, quantile and cumulativeprobability as parameters for data distributions identification training sets in experience. The combined structuremodel is divided into two layers. In the first SOM layer, different reliability distr ibutions tr aining sets are clusteredinto groups using SOM. In the second SVM layer, the clusters are learned and classified respect ively in each groupusing novel multi2class support vector machines. Random data time ser ies of 23 types of probability distr ibutions aretesting identified in the trained model. The results indicate that the identification rates are higher by the intelligentmodel compared to BP neural networks and probability networks models.

Key words: neural networks, support vector machines, machine learning, reliability, probability distribution, pattern recognition