This paper presents the development of a set of hybrid metaheuristic based on the use of evolutionary strategies in conjunction with classical and quantum multi-swarm optimization with master-slave approach. These algorithms are named Competitive Evolutionary Multi-Swarm Optimization (CEMSO) and Competitive Quantum-Behaviour Evolutionary Multi-Swarm Optimization (CQEMSO). For comparison and validation of the results, four engineering problems encountered in many publications scientific are used: Welded Beam Design (WBD); Minimization of the Weight of a Tension/ Compression Spring (MWTCS); Speed Reducer Design (SRD); Design of a Pressure Vessel (DPV). The algorithms were developed under the CUDA architecture, which provides a massive parallel computing environment that enables a more appropriate data allocation regarding the organization of swarms, as well as contributing to the significant decrease in processing time.
With the application of evolutionary strategies in the PSO and QPSO algorithms, as well as the proposed boundary conditions, the solutions described in this document offer several advantages. We can highlight improvements in the ability to search, increasing the convergence rate and high degree of parallelism. These facts are confirmed by the data obtained (i.e. Execution time, best solutions obtained, mean and variance of results) by CQEMSO and CQEMSO algorithms when compared to those obtained from multi-swarm approach for PSO (COMSO), EPSO (COEMSO) and QPSO (COQMSO). All of these algorithms were implemented and subjected to performance analysis through experiments with engineering problems described above.
|
