Wave Height Prediction in Maritime Transportation Using Decomposition Based Learning

Automation in the area of ship navigation and course planning is adversely affected by oceanic conditions. It leads to deviation from the set course, damages the ship structure, and reduces overall efficiency. Autopilot in ships can keep the ship on course but is unable to choose an efficient path in real-time. Varying wave height is one of the most prominent causes that lead to this inefficiency in the ship’s autopilot system. Current state-of-the-art methods in the domain include building machine and deep learning-based models to estimate wave height. However, the existing systems have several limitations, such as difficulty in handling abrupt non-linear and chaotic variations present in the data, low generalizability, noise sensitivity, and many more. To resolve this issue, the current study proposes a hybrid approach involving a combination of Variable Mode Decomposition and Bidirectional Long Short-Term Memory model (VMD - BiLSTM) and its integration to the ship’s autopilot system. The VMD-based data decomposition enables deep learning models to smoothly and accurately capture observed variational components present in the data, contributing to improved accuracy. Performance comparison with state-of-the-art prediction models validates the efficiency and reliability of the proposed prediction approach.