Structure Learning Of Bayesian Network By Use Of The Hy-brid Fish Swarm Optimization Algorithm

Bayesian network(BN)is a graphical model which is developed by the probability theory with graph theory,it can intuitively explain dependent and independent relationship between random variables by the graphical language.Because of its unique advantages in encoding and reasoning uncertain knowledge,it plays an important role in artificial intelligence,automatic control,machine learning,etc.However,the construction of Bayesian network is a complicated process,and structural learning for Bayesian networks from data is a non-deterministic polynomial hard problem.Therefore,the core of the Bayesian network theory research is to explore the effective learning algorithm for structural learning in Bayesian networks.Usually,the main search algorithm of structural learning in Bayesian networks is heuristic searching algorithm,such as genetic algorithms,particle swarm optimization,ant colony algorithm,artificial fish-swarm algorithm,etc.Particle swarm optimization(PSO)and artificial fish-swarm algorithm(AFSA)are two typical swarm intelligence algorithms.They have been successfully applied to the structure learning of Bayesian networks.Based on the characteristic of PSO and AFSA,an improved algorithm is proposed and introduced in Bayesian network structure learning,which is named P-AFSA.The main works can be summarized as follows:(1)We give a comprehensive introduction to theories of Bayesian network,and the state-of-art research,specifically on the Bayesian network structure learning,and the score search methods.(2)The application of learning Bayesian network structure by AFSA is introduced.In order to improve the accuracy,the limitations of searching in the GA-AFSA are analyzed.The individual remembering capacity and communicating capacity of particle swarm optimization algorithm are introduced into the AFSA.We use the ideas of superiority inheritance to retain the superiority individual as far as possible.Therefore,a new method named P-AFSA is proposed,based on the hybrid Fish swarm optimization algorithm.(3)We apply the P-AFSA to the Bayesian network structure learning.The initial undirected graph is generated by the mutual information and the maximum spanning tree,which is the foundation of the initial population.Based on the particle's capacity of experience learning and capacity of information sharing between swarms of PSO,we designed two behaviors,backtracking and collaborating,which enlarge the behavior model of AFSA and avoid the blindness of searching.And the updating strategy of the individual position based on superiority inheritance operator is designed.Compared with the BPSO and GA-AFSA,simulation experiments results show that the improved algorithm has better performance.