A Graph Neural Network Surrogate Model for the Prediction of Turbine Interaction Loss
James Bleeg
Abstract
Abstract The current generation of wind farm flow models lacks an option that can efficiently and reliably account for both wake and blockage effects when calculating turbine interaction loss. Traditional wake models are fast but ignore blockage effects. High-fidelity flow models are more complete, but turnaround times can be relatively long. The objective of this study is a model that combines the speed of traditional models with the accuracy of higher-fidelity approaches. To this end, we use a graph neural network (GNN) as a surrogate model for a steady-state Reynolds Averaged Navier-Stokes (RANS) model. Comparisons reveal good agreement between the GNN and RANS results for the atmospheric conditions considered.
Topics & Concepts
Reynolds-averaged Navier–Stokes equationsWakeSurrogate modelFidelityTurbineArtificial neural networkComputer scienceFlow (mathematics)GraphComputational fluid dynamicsMechanicsArtificial intelligenceMachine learningEngineeringAerospace engineeringPhysicsTheoretical computer scienceTelecommunicationsWind Energy Research and DevelopmentEnergy Load and Power ForecastingTurbomachinery Performance and Optimization