Monopile strain prediction from wind turbine structural responses using optimised ANFIS models

Monopile-supported wind turbines serve as a key technology for harnessing wind energy. However, the response of embedded monopile foundations is difficult to monitor directly after installation, whereas the dynamic responses of the wind turbine superstructure can be measured more easily during operation. This study investigates the feasibility of applying machine learning techniques to predict monopile strain based on measurable superstructure responses. To this end, an optimisation framework is developed, in which metaheuristic algorithms are used to optimise the parameters of the Adaptive Neuro-Fuzzy Inference System (ANFIS), a well-established neuro-fuzzy modelling approach. An upgraded Energy Valley Optimiser, termed UEVO, is further proposed as the main optimiser, in which the Brownian random movements used in the standard EVO search process are replaced by Lévy flights to improve convergence behaviour. A supervised prediction scheme is then established from synchronised acceleration and strain measurements obtained in the scaled wind turbine model test, where acceleration responses are used as inputs and monopile strain responses as targets. The results show that UEVO improves the accuracy and robustness of monopile strain prediction and outperforms the benchmark algorithms. These findings highlight the potential of using measurable superstructure responses for indirect monitoring of embedded monopile foundations in wind turbine systems.