摘要
本文针对动态释放、异质性强的B2C电商波次订单的物流配送决策难题,以提高车辆满载率、降低配送成本为目标,研究考虑波次订单动态释放的B2C电商订单城区物流合并配送问题。首先,将原动态问题建模为一阶马尔可夫决策过程以表达其多阶段时序序贯决策的特点;其次,基于状态转移具有的时序特征,提出一种基于时序预测的前向动态规划方法用于寻找最优策略,将时序预测信息融入到合并配送决策的模型中,并结合定性启发式规则与定量优化模型以兼顾方法的决策效率与优化能力;最后,基于标准算例下的数值实验和某B2C电商平台的实例分析,验证了所提方法的有效性和实用性。此研究可为B2C电商订单物流配送提供决策支持,同时对于状态转移具有时序特征的马尔可夫决策问题提供了求解的新思路。
B2C e-commerce has explosively grown during the epidemic and post-epidemic era due to its massive storage SKUs(Stock Keeping Units),which can mostly cover all aspects of home life.However,these massive storage SKUs which can be purchased in any combination make B2C e-commerce orders have the characteristics of heterogeneous,small volume,multiple batches,and dynamic arrival.Therefore,the wave-picking method is mostly applied in a warehouse to improve the order-picking efficiency by increasing the density of SKUs to be picked,which brings new challenges to order distribution decisions.These challenges include incomplete information for order distribution decisions,high uncertainty about the forthcoming wave-picking orders,inconsistency of order production and distribution,etc.Therefore,this paper focuses on the order consolidated distribution decision-making problem considering the dynamic release of the forthcoming wave-picking orders with the incomplete decision information.Due to the multi-stage sequential feature of the B2C e-commerce wave-picking order consolidated distribution problem,we first transform the original dynamic decision-making problem into a first-order MDP model and then analyze its complexity.Secondly,in view of the difficulty of solving the model with high complexity,based on the data trend analysis,we deduced that the system states of the model are time-series transformed.Then,a forward dynamic programming method based on time-series forecasting is designed to solve this model.In a single decision phase,the NP-complete hard original problem is transformed into the two-stage“Pack_forecast+TSPTW”quantitative optimization model with lower complexity through qualitative heuristics,which can realize the rapid generation of the consolidated distribution plan.Finally,we verified the effectiveness and efficiency of the proposed method by numerical experiments and case analysis.In the Section 1 of the paper,due to the features of multi-stage sequential and no after-effect decisions in the process of order distribution,the original dynamic decision-making problem is transformed into a first-order MDP model.Then,the decision stage,the system state,the decision set,the transition probability,the decision cost,and the objective function of the MDP model are given.The complexity of solving the model is analyzed as well.In the Section 2 of the paper,due to the high complexity of solving the MDP model,based on the trend analysis of historical data,we deduced that the release volume of wave-picking orders is time-series related,which makes the system state transformation of the MDP model is time-series related.Then,we design a forward dynamic programming method based on time-series forecasting to search the optimal policy of the MDP model.We use qualitative heuristics to transform the NP-complete hard model of a single decision phase into the two-stage quantitative optimization model,“Pack_forecast+TSPTW”,with lower complexity.The ARIMA time-series forecasting method is used to predict the release volume of forthcoming wave-picking orders.Finally,the consolidated distribution plan can be generated quickly by incorporating the time-series forecasting information into the two-stage model.In the Section 3 of the paper,we first adapt the benchmark instances of classic VRPTW to our problem.Then,it is verified on small-scale instances that the consolidated distribution of wave-picking orders is superior to the separated distribution,and the proposed method has the ability to approach the lower bound obtained using complete decision information.At last,we conducted a real case study based on the real data of a large B2C e-commerce platform.The practicability of the proposed method and the effectiveness of dynamic decision-making based on time-series forecasting information are demonstrated through comparative experiments.In summary,although the wave-picking production mode can effectively deal with the characteristics of dynamic arrival and strong heterogeneity of B2C e-commerce orders,it brings new challenges to the distribution of wave-picking orders.To solve the consolidated distribution decision-making problem of B2C e-commerce wave-picking orders,this paper constructs a multi-stage sequential decision-making MDP model and designs a forward dynamic programming approach based on time-series forecasting to solve the model.The research results show that the combination of qualitative heuristics and quantitative optimization models can effectively balance the timeliness of the decision and the optimality of the decision’s objective,which can facilitate the obtainment of a sub-optimal solution of complex dynamic problems in a short time.Besides,it is demonstrated that decision-making problems with time-series sequential features can be effectively solved by means of short-term time series forecasting.Our research can provide decision-making support for solving e-commerce order consolidated distribution problem with the feature of dynamic order release,and can be a reference for complex dynamic problems with multi-stage sequential decision-making as well.
作者
石海洋
孙丽君
胡祥培
SHI Haiyang;SUN Lijun;HU Xiangpei(School of Economic and Management,Dalian University of Technology,Dalian 116024,China)
出处
《管理工程学报》
CSSCI
CSCD
北大核心
2024年第2期152-165,共14页
Journal of Industrial Engineering and Engineering Management
基金
国家自然科学基金项目(71971037、71971036)
中国留学基金委项目(201906060096)。
关键词
B2C电商
波次拣选
合并配送
时序预测
前向动态规划
B2C e-commerce
Wave-picking
Consolidated distribution
Time-series forecasting
Forward dynamic programming
作者简介
通讯作者:孙丽君(1979-),女,山东烟台人,大连理工大学经济管理学院教授,博士生导师,研究方向:数据驱动的智能决策、物流系统优化等。
