The analysis of multivariate time series for the identification of ship operational conditions

Paper in proceeding, 2025

Accurate identification of a ship’s operational conditions is crucial for extracting steady-state sailing scenarios from monitoring data, thereby improving the quality of ship performance assessments. Traditional methods often rely on manual thresholds and complex signal filtering, which can be labor-intensive and error prone. This study presents a data-driven approach for automatically classifying key operational states, including sea passage, anchoring/drifting, berthing, and maneuvering, based on multivariate monitoring data. The proposed approach employs the state-of-the-art Toeplitz Inverse Covariance-based Clustering (TICC) algorithm, an advanced multivariate time series clustering technique that provides a simple, fast, and efficient way of identifying these conditions. It analyzes multivariate time series data, such as propulsion power, engine RPM, ship speed, and ship heading, to automatically distinguish between these scenarios. We demonstrated the method based on the full-scale measurements of a worldwide sailing chemical tanker. The results show that the proposed method can accurately extract typical operational conditions, as confirmed by a comparison with the actual measurement labels. This approach not only simplifies the identification process and improves its reliability, but it is also particularly valuable for ships that do not have labeled operational states, offering strong potential for practical applications in ship performance monitoring.