Abstract: In practice, enterprises often implement multiple projects that are geographically dispersed in parallel. The implementation of these projects requires both shared enterprise resources and exclusive local resources. In the context of this kind of multi-project management, the decentralized resource-constrained multi-project scheduling problem (DRCMPSP) is proposed. In DRCMPSP, the frequent transfers of limited global resources among dispersed sub-projects often consume a lot of times, which have a great impact on the schedule of multiple projects. However, existing studies usually assume that resources can be transferred instantaneously, and the scheduling schemes in this case are often not feasible in the actual project implementation process. In this paper, the decentralized resource-constrained multi-project scheduling problem with transfer times (DRCMPSPTT) is studied based on dynamic multi-project environment. Different from the static project environment in the previous literatures, the continuous arrival and unpredictable arrival time of new projects are considered in this paper. The dynamic DRCMPSPTT consisted of several single projects that arrived over time, and these single projects are independent of each other. The finish-to-start precedence relationships with zero-lag existed only between the activities of a single-project. The execution of activities required several types of global and local resources, and the transfers of the global resources among different projects consumed times. The goal of DRCMSPSTT is to minimize the average project delay of multiple projects under the constraints of the priority relationship between the activities and the constraints of local and global resources limit. In view of the dynamic environment, the DRCMPSPTT model is built based on multi-agent system, and an auction mechanism with time windows is designed. The distributed multi-agent system based on auction with time windows (DMAS/ATW, in short) could allocate global resources for multiple projects under the dual constraints of the amounts of global resources and the transfer times. By solving a concrete DRCMPSPTT example, the dynamic scheduling process of DMAS/ATW algorithm is analyzed in detail.And numerical experiments are carried out based on the DRCMPSP instances in multi-project scheduling problem library MPSPLIB.The experimental results show that the average project delay obtained by DMAS/ATW algorithm can be reduced by up to 42% and 26% on average compared with that obtained by DMAS/RIA algorithm in relevant literature when there is no resource transfer time constraint. Moreover, for 1/3 instance sets, the results of DMAS/ATW algorithm with resource transfer time constraint are better than those of DMAS/RIA algorithm without the resource transfer time constraint, which verifies the superiority of DMAS/ATW algorithm in this paper. The experimental analysis of the size and global resource utilization factor of the instances shows that the DMAS/ATW algorithm has good adaptability to the examples with different scales and resource constraints degree. The research problem and methods in this paper provid reference value for the future research of multi-project scheduling problem in dynamic environment.

Key words: decentralized multi-project; resource transfer times; multi-agent system;auction mechanism; dynamic multi-project environment