Combining DCOP and MILP for Complex Local Optimization Problems

Fernanda N. T. Furukita, Fernando J. M. Marcellino, Jaime Simão Sichman

Abstract: Supply chain management, which is composed of interdependent entities that have defined roles and responsibilities, shows several characteristics in common with Multi-Agent Systems (MAS). This type of problem may be divided into several local subproblems, which can be optimized separately. However, in general, the full problem cannot be solved in a centralized way due to its complexity or the need for information privacy. This work presents a distributed heuristic method which provides an acceptable optimization of this type of complex problem when compared to the centralized approach available for the considered instances, and better than a similar approach in the literature. It is based on modeling the considered problem first as a Distributed Constraint Optimization Problem (DCOP), and then by integrating it with Mixed-Integer Linear Programming (MILP) optimization models of its subproblems. We have obtained a value which is about 5% better than a similar distributed method in the literature and only about 7% worse than the actual optimum one. We consider a promising approach for increasingly real settings.