Abstract
This paper presents a new system identification method for estimating low-frequency hydrodynamic loads from model test data, important for slowly-varying motions and mooring loads of floating structures. The difference-frequency quadratic transfer function (QTF) is represented by two-dimensional penalized B-splines, whose coefficients are determined using linear regression. This implies minimizing a cost function which embeds both fitting errors (discrepancy between measured and estimated loads) as well as penalty terms used to control the trade-off between flexibility and statistical variability. One penalty term also provides a rational means to blend the empirical QTF with a theoretical QTF from potential theory in regions of the bi-frequency domain with little support from data - far away from the peak of the wave spectrum. Well-established methods from linear regression theory are used to quantify model complexity, statistical variability and prediction performance. One advantage of regression-based system identification is its flexibility in terms of parametrization. This enables us here identifying an
-dependent QTF, including a
-proportional term representing “quadratified” cubic wave forces. The INO WINDMOOR 12 MW floating wind turbine is used as case study, and it is shown that the estimated
-dependent QTF provides good agreement between measured and estimated motions across a range of sea states.