Parameter Selection in Particle Swarm Optimization
EP '98 Proceedings of the 7th International Conference on Evolutionary Programming VII
The particle swarm optimization algorithm: convergence analysis and parameter selection
Information Processing Letters
The Journal of Supercomputing
Population structure and particle swarm performance
CEC '02 Proceedings of the Evolutionary Computation on 2002. CEC '02. Proceedings of the 2002 Congress - Volume 02
Population structure and particle swarm performance
CEC '02 Proceedings of the Evolutionary Computation on 2002. CEC '02. Proceedings of the 2002 Congress - Volume 02
A hierarchical particle swarm optimizer and its adaptive variant
IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics
| Hi-index | 0.00 |
Multi task parallel processor arrays are a common machine architecture in which, typically, the tasks running in parallel occupy disjoint subarrays of the machine. On dynamically and partially reconfigurable processor arrays the tasks can be changed during run time. This is useful for online scenarios when the relative importance of tasks might change and therefore the assignment of computational resources to the tasks should be changed. Examples are optimization tasks in an online scenario in which the results of some tasks are needed earlier than expected at initialization. For such tasks the size of their subarrays must be increased because they need more computational resources to speed up. In this paper we design flexible Particle Swarm Optimization (PSO) algorithms for 2-dimensional reconfigurable processor arrays where the algorithms can change their size and have a good optimization behaviour. Since PSO is an iterative, individual-based optimization algorithm that relies upon interactions of neighbouring particles suitable for fine-grained parallel architectures. We propose a dynamic 2-dimensional hierarchical ordering of the particles within a tasks subarray so that the best particles are concentrated in the center. This gives the best particles the strongest influence on the swarm. A further advantage is that size reductions of the tasks can easily be done by cutting off the outer parts of the swarm which contain mainly the less good particles. It is experimentally shown that the proposed algorithms perform better than standard PSO algorithms under conditions with varying supply of computing resources that are available for the tasks. Moreover, also for conditions with constant supply of processing resources and no need for size changes the proposed algorithms perform well.