Dear all,
I am currently working on predicting platform motions from fairlead tension time-series data using machine learning. However, I am facing a challenge:
while motion → tension can be predicted accurately, the inverse problem (tension → motion) does not achieve the same level of accuracy.
Can someone provide insights on why this happens or how to address it?
Best regards,
Kajal
Dear @Kajal.Thakur,
I would guess your finding is related to a similar finding that quasi-static mooring models are typically accurate enough to capture floater motion, but are not accurate for calculating mooring loads. This is related to the fact that the quasi-static mooring models capture low-frequency effects well, but do not capture high-frequency effects, where mooring dynamics are important to resolve. Mooring loads are driven by both low- and high-frequencies whereas floater motions are typically low frequency. Of course, these results may be floater dependent.
Best regards,
Dear @Jason.Jonkman ,
Thank you for the clarification. In my case, when platform motion is used as the input, I am able to predict the resulting mooring tension accurately. However, the reverse mapping—predicting motion from tension—does not achieve the same level of accuracy. This may be due to the reasons you mentioned, particularly the influence of high-frequency mooring dynamics that are not uniquely recoverable from tension measurements alone.
Could you suggest any alternative approaches or methodologies that could help address this limitation in the inverse prediction problem?
Best regards,
Kajal
Dear @Kajal.Thakur,
I’m not sure I can suggest more without knowing more about what you have done and what results you are obtaining so far.
Best regards,