Multilevel SVMD-AD-Prophet Model for Predicting the Track Irregularity of High-Speed Railways

Short-term prediction of track irregularity is essential for ensuring train operational safety and optimizing infrastructure maintenance costs. Existing prediction methods often overlook the impact of intermittent railway track maintenance, resulting in an inability to accurately forecast the evolving trends of track irregularities under maintenance influences. To address this limitation, this paper proposes a multilevel prediction approach that explicitly incorporates track maintenance into the characterization of track irregularity trends. The proposed method first utilizes the successive variational mode decomposition (SVMD) algorithm to decompose time-series track irregularity data into intrinsic mode functions (IMFs), isolating frequency components to improve short-term trend prediction. Subsequently, the development trend of each IMF is predicted using the Prophet model, augmented with an integrated time-series anomaly detection (AD) model to identify and account for maintenance dates. Furthermore, multiprediction error layers are designed to refine the model’s accuracy. The effectiveness and high precision of the proposed approach are validated through two real-world case studies based on field measurements of track irregularities. The results indicate that the proposed approach achieves minimal prediction deviation and outperforms several existing models in terms of accuracy. This approach offers valuable insights for early warning systems and timely maintenance strategies in railway track management.