Abstract: Economic globalization has promoted the development of multi-project distributed management and its management concept, and advanced information technology provides fast communication services between managers through the Internet, making distributed multi-project scheduling optimization possible. The distributed resource-constrained multi-project scheduling problem (DRCMPSP) is a project scheduling problem that studies the integrated optimization of single project scheduling and global resource coordination allocation in a distributed decision-making environment with multiple independent decision-makers and asymmetric information in multi-project scheduling. When the global resources are multi-skilled human resources, the distributed multi-skilled multi-project scheduling problem (DMSMPSP) is formed. As an extension of DRCMPSP, DMSMPSP involves the matching relationship between “activities-skills-resources” and the multi-skilled heterogeneous characteristics of global human resources in the global resource allocation process, which further increases the complexity of scheduling problems and the difficulty of research. Through the research of this paper, the depth and breadth of theoretical research on distributed multi-project scheduling problems are further strengthened. In practice, it provides a decision basis and methodological guidance for project managers to carry out multi-project scheduling in a distributed decision-making environment. Given the characteristics of the research problem, a two-layer model integrating local scheduling optimization and global coordinated decision-making is established based on the multi-agent system. In the local scheduling optimization model, each local decision-maker independently schedules the managed projects with the optimization objective of minimizing the completion time of single project. In the global coordinated decision-making model, the global decision-maker achieves the optimization goal of minimizing the multi-project total delay cost through reasonable allocation of human resources.
In addition, the local scheduling problem can be regarded as the classic resource-constrained project scheduling problem. A genetic algorithm based on forward-backward scheduling improvement is used to solve the local scheduling problem and generate the initial local baseline scheduling plans. The global coordination decision-making level primarily addresses the assignment of global multi-skilled human resources. An auction-based negotiation mechanism that takes into account the multi-skilled heterogeneous characteristics of human resources is designed. At each decision point, the global resource conflicts among multiple projects are coordinated through six stages: preparation stage, bid generation stage, temporary winner determination stage, bid modification stage, final winner determination stage, and local plan adjustment stage. We adapt the MPSPLIB instance sets and conduct experimental research. The research results indicate that the designed auction-based negotiation mechanism combined with the improved genetic algorithm can effectively coordinate the global resource allocation for problems of different sizes. Moreover, compared with the well-performing sequential game negotiation mechanisms in existing literature, the auction-based negotiation mechanisms can achieve lower average project delays, multi-project total durations, and multi-project total delay costs on most problem sets. Compared with the optimal solution obtained by the branch-and-bound algorithm, it indicates that the improved genetic algorithm can obtain high-quality initial local scheduling results. Furthermore, as the problem size or the intensity of resource conflicts increases, both the multi-project total delay cost and the number of auction rounds increase, indicating that the coordination and scheduling of multiple projects become more difficult, and the allocation process of human resources more complex.
In multi-project scheduling practice, the availability of global resources is often uncertain due to various unexpected and uncontrollable conditions, such as equipment failure. The distributed multi-skilled multi-project scheduling problem with uncertain global resource availability can be further investigated in the future. Moreover, the skill level of human resources is a dynamic concept, and the impact of learning or forgetting effects on the skill level of human resources can be further considered in research problems.

Key words: multi-project scheduling, distributed scheduling, multi-skilled resources, auction mechanism

摘要： 求解分布式多项目调度问题的关键是设计有效的资源协调机制以解决项目间的全局资源冲突。当全局资源为多技能人力资源时,调度项目活动开始时间的同时还需要满足“活动-技能-资源”的匹配关系,从而为多项目制定高质量的调度计划与全局资源指派方案。基于多Agent系统,建立局部调度优化与全局协调决策的双层模型;采用基于正向逆向调度改进的遗传算法求解初始局部调度计划;以人力资源各时段的使用权作为拍卖品,设计考虑人力资源多技能异质特点的拍卖谈判机制协调全局资源冲突。改编MPSPLIB算例集并开展实验研究,研究结果表明:本文设计的拍卖谈判机制与改进的遗传算法相结合可以有效协调不同规模问题的全局资源分配;拍卖谈判机制在测试问题集上的求解结果优于序贯博弈谈判机制;问题规模越大或资源冲突程度越强,项目协调调度难度越大,全局资源分配过程越复杂。

关键词: 多项目调度, 分布式调度, 多技能资源, 拍卖机制