The Research On Approximate Inference Algorithm For Dynamic Bayesian Networks

Bayesian network provides a powerful graph tool to express domain knowledge based on probability. It has been successfully applied to fault diagnosis, data mining, medical diagnosis, and other fields. Dynamic Bayesian network is a Bayesian network with the expansion on the factor of time, it provides a powerful tool to represent and deal with dynamic uncertain problem of stochastic processes in the field of artificial intelligence. Based on the comprehensive overview of dynamic Bayesian network, this thesis focuses on the research of approximate inference algorithm for dynamic Bayesian network. The main contents of this thesis are as follows:(1) This thesis makes a survey about the research on Bayesian network, including the origin, development, the model, the construction process, the type and the application of Bayesian network. On basis of this, dynamic Bayesian network is introduced. Moreover, the accurate inference algorithms for Bayesian network and dynamic Bayesian network are introduced in detail.(2) To the high dimension problem of traditional particle filter for Dynamic Bayesian Networks inference, based on part sample, a novel particle filter inference algorithm (PSPF) is proposed. The relative clusters of DBNs are created under the guidance of weakly interaction to reduce the dimension of problem solving. Particles are maintained over clusters of state variables, the belief state is represented as a product of particles. Ulteriorly, resampling and update over the entire state spaces of DBNs. Simulation results show that the proposed algorithm improves PF's computational efficiency notably and achieves more precise results.(3) BK algorithm and particle filter are two specific approximate algorithms for Dynamic Bayesian Networks. BK algorithm is more effective in computation, while it induces the error of inference. PF algorithm can estimate arbitrary distributions, but with high dimension problems. An Adaptive Hybrid Approximate Inference of for Dynamic Bayesian Networks (HAInf) is presented by combing the advantages of BK and particle filter. HAInf algorithm decomposes DBNs to Prototype Junction Tree for the sake of decreasing the complexity of inference. According to the size of clusters, particle filter is executed on a part of clusters, while BK is executed on the rest clusters. Ultimately, the inference is performed by propagating messages on Prototype Junction Tree. Compared with BK and PF algorithms, simulation results show that the proposed algorithm is more precise and holds time tradeoff.