On the acceleration of global optimization algorithms using a cutting plane decomposition and machine learning
A. Marousi, A. Kokossis
Computer Aided Chemical Engineering, Volume 50, pp. 617-623, 2021
global optimization; outer- approximation; cutting planes; quadratic programming; machine-learning
The paper presents an accelerated approach for global optimization that is applied on quadratic programming problems. The global optimization is based on a decomposition method using cutting planes that are generated, analysed, and screened using advanced analytics. The work capitalizes and builds on innovations by Baltean-Lugojan et al. (2019) who recently presented a generic and effective outer approximation method suitable for semidefinite relaxations. The use of data analytics is applied in populations (P) of cutting planes, experimenting with different metrics and clustering methods. The proposed approach achieves a reduction in the integrality gap by 18-30% with the largest reductions in relation to larger problems (jumbo problems: 100 variables, 75% density).