The Research Of Markov Model Forecasing Method And Application

With the development of Markov theory,developing the Markov models based on this theory which included Markov Chain Model and Hidden Markov Model.Because of the advanced characters of those model, which has aroused the in and out of abroad scholar's universal interest.Currently,Markov models have been applied to speech recognatio,stock prediction environment quality and information security and so on. The method of those concrete examples is that using initial probability distribution and the transition matrix to construct Markov Models and then dong prediction of concrete problems.In the research of Machce Learning,the method of prediction has been to hot spot among the in and out of abroad,such as the regression of BP neural network,the regression of radial neual network Generalized Regression neual network and other prediction methods.Because the structure of those neural network is so simple and easy to inplement,which has played the vital roles in the regression prediction at one time.However, the structure of those neural network is so difficulty determined and easy to fall into local extremum,which limit their development.Moreover,those neural network established in the theory of gradually which need many infinite samples to get the real simulation sample distribution function, but in fact the samples is limited,therefore the apply of regression based on neural network have congenital flaw.After that, the support vector machines based on statistics theory of learning apply into regression prediction research,which seek to the compromised method between the structure of machine learning complexity and the study precision,getting the most optimise ability.However,when do regression prediction based on support vector machine, because it's difficult to choose function parameters,and the complexity of the support vector machines regression algorithm cause to the slow running time. The training time is so long on large-scale classified question, which become the flaw that this method is unable to make up.Because the stable characteristic of Markov theory, this thesis proposed the method that unifies the Markov theory and the method of regression prediction,using multiple regression's method determined Markov transition matrix,to construct the marokv regression model,and then applies this model into predict the assignment of national income,getting good result.At the same time,I unify the Markov model and the transformation model in the economic to construct Markov switch regression Models,through experiments get low error and low running time. Simultaneously, this paper unified the Hidden Markov Model and the EM algorithm, to construct the EM-HMM model and relevant algorithm, and applied this algorithm into the biological gene heredity law.The dissertation includes:1. This thesis review the development process of Markov theory,first of all introduce the basic concep and related theorem of Markov theory,carried on the detailed analysis to the concrete structure method of Markov model which include Markov chain model and Hidden Markovian model, simultaneously review and summary to the Markov models in actual problem application.2. This paper mainly discussed the regression prediction research, first of all discussion in the artificial neural networks appliesd into regression prediction. Analysis and discussion the BP neural network,radial neual network,Generalized Regression neual network, getting the structure of those neural networks is so difficulty determined and easy to fall into local extremum, which limit their development. And then the discussion based on statistics theory of learning apply into regression prediction research,but due to it's difficult to choose function parameters,and the complexity of the support vector machines regression algorithm cause to the slow running time,thus to do new research in regression prediction algorithm.3. This paper the most important works is in two aspects,the first aspect is carrying on regression prediction,and then unifies the Markov model and the multiple regression analysis method to construct the multiple regression Markov model,and proposed the multiple regression Markov algorithm,then put it into national income assignment forecast aspects. Simultaneously,I unify Markov model and the transformation model in the economic to construct Markov switch regression Models,and apply this model into UCI data sets to verify this algorithm,and compared it with upport vector machine regression algorithm,getting smaller error and lower running time. In the second aspect, this article unify the EM algorithm and the Hidden Markov Model,to construct Hidden markov model based on EM algorithm, proposed the EM-HMM algorithm and applies it in Mendel's gene heredity law. Also, through 6 experiments to test,using a pair of independent allele, two pair of alleles and three pair of alleles to diploid and tetraploid carries on the experiment in the biology using the EM-HMM algorithm,thus the better reflection gene heredity rule.