Many Object Complex Network Clustering Based On Discrete Particle Swarm

Complex network clustering is a basic technique to study complex networks.With the advent of the era of big data,factors affecting clustering and the number of nodes in the network are increasing.The increase of clustering factor leads to the increase of clustering objective function and the increase of nodes in the network leads to the increase of variable dimension.However,the current multi-objective complex network clustering algorithm can't effectively solve the problem of spatial dimension and variable dimension.Therefore,this paper proposes a super multi-object complex network clustering model and a super multiobject discrete particle swarm optimization algorithm.The multi-objective complex network clustering model transforms clustering factors into specific objective functions.These objective functions contradict each other.By balancing these functions,the clustering of high cohesion and low coupling between classes is finally achieved.The model can also be run once to produce different levels of solutions,more in line with real-world requirements.For large-scale complex network clustering,an adaptive random grouping method based on random grouping method is proposed in this paper.Adaptive random grouping combined with the specific characteristics of the complex network divides nodes into different groups to make the grouping results more uniform.On the basis of adaptive grouping,a discrete reference vector particle swarm optimization algorithm is proposed in this paper.The discrete reference vector in particle swarm optimization can divide the solution space into several parts.The discrete particle swarm is perturbed around the reference vector.Then the optimal individuals and the global optimal individuals are selected according to the Angle between the population and the reference vector.The selected individuals have the advantages of avoiding falling into local optima and fast convergence.The remaining individuals adjust their speed and direction based on global and individual optima.Finally,this thesis by improving the discrete reference vector particle swarm optimization algorithm proposes the dynamic decomposition particle swarm optimization algorithm based on the law of large Numbers.The dynamic decomposition particle swarm optimization algorithm increases the diversity of population by dynamic screening individuals.In order to improve the calculation efficiency,some nodes randomly selected are substituted into the objective function to reduce the calculation time and improve the calculation efficiency,and the rationality and feasibility of this strategy are demonstrated theoretically.In addition,theoretical,practical,low-dimensional and high-dimensional data sets are used to verify the superiority of the reference vector based discrete particle swarm optimization and dynamic decomposition particle swarm optimization in the clustering field.