Efficient monitoring and prediction of the motion responses of compliant floating photovoltaic arrays: Case study of a hydrodynamic model test with erroneous sensor data

• Two-layer workflow restores monitoring integrity in floating solar arrays. • Cross correlation reassignment removes camera switching artifacts. • LSTM fills data gaps while preserving key statistics and extremes. • Benchmarked against spline, autoregressive, and persistence methods. • Stable accuracy when look back is five times the prediction horizon. Floating photovoltaic (FPV) arrays in irregular and harsh offshore seas face monitoring failures from simultaneous multi-float dropouts and optical switching between neighbouring floats. This study introduces a two-layer strategy that combines advanced preprocessing, low-pass filtering, Z-score outlier removal, and cross-correlation correction of switched signals with an optimized Long Short-Term Memory (LSTM) model capable of reconstructing both isolated and multiple simultaneous gaps. Unlike prior reconstruction studies on single-platform marine measurements, this work addresses compliant multibody FPV arrays where irregular waves create complex patterns of sensor faults. Validation on a scaled 216-float array under regular and storm-like irregular seas shows 95–100% fidelity for isolated gaps and superior performance to spline, autoregressive, and persistence baselines in multi-gap conditions. The framework preserves reliability-critical statistics such as upcrossing rates and extremes, reduces intensive simulation and testing time, and improves monitoring robustness for FPV systems under harsh marine environments.