带有资源依赖学习效应的多模态项目调度交互式群智能方法

Resource-dependent learning effects are not only an unsolved problem widely existing in the MRCPSP (multi-mode resource-constraint project scheduling problem), but also the primary factor affecting the correctness of processing time. Processing time and resource availability are key to the effectiveness of MRCPSP. The proposal mainly focuses on the critical problems in MRCPSP, which are learning effect modelling, dynamic resource allocation, and multi-mode multi-objective optimizaiton. The experience value of a resource, the fundamental parameter of learning effects, is obtained by evaluating the historically used information. The linked-list available time-fragment is introduced to dynamically distribute and recover resources. The real-time allocation and estimation strategy is developed. Appropriate available fragments are selected and combined into a whole one for the requirement. The resource-dependent learning effect model is built according to the experience value and the learning ability of a resource.An induced learning effect model is introduced for resources have variable learning ability. The internal related learning effects among different resources are presented according to their modes in an activity. As well, the external related learning effect is analyzed for resources shared by serial activities and concurrent activities. Based on the learning effect models, the comprehensive mathematical model is constructed for each resource to accurately estimate the processing time. The resource-constrained multi-mode project scheduling model with the resource- dependent learning effect is investigated to minimize cost and makespan with time-coast trade-off, for which an interactive swarm intelligence optimization algorithm is proposed. Accurate processing time and available resources could be obtained. Cost of the whole project could be reduced and reaource utilization could be improved, greatly.

资源依赖学习效应不仅是广泛存在于项目调度但尚未研究的新问题，也是影响加工时间准确性的根源；加工时间和资源可用状况又是保证多模态项目调度高效的关键。本课题针对学习效应建模、资源动态分配、多模态多目标优化等项目调度的关键问题，提出依据被使用历史信息评估资源熟练度的方法，设计基于链式可用时间段的资源快速分配和回收机制，构建可用资源 "碎片""按需"整合的评估策略；根据具有不同学习能力资源的学习率和熟练度，构造资源依赖学习效应模型，建立可改变学习能力资源的诱导式学习效应模型；考虑活动内各模态对应资源间的内相关学习效应、共享资源的串行活动和并行活动的外相关学习效应，构建准确估计加工时间的数学模型；考虑时间-成本权衡的总成本和最大完工时间最小化问题，建立资源依赖学习效应的多模态项目调度模型，提出多模态项目调度的并行交互式群智能算法。获取准确的加工时间和资源可用状态，降低项目成本、提高资源利用率。

带有资源依赖学习效应的多模态项目调度是广泛存在于复杂工程应用但尚未研究的新问题。本课题主要研究准确的加工时间估计、高效的多模态项目调度。以资源为核心，实现资源的全局化管理；提出一种同时考虑学习和（或）恶化效应的模型，将学习和恶化效应函数建模为关于位置和累积时间的函数，优化考虑恶化效应的单机成组调度；针对可置换两机流水作业调度问题，提出通用学习效应模型和带先验知识学习和遗忘效应的流水调度问题模型和优化方法；将带有学习效应的调度问题转化为带有学习效应的最短路径问题，提出一种求解带有学习效应最短路径问题的高效启发式精确算法；考虑时间-成本权衡的总成本和最大完工时间最小化问题，建立资源依赖学习效应的多模态项目调度模型；基于资源可用状态和学习效应确定的加工时间，提出多目标项目调度的智能优化算法。