Time series prediction using optimal theorem and dynamic Bayesian network

A novel multi-step-ahead time series prediction model is proposed based on combination of the multi-information fusion optimization model and the dynamic Bayesian network (DBN). Our contribution includes: (1) a theorem of multi-information fusion prediction is proposed and proved. We can obtain the optimal estimate value of prediction based on the proposed fusion estimation theorem. (2) Based on proposed theorem, we consider using the recursion-based DBN to enhance performance of the optimal-based direct prediction model. A novel graph model named the R-DBN that generated from combination of multi-information fusion prediction and DBN is developed to predict multi-step-ahead time series data. The simulation and comparison results show that the proposed model is more effectiveness and robustness.