Tackling Uncertainty in Online Multimodal Transportation Planning Using Deep Reinforcement Learning

In this paper we tackle the container allocation problem in multimodal transportation planning under uncertainty in container arrival times, using Deep Reinforcement Learning. The proposed approach can take real-time decisions on allocating individual containers to a truck or to trains, while a transportation plan is being executed. We evaluated our method using data that reflect a realistic scenario, designed on the basis of a case study at a logistics company with three different uncertainty levels based on the probability of delays in container arrivals. The experiments show that Deep Reinforcement Learning methods outperform heuristics, a stochastic programming method, and methods that use periodic re-planning, in terms of total transportation costs at all levels of uncertainty, obtaining an average cost difference with the optimal solution within 0.37% and 0.63%.