Digital twins for a floating photovoltaic system with experimental data mining and artificial intelligence modelling

Abstract

Floating photovoltaic (FPV) systems face complex multi-physics interactions from wave-induced hydrodynamics and solar variability, yet there is a lack of a digital framework to balance the large amount of data, modelling accuracy, and real-time adaptability. This study addresses this gap by developing an AI-driven digital twin framework that integrates physical experimentation, data integration, and neural network-based modelling. A novel FPV system was experimentally tested under 150 + scenarios in different solar irradiance and water wave conditions, capturing hydrodynamic, thermal, and power performances. A two-tier artificial neural network architecture was implemented, providing a high-fidelity model for detailed analysi and a reduced-order model for real-time applications. The virtual twin can predict key outputs, including heave, surge, pitch, mooring forces, PV temperature, and power output, which potentially reduces the need for sensors on the physical twin and provides more comprehensive information. In summary, the proposed digital twin framework enables remote monitoring, prediction, and intelligent management of FPV systems. Moreover, it holds the potential to support wave-adaptive panel control energy system management, and predictive maintenance. These functions position the digital twin as a core enabler for efficient, reliable, and scalable offshore solar energy deployment.