Maritime transportation serves as the backbone of global trade, carrying more than 80% of the world’s cargo by volume. Ensuring shipping safety is a top priority for the maritime industry. To uphold safety standards, Port State Control (PSC) inspections, established by the International Maritime Organization (IMO), are conducted by national ports to verify that foreign visiting ships comply with international and local regulations and are adequately manned. Given the limited inspection resources at ports and the need to avoid excessive inspections that could disrupt the fast turnover of the maritime supply chain, accurately predicting a ship’s inspection in PSC, particularly the deficiency and detention conditions, is crucial for improving the reasonability of the ship inspection process. However, the existing models usually treat detention and deficiency prediction tasks separately, while advanced models such as deep learning are seldom developed for prediction. To address these limitations, we propose Dual-RSRAE, a novel multi-task Dual Robust Subspace Recovery Layer-based auto-encoder for ship risk prediction. This approach integrates the prediction of deficiencies and detentions within a unified, end-to-end pipeline, making it the first attempt to explore the inherent connections between these tasks. Our evaluation, conducted on 31,707 real PSC inspection records from the Asia-Pacific region, demonstrates that Dual-RSRAE outperforms state-of-the-art methods, achieving at least an 13% improvement in detention prediction and a 12% improvement in deficiency prediction accuracy.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号