Intelligent forecasting and distribution in cross-border e-commerce import trade: A deep-learning-based iterative optimization approach

Abstract

The dramatic growth of cross-border e-commerce (CBEC) trade promotes the vigorous development of bonded warehouses, providing overseas suppliers with an opportunity to lay out distribution networks to meet the domestic consumers’ growing logistics efficiency requirement. This paper considers the distribution network design problem with iteratively updated demand. Specifically, we first construct a hybrid deep learning model, which integrates a convolutional neural network for extracting recessive features and long short-term memory for a retrograde time extension to forecast the consumers’ demand. Then, a mixed integer linear programming (MILP) model is developed to formulate the distribution network design, which aims to rent the appropriate storage capacities of some warehouses in different locations and make the product allocation plans with minimum operation cost. The Benders decomposition algorithm is appropriately adopted as the solution approach to the proposed model. When the warehouse locations and distribution plan are initially developed, the logistics timeliness of some destinations will be improved, potentially leading to a redistribution of consumers’ demand. Therefore, we integrate the prediction and MILP models to construct a forecasting-distribution iterative optimization process to explore the optimal solution dynamically. A real case study is used to verify the effectiveness of the proposed integrated approach. Our research formulates characteristic distribution network design solution for overseas suppliers engaged in CBEC import trade, providing valuable insight to achieve an iterative optimization process through organically linking deep-learning-based forecasting and optimization processes.