Data-driven Surge Modelling of Floating Offshore Wind Turbines with Various Input Selection for Condition Monitoring

摘要

Floating offshore wind turbines (FOWTs) rely heavily on
mooring systems to maintain position and ensure safe operation under
complex environmental loading conditions. However, accurate modeling
and condition monitoring of mooring dynamics remain challenging due to
strong nonlinear behavior, coupled multi-degree-of-freedom interactions,
and the limited availability of high-quality experimental data. This study
proposes a data-driven modeling framework for condition monitoring of
FOWT mooring systems under limited data conditions. A time-delay
neural network (TDNN) is developed to capture the nonlinear dynamic
relationships between environmental excitations, platform motions, and
surge response which closely correlates with mooring forces. The model is
evaluated using the normalized root mean square error (NRMSE) under
wave-only, wind-only, and wave–wind coupled scenarios. Results indicate
that appropriate selection of input features significantly influences pre
diction accuracy. In wave-only cases, pitch motion provides the most rele
vant additional information for improving surge prediction. In wind-only
cases, the combination of heave and pitch yields the best performance,
as these motions contain useful low-frequency dynamics correlated with
surge. However, in wave–wind coupled cases, model performance deterio
rates due to increased system complexity and limited data length, which
reduces generalization capability. Overall, the proposed framework high
lights the importance of physically informed input selection and demon
strates its effectiveness for data-driven condition monitoring of FOWT
mooring systems using limited experimental measurements.