Discrepancy-Based Method for Hierarchical Distributed Optimization

Distributed constraint optimization is increasingly used for problem solving by multiple agents. However, there are situations where the system is made up of heterogeneous agents, for which the context, the structure, and the business rules define the interactions that are possible between them. As an example, supply chains are made up of interdependent business units having some form of customer-supplier hierarchical relationships. The coordination space for these hierarchical situations can be described as a tree. Therefore, we propose a distributed algorithm (MacDS) that performs discrepancy-based search which is known to perform well for centralized problems. The proposed algorithm is complete and aims at producing good solutions in a short amount of time. It allows concurrent computation and is tolerant to message delays. It has been evaluated using real industrial supply chain problems, for which it showed good performance.