Please use this identifier to cite or link to this item: http://localhost:8081/xmlui/handle/123456789/2253
Title: PERFORMANCE STUDY OF VARIANTS OF PSO AND THEIR PARALLEL IMPLEMENTATIONS
Authors: Reddy T, Tirupathi
Keywords: PARALLEL IMPLEMENTATIONS;ELECTRONICS AND COMPUTER ENGINEERING;STUDY PSO;VARIANT PSO
Issue Date: 2012
Abstract: Particle Swarm Optimization (PSO) is gaining popularity for solving computationally intensive problems. There are many variants of PSO. In this thesis, we first propose a new archiving technique for Multi-Objective Particle Swarm Optimization (MOPSO) and compare it with existing techniques. We then study the performance of parallel implementations of Standard Particle Swarm Optimization (SPSO) for solving some well known combinatorial problems. While solving problems with MOPSO, the quality of solution set depends on the selection of global best from archive. In this report, an improved radius-based archive technique is proposed for multi-objective problems using PSO. The exploration of extreme solutions over Pareto front gained more importance, while solving real-world problems such as permutation flow-shop scheduling. Experiments are conducted on the known permutation flow-shop scheduling problems taken from literature and the results of proposed technique are compared with other archive techniques such as crowding distance and radius-based guide selection techniques. Due to complex business strategies involved in cloud computing environment, job scheduling that satisfies QoS requirements of user jobs is a challenging issue. This requires allocation of computing resources to incoming user jobs that achieves maximum benefit, and at the same time satisfies various QoS requirements of user's jobs. Since, the scheduling problems belongs to the family of combinatorial problems, designing of a algorithm that achieves optimal solutions in polynomial time is not possible. In this report, a standard particle swarm optimization based approaCh for multi-QoS job scheduling has been proposed. Travelling Salesman problem (TSP) also belongs to the family of combinatorial problems, and it is not possible to design a polynomial time algorithm to solve this. In this report, we present and analyze the application of Standard Particle Swarm Optimization for Travelling Salesman Problem. Experiments are conducted on benchmark datasets obtained from TSPLIB and the results are compared with PSO-based approach. For large scale problems, even the variants of PSO (i.e., PSO, SPSO, and MOPSO) require long running time to get the solutions. To achieve the computational speedup, the parallelization of above proposed methods is also presented in the report. Finally, we compared the performance of parallel implementations with serial implementations in terms of execution time and speedup.
URI: http://hdl.handle.net/123456789/2253
Other Identifiers: M.Tech
Research Supervisor/ Guide: Mishra, Manoj
metadata.dc.type: M.Tech Dessertation
Appears in Collections:MASTERS' THESES (E & C)

Files in This Item:
File Description SizeFormat 
ECDG21999.pdf4.96 MBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.