关键词: 生鲜电商, 前置仓选址, 两阶段随机规划, 样本平均近似, Benders分解

Abstract:

With the popularization of online shopping and the implementation of stay-at-home orders， China’s fresh food e-commerce market has been growing rapidly and it has changed our buying habits for fresh food. Currently， there are mainly three forms of fresh food e-commerce in China， namely front warehouse， in-store as warehouse， and community buying group. Among them， as a popular form， front warehouse plays a vital role in ensuring the freshness and on-time delivery rates of fresh foods because it can well address the last three-kilometer delivery problem. However， high investment cost of front warehouses in warehouses location， order fulfillment and inventory holding has become one of the main bottlenecks in restricting its further development for fresh food e-commerce. To deal with the problem with front warehouses， the front warehouse location and order fulfillment problem for fresh e-commerce are addressed， considering the uncertainties in fresh product demands and the shelf-life constraints of fresh products. This problem can be formalized as a two-stage stochastic programming model where the warehouse location and inventory replenishment decisions can be made in the first stage before the realization of uncertain customer demands， and the order fulfillment decisions are made in the second-stage after uncertain customer demands are observed. Due to the computational difficulties and non-linearity in solving the two-stage stochastic programming model， a sample-average-approximation based Benders decomposition algorithm （SBD） is proposed to transform the stochastic model into a sample approximation model by using Latin hypercube sampling method. As a result， this approximation model is a mixed integer programming model and thus can be solved by Benders decomposition algorithm. Finally， a case study about a fresh food e-commerce company in Shanghai， China， which aims to deploy a front-warehouse distribution network for online fresh products， is used to verify the feasibility and effectiveness of the proposed algorithms. It demonstrates that the presented two-stage stochastic programming model can effectively reduce order fulfillment costs for fresh food e-commerce when uncertainties are dealt with. Furthermore， the experimental results reveal that the SBD algorithm performs better than the commercial solver CPLEX， both in small-scale instances and large-scale instances. In addition， the sensitivity analysis indicates that the unit holding cost， expired cost and shortage cost have a significant effect on total order fulfillment cost for fresh food e-commerce. In summary， the proposed two-stage stochastic programming model and corresponding SBD algorithms can well handle the decisions problem with front warehouse locations and order fulfillment for online fresh food e-commerce when uncertainties are encountered.

Key words: fresh food e-commerce, front warehouse, two-stage stochastic programming, sample average approximation, Benders decomposition