加权模糊隶属度多模型RBF网络及其应用-通信与信息系统专业论文.docxVIP

武汉理工大学硕士学位论文Abstract 武汉理工大学硕士学位论文 Abstract Artificial neural network(ANN)alreMy has been applied in many fields and achieved plentiful fruits already．However,practical problem that has come to prominence is that ANNs ale unstable predictors．Neural network．ensemble is a leaning paradigm where a collection of a finite number of ANNs is trained for．sallle task．It originates from research work ofHansen and Salemon on 1990，which shows that the generalization ability of a neural network ensemble system Can be significantly improved through ensembling a nember ofANNs． Objective weather element forecasting has the very important effect in certain area weather forecasting．Tradition means such as statistic analysis，MOS forecasting expgrience forecasting etc are usually adopted．In view of the weaknesses of simple neural ne铆ork for weatherc forecasting,a new and more effective weather forecasting model called multi-model RBF neural networks is developed in this paper based on time series analysis and neural network ensembling．Then main contribmions ofthis dissertation are summarized as follows： Firstly,several weather forecasting methods have been introduced in this paper． It is important that introduced ANN technology development and application in atmosphere fields． Secondly,the structure and algorithm of RBFNN has been introduced．The theory and solving methods of improving generilaztion ability for neural network has been analysed．It is important that inoduced neural network ensemble technology theory and application．Nurual network ensemble is the most effective method in improving genefilaztion ability for neural network at present． Thirdly,aiming at the learning problem using neural network in complex problems，a expert network RBF model MRBFN and a multi—model RBF network model FWM based on weighted fuzzy degree ofmemberships are presented．The two models are given algorithm and simulation results． Lastly,make u∞of the time series analysis，lcad into FWM network model，to Ⅱ 武