Prediction of TLP tensioner dynamics under second-order waves via wavelet-enhanced deep learning

The semi-rigid and semi-compliant configuration of a Tension Leg Platform (TLP) avoids resonance at common ocean-wave frequencies. However, second-order wave actions can induce high-frequency resonance and slow-drift phenomena, degrading platform motion and tensioner fatigue life. Accurate prediction of TLP dynamic responses is therefore essential for safe deep-sea operations. This paper proposes the Floating Platform Motion Prediction Network (FPM-Net), a hybrid deep learning model combining wavelet transform and spectral enhancement, to predict TLP tensioner-system dynamics under four typical wave forces. FPM-Net integrates Wavelet Transform Convolution (WTC), an Interactive Adaptive Spectral Block (IASB), and enhanced Long Short-Term Memory (LSTM) for multiscale feature extraction and frequency-domain enhancement of non-stationary signals. Compared with benchmark models, FPM-Net reduces displacement-prediction RMSE and MAE by averages of 72.46% and 77.41%, respectively, achieving 0.19 and 0.14. For pressure prediction, average RMSE and MAE reductions are 34.53% and 27.67%, demonstrating superior accuracy and robustness for deep-sea platform operation.