Study on Several Issues of the Bayesian Dynamic Models

Abstract: In this paper, the knowledge about Genetic Algorithm and Neural Network is introduced at first. And considering the application in the paper, the method for improving Neural Network using Genetic Algorithm is also introduced. Then the method for solving the Bayesian Dynamic Model with a nonlinear observation equation using Evolutional Neural Network (ENN) is studied. Where the weights of the neural network is decided by Genetic Algorithm. The method avoids the local extreme minimum points in the target function using Neural Network algorithm. MCMC method is a kind of useful way for dealing with Nonlinear Bayesian Dynamic Model. If the target function is not easy to sample, we can consider using Metropolis-Hastings algorithm to sample from the target function by selecting a proposal distribution, Where the proposal distribution is easy to sample. The AM algorithm introduced in this paper is a kind of Metropolis algorithm. Using AM algorithm to simulate nonlinear dynamic models is not only simplifying the sampling process, but also has better convergence property. At last, on the basic of the standard GDLM and MGLM, a kind of new model called Mixed General Linear Dynamic Model (MGDLM) is put forward. Then some primary study for two special kind of MGDLM is introduced…
Key words: Nonlinear Bayesian Dynamic Model; Genetic Algorithm; Neural Network; Adaptive Metropolis Algorithm; Mixed General Dynamic Linear Model

This entry was posted in Master Thesis. Bookmark the permalink.