Scribd Logo
Upload

Welcome to Scribd!

  • 8 views

    Engopt2012-0533 - Development and Implementation of Parallel Versions of The Particle Swarm Algorithm On Clusters Using Mpi

    Uploaded by

    Tg Wallas
    This document describes the development and implementation of parallel versions of the particle swarm optimization algorithm on computer clusters using MPI (Message Passing Interface). Three parallel PSO algorithms were created that use a master-slave paradigm and differ in their communication methods and particle updating approaches. The parallel algorithms were tested on benchmark problems and showed good performance and scaling. In particular, an asynchronous and immediate updating parallel PSO algorithm demonstrated excellent performance with over 90% parallel efficiency on the problems tested. This algorithm was then applied to an actual parameter estimation problem and compared to the results from another optimization method.

    Copyright:

    Attribution Non-Commercial (BY-NC)
    Download as PDF, TXT or read online from Scribd

    Engopt2012-0533 - Development and Implementation of Parallel Versions of The Particle Swarm Algorithm On Clusters Using Mpi

    Uploaded by

    Tg Wallas
    8 views1 page
    This document describes the development and implementation of parallel versions of the particle swarm optimization algorithm on computer clusters using MPI (Message Passing Interface). Three parallel PSO algorithms were created that use a master-slave paradigm and differ in their communication methods and particle updating approaches. The parallel algorithms were tested on benchmark problems and showed good performance and scaling. In particular, an asynchronous and immediate updating parallel PSO algorithm demonstrated excellent performance with over 90% parallel efficiency on the problems tested. This algorithm was then applied to an actual parameter estimation problem and compared to the results from another optimization method.

    Original Title

    533

    Copyright

    © Attribution Non-Commercial (BY-NC)

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    This document describes the development and implementation of parallel versions of the particle swarm optimization algorithm on computer clusters using MPI (Message Passing Interface). Three parallel PSO algorithms were created that use a master-slave paradigm and differ in their communication methods and particle updating approaches. The parallel algorithms were tested on benchmark problems and showed good performance and scaling. In particular, an asynchronous and immediate updating parallel PSO algorithm demonstrated excellent performance with over 90% parallel efficiency on the problems tested. This algorithm was then applied to an actual parameter estimation problem and compared to the results from another optimization method.

    Copyright:

    Attribution Non-Commercial (BY-NC)
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    8 views1 page

    Engopt2012-0533 - Development and Implementation of Parallel Versions of The Particle Swarm Algorithm On Clusters Using Mpi

    Uploaded by

    Tg Wallas
    This document describes the development and implementation of parallel versions of the particle swarm optimization algorithm on computer clusters using MPI (Message Passing Interface). Three parallel PSO algorithms were created that use a master-slave paradigm and differ in their communication methods and particle updating approaches. The parallel algorithms were tested on benchmark problems and showed good performance and scaling. In particular, an asynchronous and immediate updating parallel PSO algorithm demonstrated excellent performance with over 90% parallel efficiency on the problems tested. This algorithm was then applied to an actual parameter estimation problem and compared to the results from another optimization method.

    Copyright:

    Attribution Non-Commercial (BY-NC)
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 1

    ENGOPT2012-0533 - DEVELOPMENT AND IMPLEMENTATION OF PARALLEL VERSIONS OF THE PARTICLE SWARM ALGORITHM ON CLUSTERS USING MPI

    Moraes, Antonio de O Samel* (PEQ/COPPE/UFRJ - BRAZIL) Lage, Paulo Laranjeira da Cunha (PEQ/COPPE/UFRJ - BRAZIL) Secchi, Argemiro Resende (PEQ/COPPE/UFRJ - BRAZIL)
    Abstract: In despite its recognized strenght in global optimization problems, the high number of objective function evaluations requested by Particle Swarm Optimization (PSO) method severely limits its aplication when the computational cost of the objective function calculation is large, as in most real engineering problems. In order to overcome this deficiency, enabling the efficient use of the PSO algorithm in the large-scale engineering problems, the parallel computation in clusters appears as an excellent alternative for reducing the optimization time. By its nature, this algorithm can be easily parallelized, because the objective function is calculated independently for each particle in the swarm. Therefore, three parallel versions of the PSO algorithm, using the set of functions of the MPI (Message Passing Interface) library, were developed in this work. These parallel algorithms use the master-slave paradigm to exchange information between the processors, differing among themselves by the communication mode among the master and the slave processors (synchronous or asynchronous) and the particles update in swarm (immediate or by swarm). A set of benchmark tests was conducted to validate and to analyze the perfomance of the developed methods. In particular, the AIUPPSO algorithm (Asynchronous and Immediate Update Parallel PSO) showed excelent perfomance, with linear speed-up and parallel eficiency higher than 90 % for all problems tested. The results were obtained in the MIMD parallel computers with distributed memory and Infiniband network of the cluster in the Termofluidynamics laboratory (LTFD of PEQ/COPPE/UFRJ). Finally, an actual parameter estimation problem with a costly objective function was solved and the solution was compared to that provided by the ODRPACK95 optimizer.

    You might also like