A hybrid particle swarm with velocity mutation for constraint optimization problems

Files

RA_hdl_83947.pdf (589.48 KB)
  (Restricted Access)

Date

2013

Authors

Bonyadi, M.
Li, X.
Michalewicz, Z.

Editors

Blum, C.

Advisors

Journal Title

Journal ISSN

Volume Title

Type:

Conference paper

Citation

Proceedings of the 15th Annual Conference on Genetic and Evolutionary Computation, GECCO'13, 2013: pp.1-8

Statement of Responsibility

Mohammad Reza Bonyadi, Xiang Li, Zbigniew Michalewicz

Conference Name

Genetic and Evolutionary Computation Conference (15th : 2013 : Amsterdam, Netherlands)

Abstract

Two approaches for solving numerical continuous domain constrained optimization problems are proposed and experimented with. The first approach is based on particle swarm optimization algorithm with a new mutation operator in its velocity updating rule. Also, a gradient mutation is proposed and incorporated into the algorithm. This algorithm uses ε-level constraint handling method. The second approach is based on covariance matrix adaptation evolutionary strategy with the same method for handling constraints. It is experimentally shown that the first approach needs less number of function evaluations than the second one to find a feasible solution while the second approach is more effective in optimizing the objective value. Thus, a hybrid approach is proposed (third approach) which uses the first approach for locating potentially different feasible solutions and the second approach for further improving the solutions found so far. Also, a multi-swarm mechanism is used in which several instances of the first approach are run to locate potentially different feasible solutions. The proposed hybrid approach is applied to 18 standard constrained optimization benchmarks with up to 30 dimensions. Comparisons with two other state-of-the-art approaches show that the hybrid approach performs better in terms of finding feasible solutions and minimizing the objective function.

School/Discipline

Dissertation Note

Provenance

Description

Access Status

Rights

Copyright © 2013 ACM

License

Grant ID

Call number

Persistent link to this record