Multi-Step Significant Wave Height Forecasting Using an Artificial Intelligence Model Optimized by Dynamic Time Warping

Abstract Significant wave height is a critical parameter in the design of marine structures. Accurate and reliable multi-step forecasting of significant wave height is essential for ensuring the safety of offshore construction. In this paper, we propose a novel artificial intelligence model, which integrates a convolutional neural network (CNN), bidirectional gated recurrent unit (BiGRU), and dynamic time warping (DTW), to enhance multi-step significant wave height forecasting. Unlike traditional methods, the proposed model leverages DTW to analyze the dynamic similarity between target outputs and historical wave data, providing additional reference information to the forecasting process. Initially, DTW is employed to identify similar patterns in historical data that correspond to the target outputs. These patterns are then embedded into a similarity feature vector via a CNN block. Subsequently, this similarity feature is combined with the lagged sequence and used as input to the BiGRU for multi-step forecasting. The forecasting performance is evaluated using three measured datasets from the NOAA National Data Buoy Center, and the proposed model is compared against three benchmark models. The experimental results demonstrate that the proposed model delivers accurate and stable forecasting performance, surpassing all benchmark models in terms of accuracy and reliability.