Mooring tension estimation for multi-connected floating photovoltaic arrays via LSTM networks

초록

Accurately estimating mooring‑line tension is essential for the safe operation of large, multiconnected floating‑photovoltaic (FPV) arrays, yet installing load cells on every line is impractical. This study develops and evaluates data‑driven tension estimators that use only motion responses generated from time‑domain hydrodynamic simulations. A long short‑term memory (LSTM) network trained on displacements provides the reference performance. When trained instead on raw accelerations, the model performs noticeably worse, reflecting the spectral mismatch between acceleration and tension signals. Adding directional spreading to the training data restores robustness for the displacement‑based model under oblique seas, but offers limited benefit for the acceleration‑based model. In this study, a physics‑guided LSTM is proposed to reduce reliance on displacement sensors, in which a learnable filter transforms accelerations into displacement‑like features. This hybrid model narrows the performance gap, achieving stable and robust prediction performance. The proposed model attains accuracy comparable to displacement‑based estimation, demonstrating its effectiveness with accelerometer input alone and highlighting its potential as a cost‑efficient tool for structural health monitoring of large‑scale FPV systems. © 2025 Elsevier B.V., All rights reserved.

키워드