Application For The Combinatorially Forecasting Model Based On Lssvr

In the classification field,Support Vector Machine(SVM)has being well developingin theories and applications.In regression field,there are still applications worth study-ing.Least Squares Support Vector Machine For Regression(Lssvr)is a machine learningmethod developed by Suyken and his co-workers,and is based on Statistical LearningTheory which is developed by Vapnik and his co-workers.In a word,Lssvr is also a newmethod of SVM.It has better learning ability, and solves the problems ,such as non-linearmodel, small sample and local extremum.It has also few parameters,and solves linearsystem to get problems'solution. So far SVM has become popular in all fields.In this thesis,first described SVM development and related some research results. Sec-ondly,the main content of Statistical Learning Theory is brie?y introduced.There presentsThe basic principle and process of Lssvr.The last,considered advantage of traditionalARIMA method, Lssvr, and Markov Chain in time series predication,there makes up oftwo models: ARIMA-Lssvr combined model, Lssvr-Markov Chain combined model. Theformer model divides time series into linear part and non-linear part.The later modeldivides time series into certain part and random part.