PWPNet: A Deep Learning Framework for Real-Time Prediction of Significant Wave Height Distribution in a Port

In this paper, a 2-stage cascaded deep learning framework, Port Wave Prediction Network (PWPNet), is proposed for real-time prediction of significant wave height (SWH) distribution in port. The PWP-out model the first stage, predicting port-entrance parameters, utilizes three branches, branch using Long Short Term Memory (LSTM) module to learn temporal dependencies time sequences second and Wind field Feature Extraction (WWFE) modules, composed residual network with spatial channel attention, capture spatiotemporal characteristics outside-port 2D wind data, third multi-scale encoding periodic waves wind. PWP-in estimating in-port SWH distribution, uses parameters based on customized Artificial Neural (ANN) takes PWP-out’s output as its input. A comparison performance mainstream machine models including LSTM, GRU, BPNN, SVR, ELM, RF at Hambantota shows that outperforms all other regarding medium-term (25–48 h), med–long-term (49–72 long-term (73–96 h) predictions, ablation experiments proved effectiveness branches. Furthermore, our PWPNet those predictions