基于数据驱动的生鲜电商联合补货与配送协同优化研究

How to expand profit margins by reducing operational costs has become a problem plaguing the development of fresh products E-commerce industry, and the arrival of big data era provides opportunities for them to further optimize the inventory replenishment and delivery decisions. Integrating data-driven approaches to conduct forecasting, this project studies the coordinated multi-product replenishment-delivery optimization model and algorithm by taking the advantages of online data. This project mainly includes: (i) Considering the diversity and uncertainty of online demand of fresh products, a demand forecasting method driven by clickstream data is proposed, and a forecasting model based on convolutional-Gated Recurrent Unit neural network and its corresponding training method are developed, respectively. (ii) Considering the deterioration and short fresh-time of fresh products, a coordinated joint replenishment and delivery model based on data-driven is developed, and the investment on preservation technology is also taken into account. (iii) Considering the high-dimensional and strong non-linear characteristics of the collaborative optimization model, a novel intelligent algorithm called improved teaching-learning based optimization (TLBO) algorithm is developed to solve this complex coordinated model after targeted improvements. A large number of numerical experiments are carried out to verify the effectiveness of the proposed algorithm. (iv) The proposed data-driven coordinated model and the TLBO algorithm are applied to the typical fresh product online retailer for the development of decision support system, providing guidelines for online retailers to take advantage of data in the process of optimizing inventory replenishment and delivery, then the operational costs of fresh product online retailer could be reduced.

如何通过降低运营成本来拓展利润空间成为困扰生鲜电商发展的难题，数据时代的到来为其进一步优化库存补货与配送过程创造了条件。本项目利用生鲜电商的网络数据优势，在集成数据驱动需求预测的基础上，研究生鲜电商多品种补货-配送协同优化模型与求解算法。主要内容包括：(1)针对网络环境下生鲜产品需求多样化和不确定性特征，设计点击流数据驱动的需求预测方法，构建基于深度门控循环单元的神经网络预测模型及训练方法；(2)针对生鲜产品易变质和保鲜期短的固有特点，考虑保鲜成本投入，构建基于数据驱动的生鲜电商联合补货与配送协同优化模型；(3)针对所构建协同优化模型的高维强非线性特征，提出稳定可靠、通用性强的改进教学优化算法以求解模型，通过大量数值实验验证所提算法的有效性；(4)将所提模型和算法应用于典型生鲜电商企业，辅助其开发决策支持系统，为企业利用数据优势来优化补货与配送决策提供帮助，进而降低生鲜电商的运营成本。

数据时代的到来为生鲜电商供应链的蓬勃发展提供了机会，然而，生鲜电商企业依然在微薄的利润下艰难生存，亟需通过数据赋能来进一步优化运营成本。本项目针对网购生鲜产品的多样性和随机性特征，利用生鲜电商可获得大规模多维度数据的优势，在数据驱动生鲜产品需求预测的基础上，优化占据生鲜电商主要运营成本的补货与配送环节，构建补货-配送协同优化模型，进一步研究模型优化求解算法，以降低生鲜电商的运营成本。项目研究了四个方面的内容：（1）针对网络环境下生鲜需求的多品种和不确定特征，构建基于历史需求数据驱动的双向长短时记忆神经网络需求预测模型和基于网络点击流数据驱动集成的回声状态网络需求预测模型；（2）针对生鲜产品在补货、库存、配送过程中均需进行保鲜管理的特点，构建生鲜电商多产品联合补货与配送协同优化模型；并结合生鲜电商运营实际，将模型扩展为考虑信用支付和多仓库情形下的协同模型；（3）针对所构建协同优化模型的高维强非线性难题，在分析最优解的性质的基础上，设计高效率高质量的智能优化算法以求解模型，通过实际数据和大量随机数据实验验证了所提算法的有效性；（4）将所提模型和算法提供给典型生鲜电商企业，辅助其开发决策支持系统。本项目利用网络零售商的数据优势，构建了基于点击流数据驱动的生鲜需求预测新模型，在理论上丰富了预测领域的研究，所提出的预测方法也在风电预测领域得到了成功拓展；同时，本项目考虑生鲜的易变质特性，构建了生鲜产品补货-配送协同优化新模型，并提出了高效稳定的智能优化算法求解模型，在理论上丰富了生鲜库存和配送管理的研究，所提协同模型和算法也可扩展应用于生鲜企业的前置仓模式、生鲜社区团购模式等，在实践上为生鲜企业进行科学补货提供决策参考。