Abstract: Dynamic optimization of cloud service renting configuration is the key to achieving efficient,high-quality and on-demand service formassive stochastic manufacturing tasks in cloud manufacturing.Manufacturing tasks and manufacturing services were analyzed and defined in the cloud manufacturing environment,and the optimization objectives were set in three aspects: service renting cost,service utility and task delay.Under the condition of capacity constraints,the re-scheduling mode triggered by timing length or fixed number of tasks was established.LSTM-based demand forecasting approach was adopted to realize the forward-looking dynamic scheduling of cloud manufacturing services and the optimization for renting configuration.The experimental results showed that the demand forecasting based online-scheduling could reduce the service renting cost by 4.50%,improve the service utility by 0.86% and reduce the value of average task delay by 5.15% in comparison with the method without forecasting.

Key words: cloud manufacturing, online scheduling, demand prediction, deep learning, service renting configuration

摘要： 鉴于云服务租赁配置动态优化是云制造模式下实现大量随机制造任务高效、高质、及时按需服务的关键,对云制造环境下的制造任务和制造服务进行分析定义,并以服务租赁成本、服务使用效用和任务延期值3个指标为优化目标,考虑制造服务容量的约束条件,建立了基于定时长或定任务数触发的重调度模型;通过基于长短期记忆方法的后续云服务需求容量预测,实现前瞻性的云制造服务动态调度及其租赁配置优化。实验结果表明,相比不带预测的方法,基于需求预测的在线调度方法在统计意义上可以缩减服务租赁成本4.50%,提高服务使用效用0.86%,缩减任务延期率5.15%。

关键词: 云制造, 在线调度, 需求预测, 深度学习, 服务租赁配置