Quantum-behaved Particle Swarm Optimization Algorithm And Applying IT On Learning Bayesian Networks

Since learning Bayesian networks from data is a NP-hard problem, especially when the set ofdata is big, the process of learning Bayesian networks structure is prone to fall into prematureconvergence and obtain a local optimal Bayesian networks under Particle Swarm Optimizationalgorithm and Hill Climbing algorithm, K2algorithm must know the node order while MWSTalgorithm has a bad accuracy. So a new approach is proposed. The Binary Quantum-behavedParticle Swarm Optimization algorithm and Bayesian Information Criterion score are applied toobtain an optimal Bayesian networks. Besides, in order to get a more optimal network structure, theprocess of searching has deleted cycle. A quantum Bayesian network based on quantum mechanicsis also mentioned in our paper. Quantum Bayesian network is one of the quantum probabilisticgraphical models, it is exploration of combining quantum theory and graph model, the mainresearch works of this dissertation can be summarized as follows:Firstly, introduce the basic theory of Bayesian networks and analyze their model instance, thenintroduce the parameter learning and structure learning of Bayesian networks. We analyze the mainstrengths and weakness of the existing algorithms. Finally, we use K2algorithm and MWSTalgorithms to learn the structure of Bayesian networks.Secondly, this paper researched on two kinds of QPSO. One is Quantumn Particle SwarmEvolutionary Algorithm(QPSEA), the other is Quantum-behaved Particle Swarm Optimization algorithm(QPSO).The simulation results of the eight benchmark functions showed that why we choose our algorithm.Thirdly, Considering the characteristics of Bayesian network, we use the proposed algorithm tolearn Bayesian network. BQPSO and BIC is applied in the process of Bayesian structure learning.In order to get a more optimal network structure, the process of searching has deleted cycle. Thesimulation results indicate that BQPSO algorithm is proposed for Bayesian structure learning has abetter performance than PSO, K2and MWST.Fourth, we researched on Quantum Bayesian network. Two kinds of network is reasched,Besides，we propose a simple Quantum Bayesian network structure. And we demonstrate how todistribute probability amplitude of each node in a Quantum Bayesian network. Like classicalBayesian networks, we prove that Quantum Bayesian networks can be expressed as the product ofthe amplitude, then analysis its nature and the areas where quantum Bayesian Network will beapplied in the future.