Multi-fidelity surrogate modeling of nonlinear dynamic responses in wave energy farms

Abstract

In wave energy farms, accurately determining the motion of each wave energy converter is essential for performance evaluation, estimating energy production, and implementing effective control strategies. The primary challenge lies in the real sea environment, where the complex nonlinear hydrodynamic phenomena make it difficult to estimate the motion of each converter precisely. High-fidelity numerical simulations, such as computational fluid dynamics, offer a detailed representation of the wave farm’s response to incoming waves. However, they are computationally intensive, making them impractical for real-time implementation and scenario evaluation. Conversely, although widely used in the industry, low-fidelity models based on linear potential flow theory lack accuracy and provide only a general solution trend. Experimental wave tank tests, while offering realistic, high-fidelity system representations, face limitations due to flexibility and costs. A multi-fidelity surrogate modeling approach presents a viable solution for designing and controlling wave energy farms. By leveraging data from various fidelities, low-fidelity numerical simulations, and high-fidelity experimental measurements, we develop a model capable of predicting the actual heave motion of each converter within a farm under diverse irregular wave conditions. This model effectively corrects the low-fidelity motion to align with each converter’s real heave response. Central to our model is the long-short-term memory machine learning method, which enables the prediction of the devices’ temporal response to incoming irregular waves. This model delivers solutions with low computational cost, making it suitable for estimating the actual device response during the design stage of a wave energy farm, facilitating real-time monitoring. • A multi-fidelity surrogate model for motion prediction within wave energy farms. • A long-short-term memory network is trained on low- and high-fidelity data. • The motion of the wave energy converters is refined to align with experimental data. • The network estimates nonlinear dynamics, keeping high computational efficiency. • Real-time monitoring of nonlinear responses of multiple wave energy converters.

Topics & Concepts