| Project oriented workflow is common in modern industry. Recently, the proliferation of Internet technology and globalization of business environment give rise to the advent of dynamic virtual alliances among complementary companies, thereby forming an extended virtual enterprise. Resource scheduling and rescheduling of multiple projects carried out in this situation cannot be effectively supported by existing centralized project scheduling approaches, mainly due to its dynamic and decentralized nature. The need for supporting such organizationally, geographically, and computationally distributed multiple projects (DMP) calls for a novel solution approach, which is distributed, collaborative, and flexible. In this research we have developed a market-based approach to address the DMP problem.; In the market-based DMP scheduling approach, the agents---representing project managers, tasks, resource managers, and resources---constitute a dynamic virtual economy, where multiple local markets are established and cleared over time. Like any other resource-constrained scheduling problem, the DMP scheduling problem also has to handle precedence constraints and resource constraints. In each local market, resource time slots are evaluated throughout by a combinatorial auction mechanism, generating resource-feasible local schedules for each resource. Meanwhile, in order to generate precedence-feasible schedules for each project, a tatonnement type procedure is applied to the temporary economy, which is a set of local markets. This iterative mechanism, called P-TATO, generates convergent and globally efficient DMP schedules. P-TATO is run in a multiagent based information system infrastructure. We design and implement the multiagent system (MAS) for the DMP control economy. The MAS design and implementation issues---including individual agent model, agent communication, and DMP control ontology---are thoroughly discussed. The solution qualities are verified through an intensive empirical analysis. The analysis results support the high levels of solution quality and computational efficiency of the mechanism.; Our approach can be directly applied to many project-oriented, decentralized scheduling problems. In addition, our problem formulation can also be considered as a decentralized extension of general resource allocation or scheduling problems. Application of our problem-solving model to other decentralized resource allocation and scheduling applications, including military logistics, supply chain management, and survivability analysis of networks are potential areas of future research. |
