【作者】 刘琳琳；

【导师】 刘锋；

【作者基本信息】 东北财经大学 ， 管理科学与工程， 2022， 硕士

【摘要】 在全球制造业转型升级的时代背景下,我国制造业市场竞争日趋激烈。新的内外部环境对企业生产管理模式提出了更高的要求,主要表现为降低生产成本、提高产品质量、缩短产品生产周期、快速响应客户多样化需求等。制造企业不仅要在技术手段上取得进步,而且要着重提高生产效率。生产调度作为制造车间提高生产效率的核心内容,自然地受到了制造业、学术界的广泛关注。而生产车间作为产品制造的直接执行者,承载着大量的生产任务,同时也是大量实时信息的交汇地。随着车间生产系统规模的扩大和运行复杂性的提高,不确定性因素急剧增加,这些因素直接影响了调度方案的可行性与科学性,而不确定加工时间作为最重要的不确定因素,如何对其进行表征并制定科学有效的调度方案尤为重要。准时化(JIT)生产的基本思想是按交货期对工件的完工时间进行严格把控。在准时生产调度环境中,早于或晚于交货期完成的作业将受到惩罚。因此,理想的调度计划是所有工件在交货期准时完成。因其能够有效降低企业生产过程成本,提高生产效率并且对生产过程进行精益化管理,准时化生产成为制造型企业追求的主要目标之一。基于以上两方面的考虑,本文针对生产调度对准时化生产、降低成本、精益化管理等实际需要,提出了以交货期为中心,以工件准时完工和准时交付为目标,考虑不确定工件加工时间,针对单机提前/拖期(E/T)问题进行研究。通过对不确定工时下的单机提前/拖期(E/T)调度问题的的分析,将其分解为工件的最优开工时间的确定和工件最优加工次序的确定两个层次,其对应的两个核心问题为不确定加工时间的处理和机器上工件加工顺序的确定。首先根据先到先服务原则,在确定次序下提出了一种处理时间不确定的单机调度的分布鲁棒优化(DRO)模型,假设每个工件的加工时间是给定分布族内的一个未知的随机变量,用均值和方差信息来描述。所提出的DRO模型旨在确定每个工件的最优开工时间,以最小化预期的最坏情况下的总提前/拖期完工成本。为了确定工件最优加工次序,使遗传算法在外层探索可行的调度空间以优化机器上工件加工序列,而在内层则使用所提出的分布鲁棒优化模型来确定满足约束条件的最佳连续解。并将根据问题的性质,将分布鲁棒单机调度模型的目标函数值的倒数作为染色体适应值。最终求得各个工件的最优开工时间和机器上工件的最优加工顺序。最后在数值实验部分证明了该算法能够有效解决该调度问题,通过实验数据证明了在分布族中合并协方差信息可以显著改善目标函数值,并且对遗传算法部分的参数进行灵敏度分析,设置不同的参数取值确定使实验结果最佳的参数。目前分布鲁棒优化方法被广泛使用于能源、金融领域,生产调度领域的研究还略显贫乏。而使用分布鲁棒优化方法在研究不确定工时下的单机调度问题时一般是建立分布鲁棒优化模型后求解模型,但对于以最小化总提前/拖期完工成本为优化目标的调度问题,其模型包含变量较多,模型复杂且求解难度大。因此,为求解该调度问题,本文将可处理的分布鲁棒优化模型与遗传算法相结合来共同求解,本文研究为不确定工时条件下的作业车间调度问题提供新思路。更多还原

【Abstract】 In the background of the transformation and upgrading of the global manufacturing industry,the competition in Chinese manufacturing market is becoming increasingly fierce.The new internal and external environment puts forward higher requirements for the production management mode of enterprises,which is mainly manifested in reducing the production cost,improving the product quality,shortening the product production cycle,and quickly responding to the diversified needs of customers.Manufacturing enterprises should not only make progress in technological means,but also focus on improving production efficiency.Production scheduling,as the core content of manufacturing workshop to improve production efficiency,production scheduling has naturally attracted wide attention from manufacturing industry and academic circles.As the direct executor of product manufacturing,the production workshop carries a large number of production tasks,and is also the intersection of a large number of real-time information.With the expansion of the workshop production system scale and the improvement of the operation complexity,the uncertainty factors have increased sharply,these factors directly affect the feasibility and scientificity of the scheduling scheme,and the uncertain processing time as the most important uncertainty factors,how to represent it and develop a scientific and effective scheduling scheme is particularly important.Just-in-time(JIT)production is one of the main goals of manufacturing enterprises,and is also an important means of low-cost operation and lean management.On-time production mode requires strict processing according to the specified time,and emphasis on time delivery.Based on the above two considerations,according to the actual needs of production scheduling alignment with time production,cost reduction,lean management,this paper puts forward the delivery time as the center,with workpiece completion on time and on-time delivery as the goal,focusing on the single machine earliness/tardiness(E/T)problem under uncertain working hours conditions.Through the analysis of the single-machine earliness/tardiness(E/T)scheduling problems under uncertain working hours,it is decomposed into the determination of the optimal starting time of workpiece and the determination of the optimal processing order of workpiece.The corresponding two core problems are the processing of uncertain processing time and the determination of the workpiece processing sequence on the machine.First of all,according to the principle of first-to-first service,a distributionally robust optimization(DRO)model of single-machine scheduling with uncertain processing time is proposed under the determination order.Assuming that the processing time of each artifact is an unknown random variable within a given family of distribution,which is described by mean and variance information.The proposed distributionally robust optimization model is designed to determine the optimal start time for each artifact to minimize the expected worst-case total earliness/tardiness completion costs.In order to determine the optimal artifact processing order,the genetic algorithm can explore the feasible scheduling space in the outer layer to optimize the workpiece processing sequence on the machine,while in the inner layer,the proposed distributionally robust optimization model is used to determine the optimal continuous solution satisfying the constraints.The inverse of the objective function value of the distributionally robust single-machine scheduling model will also be taken as the chromosome adaptation value according to the nature of the problem.Finally,the optimal start time of each workpiece and the optimal processing order of the workpiece on the machine are obtained.Finally,in the numerical experiment part proved that the algorithm can effectively solve the scheduling problem,through the experimental data proved that combining the covariance information in the distribution family can significantly improve the target function value,and to analyze the sensitivity of the parameters of the genetic algorithm part,set different parameters to determine the best experimental results.At present,distributionally robust optimization method is widely used in energy and finance fields,and the research in production scheduling field is slightly poor.However,when the distributionally robust optimization method is used to study the single-machine scheduling problem under uncertain working hours,the model generally establishes the distributionally robust optimization model.However,for the scheduling problem of minimizing the total earliness/tardiness completion cost,the model includes many variables,and the model is complex and difficult to solve.Therefore,this paper combines the tractable distribution robust optimization model with the genetic algorithm to solve the scheduling problem,which provides a new idea for the operation workshop scheduling problem under the condition of uncertain working hours.更多还原