Forecasting Vessel Power Using Variational Mode Decomposition and Convolutional Neural Networks

Vessel digitalisation can provide vital insights to support efficient deployment and optimisation of alternative/clean fuel propulsion systems to support marine vessel decarbonisation. In this context, this paper explores the effectiveness of a data-driven hybrid prognostic approach using Variational Mode Decomposition (VMD) and Convolutional Neural Networks (CNN) to forecast vessel thrust/propulsion power for a small to medium sized crewed vessel. The VMD method decomposes propulsion power data from a 27-meter Crew Transfer Vessel (CTV) into Intrinsic Mode Functions (IMFs) during typical maritime operational scenarios such as maneuvering, push on operations, transit, and loitering phases. The IMFs are reconstructed to create a refined signal representation that enables the effective training of the CNN model on a ratio of 70% training and 30% testing. The CNN model architecture is optimised for time series forecasting and is trained on the first 10 hours of data and validated on the subsequent 2-hour interval (10 to 12 hours) with a Root Mean square Error (RMSE) of 0.3 kW. The validated model is then applied for extended prognostic forecasting over the subsequent 12-hour horizon (12 to 24 hours). It is reported that the VMD/CNN approach accurately captures dynamic short-term patterns and provides insightful trend predictions, and concludes it is a promising approach for prognostic maritime energy management and operational planning applications.