Time Series Forecasting Based On Bayesian Network

Petroleum is one of the important energy, which has an impact on the world economy and politics. Now all the countries keep their eyes on the price of oil. So they can reduce the negative impact that the change of the price brings. Lots of factors will influence the price, such as supply and demand, economy, international politics, military, diplomacy and so on. The price has many characters, such as uncertainty, complex nonlinear, consists of many different things and the data contains much noise. Because of the traditional forecasting method of the price is hardly reliable, and its prediction accuracy is low, many scholars begin to have a research on a more widely used model which has a much higher prediction accuracy, much better properties and better efficiency. Bayesian network combines the Bayesian theory and the graph theory.And it provides a way to deal with the probabilistic and complicated problems. With the advantages of Bayesian network, which is strict and consistent, it has become a hot topic in the world.In this paper, by understanding of the domestic and foreign research, combining with the theory of Bayesian network, a new forecast model appears, which is based on Dynamic Bayesian Network model. Firstly, the new model makes an analysis of the influential factors in petroleum price, and then make a Dynamic Bayesian Networks based on the cause-effect graph. Through the training data, it can find out the best parameter and structure. That is to say, we find out the best Dynamic Bayesian Network model. Finally, we simulate the model with Matlab. It turns out that the Dynamic Bayesian Network has higher precision and better performance.