Constrained NLP via gradient flow penalty continuation: Towards self-tuning robust penalty schemes

Felipe Scott, Raúl Conejeros, Vassilios S. Vassiliadis*

*Autor correspondiente de este trabajo

Producción científica: Contribución a una revistaArtículorevisión exhaustiva

2 Citas (Scopus)

Resumen

This work presents a new numerical solution approach to nonlinear constrained optimization problems based on a gradient flow reformulation. The proposed solution schemes use self-tuning penalty parameters where the updating of the penalty parameter is directly embedded in the system of ODEs used in the reformulation, and its growth rate is linked to the violation of the constraints and variable bounds. The convergence properties of these schemes are analyzed, and it is shown that they converge to a local minimum asymptotically. Numerical experiments using a set of test problems, ranging from a few to several hundred variables, show that the proposed schemes are robust and converge to feasible points and local minima. Moreover, results suggest that the GF formulations were able to find the optimal solution to problems where conventional NLP solvers fail, and in less integration steps and time compared to a previously reported GF formulation.
Idioma originalInglés
Páginas (desde-hasta)243-258
Número de páginas16
PublicaciónComputers and Chemical Engineering
Volumen101
DOI
EstadoPublicada - 2017

Nota bibliográfica

Publisher Copyright:
© 2017 Elsevier Ltd

Palabras clave

  • Convergence analysis
  • Gradient flow
  • Nonlinear programming problem

Huella

Profundice en los temas de investigación de 'Constrained NLP via gradient flow penalty continuation: Towards self-tuning robust penalty schemes'. En conjunto forman una huella única.

Citar esto