Study of preferential vector of particle swarm with hierarchical reinforcement

Abstract

A preferential vector algorithm of particle swarm with hierarchical reinforcement learning is proposed to solve the balance problem of global searching range and local searching precision, and the problem of fixedly adjusting strategy of inertia weight. Firstly, the preferential particle position with crossover operation was introduced on the standard particle swarm algorithm. The sequential adding operations on single step of particle position were divided into the seeking of middle particles. The operations of crossover and mutation were combined to keep the elite particle swarm. Secondly, the adjusting strategies of inertia weight were treated as the actions. The hierarchical learning was executed in every iteration of particle swarm and the strategy with maximal discounted profit was selected. The experiment proved the validity of the algorithm.