一种生产销售系统的生产及库存控制优化研究

发布时间：2018-11-19 07:39

【摘要】：供应链上下游企业可以视为买方与卖方的关系,上游企业不考虑下游企业的需求容易导致企业之间的供需失衡。因此研究上下游企业之间的生产销售问题,以实现长期运行情况下每个企业利润最高,对于实际供应链有重要意义。首先,论文建立了一种基于单零售商需求驱动的生产销售系统。该系统包括一个上游企业(生产商)和一个下游企业(零售商),两个企业拥有独立的库存单元和决策过程。生产商可以视为基于需求驱动的传送带给料加工站系统,成品库库存状态受零售商补货订单和自身生产能力共同影响。生产商的控制变量为加工站点的前视距离,优化目标是寻找到最优的前视控制策略,考虑到系统中工件到达时间、工件加工时间难以准确获知,论文采用一种与模型无关的强化学习算法对生产商的最优或者次优前视控制策略进行求解。零售商采取动态补货策略,库存状态受随机的顾客需求和生产商的实际补货量共同影响。零售商的控制变量为补货点,优化目标是寻找到最优的库存控制策略,由于顾客需求的动态随机性,论文采用强化学习算法对零售商的优化控制问题进行求解。考虑到实际情况中生产商通常给多个零售商补货,论文进一步研究了一种多零售商生产销售系统的优化控制。系统中多个零售商之间存在静态博弈关系,即每个零售商的补货需求在时间上有先后顺序,并且各零售商不知道其余零售商的补货策略。生产商根据补货订单的时间顺序即时完成配货后,成品库状态相应发生改变。最后论文采用强化学习算法分别对生产商和多个零售商的最优或者次优控制策略进行求解,并从顾客到达率、仓库容量和补货期三方面对多零售商之间的博弈关系进行了分析。
[Abstract]:The upstream and downstream enterprises of supply chain can be regarded as the relationship between buyer and seller, and the upstream enterprises do not consider the demand of downstream enterprises, which leads to the imbalance of supply and demand between enterprises. Therefore, it is of great significance to study the production and sales problems between upstream and downstream enterprises in order to realize the highest profit of each enterprise in the long run. Firstly, a production and sales system based on demand driven by single retailer is established. The system consists of an upstream enterprise (manufacturer) and a downstream enterprise (retailer). The two enterprises have independent inventory units and decision-making process. The manufacturer can be regarded as a demand-driven delivery and delivery station system, and the inventory status of the finished product store is affected by the retailer's replenishment order and its own production capacity. The manufacturer's control variable is the forward distance of the processing station. The optimization goal is to find the optimal forward view control strategy. Considering the arrival time of the workpiece in the system, the processing time of the workpiece is difficult to be accurately determined. In this paper, a model independent reinforcement learning algorithm is used to solve the optimal or suboptimal forward view control strategy. The retailer adopts dynamic replenishment strategy and the inventory status is affected by the random customer demand and the actual replenishment volume of the manufacturer. The retailer's control variable is restocking point, and the optimization goal is to find the optimal inventory control strategy. Because of the dynamic randomness of customer demand, the reinforcement learning algorithm is used to solve the optimal control problem of retailer. Considering that manufacturers usually replenish more than one retailer in practice, this paper further studies the optimal control of a multi-retailer production and sales system. There is a static game relationship among many retailers in the system, that is, the replenishment demand of each retailer has an order in time, and each retailer does not know the replenishment strategy of the other retailers. The state of the finished goods warehouse changes immediately after the manufacturer completes the distribution according to the time order of the replenishment order. Finally, the reinforcement learning algorithm is used to solve the optimal or sub-optimal control strategy of manufacturers and multiple retailers, and the game relationship between multiple retailers is analyzed from three aspects: customer arrival rate, warehouse capacity and replenishment period.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F274

【参考文献】