Stochastic assessment of aerodynamics within offshore wind farms based on machine-learning. (December 2020)

• Record Type: Journal Article
• Authors: Richmond, M.; Sobey, A.; Pandit, R.; Kolios, A.
• Is Part Of: Renewable energy. Volume 161(2020)
• Page Start: 650
• Page End: 661
• Publication Date: 2020-12
• Subjects: Artificial intelligence -- Computational fluid dynamics -- Wind turbines -- Power prediction; Renewable energy sources -- Periodicals; Power resources -- Periodicals; Énergies renouvelables -- Périodiques; Ressources énergétiques -- Périodiques; 333.794
• DOI: 10.1016/j.renene.2020.07.083
• Languages: English
• ISSNs: 0960-1481

Abstract

Abstract: Wind turbine flow field prediction is difficult as it requires computationally expensive computational fluid dynamics (CFD) models. The contribution of this paper is to propose and develop a method for stochastic analysis of an offshore wind farm using CFD and a non-intrusive stochastic expansion. The approach is developed through testing a range of machine-learning methods, evaluating dataset requirements and comparing the accuracy against site measurement data. The approach used is detailed and the results are compared with real measurements obtained from the existing wind farm to quantify the accuracy of the predictions. An existing offshore wind farm is modelled using a steady-state CFD solver at several deterministic input ranges and an approximation model is trained on the CFD results. The approximation models compared are Artificial Neural Networks, Gaussian Process, Radial Basis Function, Random Forest and Support Vector Regression. RBF achieves a mean absolute error relative to the CFD model of only 0.54% and the error of the SVR predictions relative to the real data, with scatter, was 12%, compared to 16% from Jensen. This approach has the potential to be used in more complex situations where an existing analytical method is either insufficient or unable to make a good prediction. Highlights: A method for stochastic expansion of computational fluid dynamics (CFD) results is developed. Five machine learning approaches are compared in their approximation of the model results. Results show that the CFD results can be accurately replicated but are limited by the CFD model. This approach can be useful for wind farm site wind load evaluation.