ConvGRU-RMWP: A Regional Multi-Step Model for Wave Height Prediction

Accurate large-scale regional wave height prediction is important for the safety of ocean sailing. A multi-step model (ConvGRU-RMWP) based on ConvGRU designed two problems difficult spatial feature resolution and low accuracy in navigation prediction. For prediction, a multi-input multi-output strategy used, direction period are used as exogenous variables, which combined with historical data to expand sample space. features, convolutional gated recurrent neural network an Encoder-Forecaster structure extract resolve multi-level information. In contrast time series forecasting methods that consider only backward forward dependencies dimension single assessment properties predictor variables themselves, this paper additionally considers correlations implied among meteorological variables. This uses information past 24 h predict next 12 h. The results both space show can effectively temporal obtain good results. proposed method has lower error than other five verifies applicability three selected sea areas along global crude oil transportation route, all have error.