Resilient synchromodal transport through learning assisted hybrid simulation optimization model

Disruptions and uncertainties can significantly reduce the efficiency of conventional intermodal transport, often leading to severe economic losses and deterioration in service levels. To mitigate the negative impacts of disruptions on the shipments, our research leverages the flexibility of synchromodality and develops a learning-based modular framework for disruption management. By utilizing a hybrid simulation-optimization modeling approach, the framework effectively captures disruptions and generates dynamic response strategies. Through the integration of Reinforcement Learning (RL), the proposed approach re-plans under disruptions, accounting for their stochastic characteristics, enabling swift and effective decision-making in real-time scenarios. Results are compared against two policies, always wait and always reassign, highlighting the superior performance of the RL approach, when exposed to a certain disruption profile, with comparable or better decisions compared to other policies in response to disruptions. Additionally, results are compared against a benchmark policy to test an alternative reward mechanism, demonstrating that integrating a cost-based reward mechanism increases its resilience and results in lower costs, especially in the case of more frequent and low to moderately severe disruptions.