The influence of flow-field representation on electrical power and fatigue-load surrogate modelling in offshore wind farms

Abstract. High-fidelity wind farm simulations are widely used to generate data for surrogate modelling of offshore wind turbine fatigue loads and electrical power output, often relying on high-dimensional spatial representations of rotor-plane wind fields. However, the effective dimensionality required for accurate prediction using temporally aggregated wind-field data remains unclear. This study investigates whether compact, physically motivated spatial descriptors can provide competitive predictive performance relative to full spatial wind-field representations when only temporally aggregated inputs are available. Mean wind speed and standard deviation fields aggregated over 10-minute intervals on a 30 × 30 rotor-plane grid are considered. A deterministic set of 31 spatial descriptors summarizing wake effects, shear, asymmetry, and spatial variability is constructed and used to train surrogate models based on a residual multilayer perceptron. Model performance is evaluated using group-based five-fold cross-validation. The results show that the compact descriptor set achieves predictive accuracy comparable to full spatial inputs for fatigue load components, indicating that low-order spatial representations are sufficient for capturing load-relevant flow characteristics in the aggregated data regime considered. In contrast, for electrical power prediction, models based on compact descriptors exhibit reduced accuracy relative to those using full or hybrid spatial representations, suggesting that energy-related quantities retain sensitivity to higher-dimensional spatial structure even after temporal aggregation. These findings highlight a task-dependent trade-off between input dimensionality and predictive performance and clarify when compact spatial representations are appropriate for surrogate modelling of wind turbine response.