Data-driven segmentation of ship operational conditions using multivariate time-series clustering

Artikel i vetenskaplig tidskrift, 2026

Accurate identification of ship operational conditions is essential for extracting steady sailing periods from monitoring data and improving the reliability of ship performance assessments. The approach employs the Toeplitz Inverse Covariance-based Clustering (TICC) algorithm to segment operational states, including sea passage, maneuvering, anchoring/drifting, and at berth, from full-scale monitoring data. The method was validated using more than two years of measurements from a globally operating chemical tanker, including both coastal short sea and transatlantic open ocean voyages. The TICC-based method showed good agreement with the onboard reference labels for broad operational states, and further identified different steady operating settings within the sea passage label. The refined segmentation provides a useful basis for downstream applications such as energy efficiency analysis, performance modeling, and predictive maintenance.