Data-driven modeling of hydrodynamic loading using NARX and harmonic probing

Abstract

Optimizing floating wind turbines and their mooring systems requires validated computational models that predict wave-frequency and low-frequency hydrodynamic loads. Low-frequency loads are crucial for determining extreme offsets and tension in mooring lines and are generally described by a quadratic transfer function. The quadratic transfer function, obtained with numerical tools, accurately predicts low-frequency loads in mild sea states. However, since the existing numerical methods are based on potential and perturbation theory, they generally fail to accurately predict low-frequency loads in moderate-to-extreme sea states where current, viscous, and beyond-second-order potential effects become significant. Developing a procedure for empirical transfer function estimation is, therefore, necessary to overcome these limitations. This paper describes an existing framework for estimating any higher-order transfer functions from experimental data. The framework employs a nonlinear auto-regressive model based on Kriging to establish a causal input/output relationship between the wave-elevation and hydrodynamic force histories exerted on the floater. Then, higherorder transfer functions are extracted using harmonic probing. The procedure was validated by estimating the linear surge transfer function of the INO WINDMOOR 12 MW floater using synthetic data. The data-driven results showed an excellent agreement with the theoretically computed transfer function.

Data-driven modeling of hydrodynamic loading using NARX and harmonic probing