Real-time dynamic selection algorithm of RCPSP scheduling priority rules based on deep learning

Abstract

<p>For the resource-constrained scheduling problem, a deep learning-based real-time dynamic selection<br>algorithm of scheduling priority rules is designed to minimize the project's makespan. Moreover, each scheduling stage selects priority rules in real-time for activity scheduling. Through constructing a deep neural network model, the mapping relationship between the project states and the best priority rule in each scheduling stage of the scheduled project is determined. Then the priority rule is dynamically selected for the scheduled project in real-time.<br>The final scheduling plan is obtained by combining the serial schedule generation scheme. Experimental research shows that the real-time dynamic selection priority rule algorithm outperforms the single priority rule heuristic and the hybrid priority rule heuristic covered in the paper and has better generalizability. In addition, compared with the meta-heuristic algorithm, the algorithm has a higher solution efficiency than the meta-heuristic.</p>

针对资源受限项目调度问题，以最小化项目完成时间为目标，设计基于深度学习的调度优先规则实时动态选择算法，在每个调度阶段实时选择优先规则进行活动安排。通过构建深度神经网络模型，确定已调度项目在各阶段的项目状态与最佳优先规则之间的映射关系，再据此为待调度项目实时动态选择优先规则，结合串行调度机制生成最终调度计划。 实验研究表明：实时动态选择优先规则算法表现优于文中所涉及的单一优先规则启发式算法及混合优先规则启发式算法，且具有更好的泛化性；此外，与元启发式算法相比该算法具有更高的求解效率。