Modeling the Management of Water Resources Systems Using Multi-Objective DCOPs
Francesco Amigoni, Andrea Francesco Castelletti, Matteo Giuliani · Virtual Community of Pathological Anatomy (University of Castilla La Mancha) · 2015
Multi-agent systems are being increasingly used in environmental modeling applications to characterize human behavior and interactions with natural processes. A model based on Distributed Constraint Optimization Problems (DCOPs) has been recently proposed for studying the management of water resources systems from the point of view of a regulating institution in charge of coordinating multiple distributed decision-makers (agents). However, this DCOP-based model does not explicitly account for the variety of stakeholders' and regulators' interests that are generally involved, representing incommensurable and often competing objectives. In this paper, we provide a Multi-Objective DCOP (MODCOP) model that supports distributed water resources management through the exploration of tradeoffs across different agents' objectives. Among the available algorithms for solving the resulting MO-DCOP, we choose a variant of the B-MOMS algorithm because it allows identifying (an approximation of) the whole Pareto frontier for the problem. Experimental results conducted on a number of randomly generated water systems show that the approximation introduced by the solving algorithm is limited for most of the systems.