Huo, Jialei (Wuhan University of Technology), Li, Chen (Wuhan University of Technology), Wang, Tengfei (State Key Laboratory of Maritime Technology and Safety，Wu), Cai, Yuxuan (Wuhan University of Technology), Li, Ruiyao (Wuhan University of Technology), Mei, Qiang (Jimei University)

Integrating Frequency Adaptive Normalization into Inland Ship Trajectory Prediction: A Non-Stationary Time Series Forecasting Approach

Scheduled for presentation during the Video Session "On-Demand Video Presentations" (VP-VP), Saturday, November 22, 2025, 08:00−18:00, On-Demand Platform

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on April 2, 2026

Abstract

Improving the accuracy of ship track prediction is indispensable for ensuring ship voyage security and the organization and scheduling of traffic flow. However, in the context of inland waterway traffic, ship AIS data exhibit strong non-stationary characteristics due to environmental factors such as the natural curvature of the channel and complex navigational flow patterns. Traditional forecasting models struggle to accurately capture the intricate spatiotemporal dependencies, leading to cumulative errors, particularly around turning points. In response to the aforementioned problems, we innovatively propose a time-series forecasting approach for non-stationary tasks of inland river ship trajectories based on frequency normalization adaptation. The stationary and non-stationary characteristics of the ship AIS information are dynamically decoupled through the Fourier Transform (DFT). We construct a hybrid network architecture combining Temporal Convolutional Network (TCN) and Transformer: to enhance the modeling ability of long-term spatial dependencies while ensuring the capture accuracy of local temporal features. The final output of this network is a trajectory sequence combining stationary prediction and non-stationary residuals. The results indicate that this approach is more effective than other classical models in terms of prediction performance, verifying the effectiveness of the non-stationary components decoupling strategy.