Wave runup prediction for a semi-submersible based on temporal convolutional neural network

Abstract

Nonlinear wave runup could result in serious wave impact on the local structures of offshore platforms in rough seas. The reliable and efficient wave runup prediction is beneficial to provide essential information for the design and operation of offshore platforms. This work aims to develop a novel data-driven method to achieve the nonlinear mapping underlying the wave-structure interactions. The Temporal Convolution Network (TCN) model was employed to predict the wave runup along the column of a semi-submersible in head seas. The incident wave and vertical motions including heave, roll, and pitch were fed into the TCN model to predict the wave runup. Experimental datasets were provided for training and test. Taking both temporal and spatial dependency into consideration, the input tensor space was optimized from the perspective of physical meaning and practicality. Sensitivity analyses were conducted to obtain the optimum length of time window and evaluate the relative importance of input variables to wave runup prediction. Moreover, the effects of characteristics and size of the training dataset on the model performance were investigated to provide guidelines for training dataset construction. Finally, upon validation, the generated TCN model showed a strong ability to provide stable and accurate wave runup results under various wave conditions, and it is a potential alternative tool to achieve efficient but low-cost wave runup prediction.