Coarse-grained Distributed Computing Model Of Particle Swarm Optimization And Its Implementation In Spark Framework

In terms of intelligent computing,the particle swarm algorithm has the advantages of fast search speed,low parameter setting,etc.It has been widely used in evolutionary computation.In this paper,the computational time-consuming problem is studied.On the one hand,the agent model can be adopted.On the other hand,with the development of the hardware,general purpose GPUs and distributed computing technologies are gradually attracting attention.It also supports the research of parallel algorithms.Therefore,for this purpose,it is considered to accelerate the hardware technology without loss of accuracy.This paper accelerates the particle swarm algorithm in both coarse and fine granularity.The main design model adopts the island model.On the one hand,using Spark for distributed coarse-grained computing,the population is divided into several sub-populations,and the sub-populations are internally iteratively calculated.Each iteration calculation updates the global optimal position of the total population.On the other hand,through the GPU's powerful single-instruction multi-stream parallel processing method,each iteration of each sub-population is further fine-grained,and each individual in the sub-population is calculated in parallel,and finally the overall task calculation time is shortened.The core issue in this paper is how to design the particle swarm algorithm to maximize parallelism.Finally,simulation experiments verify the superiority of the GPU cluster acceleration effect.When the problem size is larger and larger,the acceleration will increase,eventually reaching the time-consuming problem of solving the calculation.