Design And Implementation Of Amazon Logistics Truck Dispatching Prediction Subsystem

As the world’s leading e-commerce company, Amazon has a complex warehouse network in China. All shipments including transfer and customer order delivery are shipped by trucks which carriers offer. For different size, the costs are different, the higher the volume the greater the cost of the truck. Every day, staffs schedule the size of truck and number in advance according to experience. If the prediction of cargo volume is too large, the truck space utilization will be low and waste money in unused space. On the other hand, if the forecasted volume is too small, all shipments can’t be shipped out on time and customer satisfaction will be affected. In addition, the company is lack of truck volume rationality verification mechanism. In order to save transportation costs on the premise of meeting customers’ needs, developing a truck dispatching prediction subsystem is very necessary.Thesis topic is from real projects during author’s internship in Amazon China, the main target is to forecast daily volume of trucks, provide data for the reasonable arrangement of truck. This system is used for operational staffs primarily, which includes lane configuration module, historical data collection module, predictive parameters fitting module, real-time prediction module, report module, personal management module and back-stage management module. During the design and implementation process of this system, the author worked independently as follows:(1) Visited the Amazon fulfillment center to learn transportation processes. Through communication with operational personnel, acquainted and analyzed the demand of prediction system. Based on requirements, summarized and divided system modules and confirm with system’s users.(2) Finished the system’s design, selected techniques framework according to the specific situation of the project; based on the analysis of real data, selected unary linear regression model, tested prediction model and made appropriate use of the selected prediction model for project specific issues; through analysis of actual business requirement to abstract entities, attributes and entities relationship, completed the database design.(3) According to the final preliminary design, completed the detailed design, used generalized least squares to do linear fitting, and finished system testing independently. After the system coming into use, analyzed the effect and found the reason of poor predictive effect. Gave proposed improvement needs and suggestions to next version.The prediction system has been on-line now, according to analysis of real data, the system has provided reasonable arrangements of truck, that reduced waste truck space, improved truck loading rate, thus saved transportation costs.