IMPLEMENTING PARALLEL PSO ALGORITHM USING MAPREDUCE ARCHITECTURE

Abstract

ABSTRACT Optimization is the problem of finding minimum or maximum of a given objective function relative to some set, often representing a range of choices available in a certain situation. Particle Swarm Optimization (PSO) is a simple and effective evolutionary algorithm, but it may take a reasonable time to optimize complex objective functions which are deceptive or expensive. To avoid being trapped in local optima, Particle Swarm Optimization requires extensive exploration for multimodal and multidimensional functions. Expensive functions whose computational complexity may arise from dependence on detailed simulations or large datasets, takes a long time to evaluate. For such functions PSO must be parallelized to use multiprocessor systems and clusters efficiently. Parallelization of PSO can lead to scalable speedup in performance. PSO can be naturally expressed in Google’s MapReduce framework to develop a simple and robust parallel implementation. To improve optimization of difficult objective functions and to improve parallel performance, modifications could be made to this flexible implementation of the algorithm. In the proposed work the classic Particle Swarm Optimization Algorithm has been implemented on Big Data platform Hadoop using MapReduce Architecture. The algorithm has been applied to optimize parameters of basic COCOMO Model need to calculate effort of the project. The experiments show that the Hadoop could carry out big data calculations which normal serial PSO could not. The proposed model would have better efficiency for intensive computational functions.