Research On Bayesian Network Structure Learning Algorithm Based On Node Chunk Sequence

Bayesian network is a powerful tool for expressing the causal relationship between complex probability knowledge and the characteristics of data sets.How to learn Bayesian network structure efficiently has always been the focus of research.Cement chiller is the key equipment for heat recovery and cooling high temperature cement clinker in cement production line.Its bayesian network model establishment and fault diagnosis directly affect the clinker quality and energy consumption of cement production.Dynamic Bayesian network is the expansion of Bayesian network in time series space,and the efficiency and accuracy of its structure learning directly affect the applicability of the network.Aiming at the problem of poor optimization efficiency of traditional Bayesian network structure learning algorithm,two Bayesian structure learning algorithms NOK2 algorithm and NCSC algorithm based on node order search are derived.And the two algorithms are applied to establish the Bayesian network and dynamic Bayesian network structure models of clinker heat exchange process parameters.The specific research work is as follows:Firstly,NOK2 algorithm is proposed based on node order optimization of Bayesian network structure learning.The fitness function for quantitative evaluation of node ordering is established by computing the weight matrix of spanning tree.Hybrid crossover strategy and isolating node operation are adopted to improve the optimal result of node order searched by genetic algorithm.And dynamic learning constants and inverted mutation strategy are used at the same time.The optimized node sequence is taken as a prior knowledge of the K2 algorithm to construct the optimal Bayesian network structure.Secondly,a local Bayesian network structure search algorithm with node chunk sequence constraints is proposed,which is NCSC algorithm.A directional maximum weight spanning tree structure is first introduced by fitness score.And the node chunk sequence is constructed on this structure.The potential parent node set of each node is determined using the node chunk sequence.Network structure is built by searching the parent node set of each node,and the optimal Bayesian network structure is obtained after illegal structure modification.Finally,two Bayesian network structure learning algorithms based on node sequence search are applied to the fault diagnosis model of clinker heat exchange of grate cooler and the dynamic Bayesian network structure learning respectively.Based on the Bayesian network structure constructed by NOK2 algorithm and NCSC algorithm,the Bayesian network model of clinker heat transfer process is established.Parameter learning and fault reasoning diagnosis are conducted on this model.According to the characteristics of dynamic Bayesian network structure,the hybrid algorithm INKABA algorithm is introduced based on NOK2 and NCSC algorithms for structural learning of dynamic Bayesian network.