Uncertainty propagation and sensitivity analysis of an artificial neural network used as wind turbine load surrogate model

Abstract

Recent studies have shown the advantage of replacing aeroelastic simulations with regression models based on Artificial Neural Networks (ANNs), which can be used as surrogate models for fast and efficient wind turbine load assessments. Once trained on a high-fidelity load simulation database covering a broad range of conditions, the surrogate model can be applied to predict loads for any site with wind climate falling within the range covered by the database. The aim of this study is to quantify the uncertainty propagation through such an ANN and to analyse how much the selected input variables influence the variance of the fatigue blade load estimations by means of a global sensitivity analysis. Results confirm that the selected ANN architecture seems suitable for this task resulting in small output uncertainties. Furthermore, the sensitivity analysis shows that the turbulence is mainly responsible for the blade load estimation, followed by the wind shear and the wind speed. The contributions of the turbulence length scale, turbulence anisotropy factor and wind veer angle are comparatively low. Comparing three different methods for sensitivity analysis shows that the partial derivative algorithm, Sobol variance decomposition and Shapley effect result in similar sensitivity measures.

Reply on RC1

“…Moreover, it has been shown that dimensionality reduction may improve the overall performance of neural network models, as non-informative variables can add uncertainty to the predictions and reduce the overall effectiveness of the model (Kuhn et al, 2013). ANN models' performance is thus highly dependant on the input data's quality, with certain inputs being vastly more relevant than others for the model's performance (Schröder et al, 2020).…”

Data-driven time series forecasting of offshore wind turbine loads

“…The ability of data-driven surrogate models to accurately emulate physics-based models while having a greatly reduced computational cost has seen them used in various ways in wind energy. This includes speeding up wind turbine modelling through reducing the expense of computational fluid dynamics for wake modelling [5,6] and characterizing hydrodynamic response [7], bypassing blade element momentum theory for aerodynamic load estimation [8], and to predict the load statistics on wind turbines [9,10,11]. For this reason, there has been significant work in producing SMs for wind farm design and planning, fatigue monitoring and design load assessment.…”

Surrogate models for the blade element momentum aerodynamic model using non-intrusive polynomial chaos expansions

“…For example, Dimitrov et al (2018), Schröder et al (2018), van den Bos et al (2018), andDimitrov (2019) used surrogate models to estimate the loads on a wind turbine based on the stochastic variables' gross parameters such as turbulence intensity, mean wind, or wind direction. Ashuri et al (2016), Murcia et al (2018), and Schröder et al (2020a) used surrogate models for uncertainty propagation through the wind turbine models. More recently, the surrogate models have been used for the wind turbines' reliability assessments (Slot et al, 2020;Schröder et al, 2020b).…”