Creation of an Offshore Wind Farm Power Output Surrogate Model for Design Optimization

Abstract

Offshore wind farms have many advantages over traditional energy systems and great potential for power generation. However, researchers are currently facing challenges in developing technologies for offshore wind farm analysis that are both computationally efficient and highly accurate. This study aims to develop a surrogate model for analyzing farms that is more efficient than existing computational models but maintains higher accuracy. It also aims to subsequently use this model to design offshore wind farms optimized for both power generation and cost of energy. Computational Fluid Dynamics (CFD) simulations were run for a design space with six features and one output and used to train a statistically robust ensemble model. This model was trained to determine relationships between the parameters and power output, tuned to achieve the greatest accuracy, and tested to ensure the legitimacy of predictions. Finally, using buoy data and Bayesian optimization, this model was used to design farms for various regions through the determination of a set of optimal structural parameters. A model with training and testing R2 values of 0.98 and 0.99 was successfully developed. It was found that, consistently among the four different locations tested with varying environmental conditions, there was an overarching optimal design: staggered orientation, 9-diameter spacing, 115 m-high hubs, and 175 m-rotor diameters. These structural parameters produced roughly 5.52−5.65 MW/m4 for each location. This consistency greatly simplifies wind-farm design going forward.