Research On The Improvement For RBF Neural Networks Based On Cloud Theory

An improved algorithm for RBF neural networks is presented based on cloud theory, with replaced Gauss radial basic function by normal cloud model ,the determine of ex and en for RBF net change into the determine of normal cloud model parameters ,so that the out of the hidden layer is vague and random , the quintessence of cloud theory has been shown well . The random of the data samples used by the net has been passed to the out layer of the net smoothly , the"uncomplete vague"defect of the classical vague theory has been overcome .Compared with cloud theory , the improved algorithm needs not the extraction of vague rules essential for cloud theory for the reservation of self-learning in RBF . At the same time, two new problems in the improved algorithm has been solved well.Firstly, in multidimensional cloud model formula when the dimension is high, the sunmry is big enough to make the out in insensitive area ,so that the net can not be trained . The problem is solved by the cloud range parameter used as weighted value . Practice has proved feasible and the scope of multidimensional cloud model is expanded .Secondly , a new method for the merger of different range cloud has been presented . By three-dimensional optimization , after merger the expect curve parameter of the cloud is got .Using factual fouling data to check up the improved algorithm for RBF neural networks based on cloud theory , the conclusion shows that the improved algorithm is superior to the classical RBF in prediction and the actual result is good . Simultaneously the improved algorithm can be transplanted to many existant improved algorithm for RBF neural networks .Finally , the analysis of the forecast accuracy has been done , and the follow-up research and improvement direction of the issue has been identified .