Experimental Evaluation of Data-Driven Deterministic Wave Generation

Kurzfassung

This paper explores the applicability of data-driven deterministic wave generation in an experimental environment. Various datasets that comprise different combinations of experimental- and synthetic data are used to train a fully convolutional neural network on the task of predicting the wave-maker control signal associated to a desired target wave sequence at a predefined target location in the wave tank. The trained models are evaluated in an experimental program, where the predicted control signals are executed in a wave tank, the actual wave is measured and compared against the predefined target wave. The standard wave modeling approach for the generation of tailored wave sequences, which is based on linear theory, is conducted for reference. The test dataset contains 950 irregular sea states with a wide range of characteristic sea state parameters, as well as seven tailored extreme waves, including the well-known "Draupner Wave". It is shown that the data-driven approach consistently outperforms the standard method in accuracy across the studied range of sea states, emphasizing its effectiveness as a flexible and precise alternative for wave basin experiments.