DeepDLP: Deep Reinforcement Learning based Framework for Dynamic Liner Trade Pricing
Xueyan Li, Yongyi Hu, Yumeng Bai, Xiaofeng Gao, Guihai Chen · 2023 17th International Conference on Ubiquitous Information Management and Communication (IMCOM) · 2023
Liner trade differs from other traffic flow in that it is characterized by substantially higher traffic volumes due to its low cost and high capacity. In the liner trade, pricing models should be distinguished by their ability to adapt to changing market conditions and provide a realistic picture of market trends. However, the prevalent manual pricing model used today in the industry can hardly adapt to the changing environment. We design a Deep Reinforcement Learning (RL) Framework for Liner Trade Problem (DeepDLP) to identify price strategy based on the characteristics of liner trade environment. We address four key problems with DeepDLP that significantly affect pricing. First, to address the issue of RL's challenging convergence due to current market variances, we use the Long-Short Time Memory (LSTM) to recognize environmental changes. Secondly, to avoid accumulated bias over time in LSTM, DeepDLP uses the price adjustment result of RL as the input of the Dynamic Price Prediction Module. Thirdly, in order to reduce noise and take past external information into account, we employ a historical information sensitive network in the Dynamic Price Prediction Module. Last but not least, we designed a balanced reward that allows the model to take into account both revenue maximization per unit of time and the sale of extra capacity. We first perform data wrangling and analysis on the actual data of a top liner trade company to identify the major routes and affecting elements in order to demonstrate the effectiveness and efficiency of DeepDLP. Next, we simulate real liner trade scenarios and conduct several experiments. The outcome shows that DeepDLP outperforms baselines and validates the significance of the designs.