Hybrid Inverted Transformer-CNN Model for Train Primary-Delay Recovery Time Prediction

The delay can easily propagate once the train is delayed, causing further delays in subsequent trains. Modeling the recovery process from train delays is essential in supporting decision-making and operational efficiency. Previous methods for predicting train delay recovery have focused mainly on temporal analysis, emphasizing operational time, delay duration, and stop time. However, models incorporating spatial information have been relatively underexplored. This paper proposes an Inverted Transformer Convolutional Neural Network (iTransformer-CNN) to predict train delay recovery time. The proposed model can more efficiently uncover the spatiotemporal relationships within the data, enhancing the accuracy of delay recovery predictions. The experiment results demonstrate that the iTransformer-CNN model can significantly improve the precision of train delay recovery predictions based on real-world data.