GPU-Accelerated Neural Networks and Computational Strategies to Predict Wave Heights

Significant Wave Height is an ocean wave characteristic that plays a major role in deriving and predicting wave energy. In this study, we aim to predict significant wave height using wind data as input to a multi-layer perceptron (MLP) neural network and analyze the developed network under several scenarios. Accordingly, the network was tested using different learning rates and numbers of layers. The computational times and results for all of the iterations and training periods were also recorded. Also, the acquired results of the models with better performances were compared with real data acquired from buoys. The results imply that all of the applied MLP networks could learn the relationships between wind and wave height and predict them. These MLP neural networks are composed of many operations that are either element-wise, or are represented as matrix-multiplications, making them good candidates for hardware (GPU) acceleration. This work illustrates the efficacy of wave height forecasting using MLP networks with GPU acceleration.