Global optimization of MIQCPs with dynamic piecewise relaxations
- 152 Downloads
We propose a new deterministic global optimization algorithm for solving mixed-integer bilinear programs. It relies on a two-stage decomposition strategy featuring mixed-integer linear programming relaxations to compute estimates of the global optimum, and constrained non-linear versions of the original non-convex mixed-integer nonlinear program to find feasible solutions. As an alternative to spatial branch-and-bound with bilinear envelopes, we use extensively piecewise relaxations for computing estimates and reducing variable domain through optimality-based bound tightening. The novelty is that the number of partitions, a critical tuning parameter affecting the quality of the relaxation and computational time, increases and decreases dynamically based on the computational requirements of the previous iteration. Specifically, the algorithm alternates between piecewise McCormick and normalized multiparametric disaggregation. When solving ten benchmark problems from the literature, we obtain the same or better optimality gaps than two commercial global optimization solvers.
KeywordsMixed-integer nonlinear programming Global optimization of quadratic programs with bilinear terms Piecewise linear relaxations Optimality-based bound tightening
Support by Ontario Research Foundation, McMaster Advanced Control Consortium, and Fundação para a Ciência e Tecnologia (Projects IF/00781/2013 and UID/MAT/04561/2013), is gratefully appreciated.
- 9.Alnouri, S., Linke, P., El-Halwagi, M.M.: Spatially constrained interplant water network synthesis with water treatment options. In: Eden, M.R., Siirola, J.D.S., Towler, G.P. (eds.) Proceedings of the 8th International Conference on Foundations of Computer-Aided Process Design, pp. 237–242. Elsevier, Amsterdam (2014)Google Scholar
- 11.Koleva, M.N., Styan, C.A., Papageorgiou, L.G.: Optimisation approaches for the synthesis of water treatment plants. Comput. Chem. Eng. (2017)Google Scholar
- 34.Nagarajan, H., Lu, M., Yamangil, E., Bent, R.: Tightening McCormick relaxations for nonlinear programs via dynamic multivariate partitioning. In: Rueher, M. (ed.) Principles and Practice of Constraint Programming: 22nd International Conference, CP 2016, Toulouse, France, September 5–9, 2016, Proceedings, pp. 369–387. Springer, Cham (2016)CrossRefGoogle Scholar