Project Details
Projekt Print View

Mixed integer nonlinear multiobjective optimization by outer approximations

Subject Area Mathematics
Term from 2020 to 2023
Project identifier Deutsche Forschungsgemeinschaft (DFG) - Project number 432218631
 
Final Report Year 2023

Final Report Abstract

The aim of this project was to develop a deterministic algorithmic solution procedure for multiobjective mixed-integer nonlinear optimization problems. We realized such a procedure for multiobjective mixed-integer convex optimization problems, which were the main focus of this project, with the Hybrid Patch Decomposition algorithm (HyPaD). This algorithm decomposes the mixed-integer problem into certain continuous subproblems by fixing the values of the integer variables. For those continuous problems it iteratively computes weakly nondominated points using techniques from multiobjective continuous optimization. These points are used to compute and improve a global upper bound set for the overall nondominated set of the mixed-integer problem as well as lower bound sets for the nondominated sets of the continuous subproblems. These lower bound sets for the subproblems only yield a global lower bound set in case all possible integer fixings are enumerated. In order to avoid such a full enumeration - especially for a large number of integer variables - simultaneously a global lower bound set is computed. For this, we solve linear relaxations of the original mixed-integer problem that are iteratively refined. Scalarizations of these relaxations yield single-objective mixed-integer linear optimization problems for which fast solvers are available. Hence, the HyPaD algorithm combines techniques from multiobjective continuous and single-objective mixedinteger linear optimization for solving multiobjective mixed-integer convex optimization problems efficiently. Since all techniques within the algorithm work entirely in the criterion space, problems with a large number of optimization variables can be solved. Besides proving correctness and finiteness of the algorithm, extensive numerical tests have been performed as well. For this, we also developed a generator for test instances for multiobjective mixed-integer nonlinear optimization problems. With the multiobjective mixed-integer branch-and-bound algorithm MOMIBB, we even developed a solution procedure for multiobjective mixed-integer nonconvex optimization problems. This algorithm makes use of subproblems based on a geometric branch-andbound procedure in the decision space. Attainable points of subproblems are used to compute a global upper bound set. What is more, the algorithm computes lower bounds for the nondominated sets of each of the subproblems. Also for this algorithm, we proved correctness and finiteness and demonstrated its capabilities in numerical tests. An implementation of all algorithms developed in the scope of this project is publicly available on GitHub.

Publications

 
 

Additional Information

Textvergrößerung und Kontrastanpassung