Issue |
RAIRO-Oper. Res.
Volume 54, Number 2, March-April 2020
|
|
---|---|---|
Page(s) | 451 - 469 | |
DOI | https://doi.org/10.1051/ro/2019096 | |
Published online | 27 February 2020 |
A metaheuristic penalty approach for the starting point in nonlinear programming
1
Modestya Research Group, Department of Statistics, Mathematical Analysis and Optimization, Institute of Mathematics (IMAT), University of Santiago de Compostela, Santiago, Spain
2
Centro de Matemática e Aplicações (CMA), FCT, UNL, 2829-516 Caparica, Portugal
* Corresponding author: david.rodriguez.penas@usc.es
Received:
20
March
2019
Accepted:
29
September
2019
Solving nonlinear programming problems usually involve difficulties to obtain a starting point that produces convergence to a local feasible solution, for which the objective function value is sufficiently good. A novel approach is proposed, combining metaheuristic techniques with modern deterministic optimization schemes, with the aim to solve a sequence of penalized related problems to generate convenient starting points. The metaheuristic ideas are used to choose the penalty parameters associated with the constraints, and for each set of penalty parameters a deterministic scheme is used to evaluate a properly chosen metaheuristic merit function. Based on this starting-point approach, we describe two different strategies for solving the nonlinear programming problem. We illustrate the properties of the combined schemes on three nonlinear programming benchmark-test problems, and also on the well-known and hard-to-solve disk-packing problem, that possesses a huge amount of local-nonglobal solutions, obtaining encouraging results both in terms of optimality and feasibility.
Mathematics Subject Classification: 49M37 / 90C59
Key words: Nonlinear programming problems / Starting point strategy / Metaheuristics / Penalty methods / Disk packing problem
© EDP Sciences, ROADEF, SMAI 2020
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.