Method Research And System Implementation Of Couriers’ Pick-up Orders Prediction Based On Trajectory Data

In recent years,with the rapid development of e-commerce,logistics and express delivery have also flourished,and people’s demand for express delivery has also increased sharply.On the one hand,the development of express logistics has made online shopping very convenient,and the construction of a nationwide express network that extends in all directions allows people to quickly receive goods purchased online no matter where they are,so more and more people choose an online platform for consumption.On the other hand,the increasingly prosperous e-commerce has also promoted the development of the logistics industry,making the logistics industry more refined and intelligent,and bringing people a better express service experience.At present,millions of parcels are collected every day in our country,and the number of couriers reaches around 3.2 million.The courier receives a certain number of package collection appointment requests every day.The courier often determines the pick-up order of those tasks according to his own experience and preferences.Thus,the logistics platform cannot know the courier’s pick-up order.In fact,if the platform can predict the pick-up order of the courier in advance,it can fully consider the order of each courier when assigning new tasks,greatly optimizing the rationality of the assignment of tasks,thereby improving the overall efficiency and reducing overall cost.Under the huge base of the logistics industry,it can generate great economic value.Therefore,based on the courier’s historical pick-up trajectory data,this paper studies how to extract the temporal and spatial correlation characteristics of couriers’ pick-up data,and predict the pick-up order of couriers’ unfinished tasks.First,for the problem of parcel pick-up order prediction,a deep model based on graph convolution and Seq2 Seq framework is proposed,called the Graph Convolutional Pointer Net.The model designs a graph convolution module with residual error to model the spatial relationship between unpick-up packages.In addition,an encoding and decoding module is designed,using the Seq2 Seq framework with the attention mechanism to extract the time correlation and the global information of all the packages,and finally predict the pick-up order of the courier.Extensive experiments are performed on the proposed model on real datasets.The data set covers about 2 months of historical data of 2,344 couriers in a certain area of Shanghai.The effectiveness of the model proposed in this paper is verified by comparative experiments and ablation experiments with existing benchmark models.Finally,this paper designs and implements a system to predict the courier’s pick-up order in the production environment.The system is designed with the idea of microservices,which can support the storage and upload of the courier’s historical collection data and the collection of items in the production environment.The deployment and maintenance of the algorithm model support the courier to use the SDK to predict the picking sequence in real-time,and also supports the visualization of the historical package data and the forecast result of the picking order.The system can not only provide realtime service of picking up order prediction,but also be used for academic research on courier picking behavior.