Reinforcement learning enabled dynamic bidding strategy for instant delivery trading. (October 2021)

• Record Type: Journal Article
• Authors: Guo, Chaojie; Thompson, Russell G.; Foliente, Greg; Peng, Xiaoshuai
• Is Part Of: Computers & industrial engineering. Volume 160(2021)
• Publication Date: 2021-10
• Subjects: Instant delivery -- Sequential auctions -- Reinforcement Learning -- Bidding strategies; Engineering -- Data processing -- Periodicals; Industrial engineering -- Periodicals; 620.00285
• DOI: 10.1016/j.cie.2021.107596
• Languages: English
• ISSNs: 0360-8352

Abstract

Highlights: Reinforced learning methods improve carrier's profits for instant delivery trading. Second price auctions are more stable and perform better than first price auctions. Deep Q networks were found to be the superior method in first price auctions. Abstract: Due to the great potential to enable collaboration and improve consolidation, auctions have been identified as a possible effective option to improve the efficiency of instant delivery. Instant delivery markets are complex and dynamic systems influenced by highly random demand. Conventional bidding strategies require perfect market information and cannot be adjusted effectively according to the evolution of requests. To address this problem, this paper proposes an auction-based trading platform to enable freight transportation procurement and develops a Reinforcement Learning (RL) enabled dynamic bidding strategy to optimize carrier's behavior in sequential auctions. In the RL enabled dynamic bidding strategy, three RL algorithms, including Q-learning, Deep Q Network and experience replay based Q-learning are used to improve carrier's bidding ability. The simulation results demonstrate that compared with the conventional bidding strategy, the RL enabled dynamic bidding strategies with any of the three RL algorithms can help carrier secure more auctions and gain more profit in a competitive marketplace. In addition, the advantages of the RL enabled dynamic bidding strategies are more obvious and the performance is more stable in more uncertain market environments.