An Efficient Algorithm For Solving Dynamic Complex DCOP Problems

Abstract

Multi Agent Systems and the Distributed Constraint Op- timization Problem (DCOP) formalism o®er several asyn- chronous and optimal algorithms for solving naturally dis- tributed optimization problems e±ciently. There has been good application of this technology in addressing real world problems in areas like Sensor Networks and Meeting Schedul- ing. Most of these algorithms however exploit static tree structures and are thus not well suited to modeling and solving problems in rapidly changing domains. Also, while in theory most DCOP algorithms can be extended to han- dle complex local sub-problems, we argue that this gener- ally results in making their performance sub-optimal, and thus their application less suitable. In this paper we present new measures that emphasize the interconnectedness be- tween each agent's local and inter-agent sub-problems and use these measures to guide dynamic agent ordering during distributed constraint reasoning. The resulting algorithm, DCDCOP, o®ers a robust, °exible, and e±cient mechanism for modeling and solving dynamic complex problems. Ex- perimental evaluation of the algorithm shows that DCD- COP signi¯cantly outperforms ADOPT, the gold standard in search-based DCOP algorithms