Reinforcement Learning-Based Multimodal Transport Prediction: Collaborative Optimization in PU Data Scenarios

Container sea-rail intermodal transportation has broad prospects in multimodal transportation due to its advantages of low cost and low carbon emissions. However, in China, road transportation still dominates due to the late development of sea-rail intermodal transportation. To accurately identify goods suitable for railway transportation, this paper proposes a collaborative optimization framework that integrates reinforcement learning and deep learning. The framework employs Deep Deterministic Policy Gradient (DDPG) reinforcement learning to dynamically select reliable negative samples and combines a CNN+Transformer classification model for freight mode prediction. Experiments based on data from Nansha Port and Guangzhou Railway Group demonstrate that the proposed method outperforms nnPU and traditional approaches in metrics such as F1-score and accuracy. The model effectively reduces noise interference and improves the prediction accuracy of freight modes in multimodal transportation, providing new insights for optimizing railway and road freight scheduling.