Towards robust and dynamic ship domains: A data-adaptive model based on Conditional Neural Network Regression

Journal article, 2026

Establishing a scientific ship domain is paramount for maritime safety and traffic efficiency, particularly in inland waterways with frequent vessel encounters and limited navigable space. However, this task is hindered by the stochastic complexity of waterway environments and the intricate non-linear kinematics of maritime traffic. To overcome these challenges, a novel Conditional Neural Network Regression (CNNR)-based ship domain model is proposed for inland waterways. The framework bridges AIS-derived empirical data with knowledge-based insights, enabling dynamic generation of ship domain boundaries that adapt to environmental conditions with both statistical precision and situational robustness. A case study using 2023 vessel traffic data from the Yangtze River estuary validated the model’s performance under varying sample sizes against traditional empirical models. Results show that the discrepancy between the empirical and proposed models increases substantially as data volume decreases, ranging from less than 5% in data-rich regimes to over 150% in intermediate and data-sparse regimes. While the proposed CNNR model consistently generates well-defined elliptical domains, empirical methods fail to form closed and regular boundaries under limited data conditions. Comparative analysis confirms the superior accuracy and robustness of the proposed model for intelligent navigation and maritime traffic management in complex inland waterways.