JSJens Peter Schøler

Jens Peter Schøler

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Jens Peter Schøler has a total of 11 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Convolutional versus graph-based surrogate models for inter-farm wake prediction using multi-fidelity transfer learning

    by Jens Peter Schøler, Jens Peter, Frederik Peder Weilmann Rasmussen, van der Laan, M. Paul, Alfredo Peña, Pierre-Elouan Réthoré

    Accurate prediction of wind farm wake interactions is important for energy yield assessment, as offshore wind farms increasingly operate in close proximity to one another. This work presents a systematic comparison of two neural network surrogate models for inter-farm wake deficit prediction: a convolutional neural network (CNN)-based attention residual U-Net (ARU-Net) and a graph neural network (GNN)-based graph neural operator (GNO). Both architectures are trained using multi-fidelity transfer learning, pre-training with low-fidelity engineering model simulations, and then fine-tuning with high-fidelity Reynolds-averaged Navier–Stokes actuator wind farm (RANS-AWF) data. The models are evaluated on procedurally generated wind farm layouts spanning diverse farm sizes, turbine spacings, wind speeds, and ambient turbulence intensities. Both architectures achieve high prediction accuracy but exhibit complementary strengths: evaluated over the wake region (δw≥10-3 m s −1 ) and averaged across both evaluation grids, the GNO achieves a lower root mean squared error (RMSE) (0.024 vs. 0.028 m s −1 ), while the ARU-Net attains a higher F 1 score (0.98 vs. 0.91), reflecting its superior wake boundary capture. Transfer learning substantially benefits the ARU-Net, while the GNO shows a smaller but measurable improvement, achieving its best performance with transfer learning.

  • Open Access

    Validation of RANS-calibrated engineering models and ANN-based surrogate for wind farm flow simulation and layout optimization

    by Jens Peter Schøler, Jens Peter, Ernestas Simutis, van der Laan, M. Paul, Julian Quick, Pierre-Elouan Réthoré

    Accurate yet efficient wake modeling is essential for wind farm layout optimization (WFLO). Wind turbine wakes are disturbed regions of flow behind a wind turbine, characterized by lower mean wind speeds and higher turbulence, which reduce downstream power production and increase structural loading. This study compares an artificial neural network (ANN)-based surrogate trained on Reynolds-averaged Navier–Stokes (RANS) data with two representative engineering wake models based on the TurbOPark and super-Gaussian formulations. The work includes recalibration of the engineering models, a systematic flow simulation study across varying turbine counts and spacings, and WFLO benchmarks validated against RANS-based annual energy production (AEP). Results show that the ANN surrogate achieves the lowest RMSE and MAPE across all scenarios in flow estimation, albeit at a higher computational cost. In WFLO, the TurbOPark-based model produced the highest RANS-validated AEP layouts, despite having lower predictive accuracy, suggesting that optimization complexity influences outcomes. Blockage modeling increased computational cost without improving accuracy.


  • Open Access

    Graph neural operator for wind farm wake flow

    by Jens Peter Schøler, Jens Peter, Frederik Peder Weilmann Rasmussen, Julian Quick, Pierre-Elouan Réthoré

    Wind farm flow simulations are computationally expensive. However, numerous simulations are often required to account for wake effects from multiple neighboring farms, which motivates the development of data-driven surrogate models. Such surrogates may also enable the consideration of neighboring farm wakes during layout optimization. Most existing data-driven surrogate approaches rely on fixed algebraic superposition principles, which may limit flexibility when extending to higher-fidelity data sources that capture nonlinear wake interactions. We propose a novel method that embeds a trainable and scalable superposition principle within a graph neural operator (GNO) architecture, enabling the model to learn complex wake combinations directly from data. The model consists of two sequential graph neural network (GNN) layers: the first encodes turbine–turbine interactions into a latent representation, while the second combines these latent turbine states to predict the wind speed at a desired location. The GNO is trained on a large dataset generated with PyWake, a steady-state engineering wake model. More than 2000 unique layouts were procedurally generated for training, with an additional 998 for testing. The GNO accurately identifies regions of strong wake interaction, although the spatial extent of wakes is slightly underestimated compared to the simulated values in cases with pronounced wake effects. Overall, the proposed GNO represents a methodological advancement in data-driven wind farm flow surrogates, introducing a new conceptual framework inspired by established engineering wake modeling principles.


  • Open Access

    Convolutional versus graph-based surrogate models for inter-farm wake prediction using multi-fidelity transfer learning

    by Jens Peter Schøler, Jens Peter, Frederik Peder Weilmann Rasmussen, van der Laan, M. Paul, Alfredo Peña, Pierre-Elouan Réthoré

    Accurate prediction of wind farm wake interactions is important for energy yield assessment, as offshore wind farms increasingly operate in close proximity to one another. This work presents a systematic comparison of two neural network surrogate models for inter-farm wake deficit prediction: a convolutional neural network (CNN) based attention residual U-Net (ARU-Net) and a graph neural network (GNN) based graph neural operator (GNO). Both architectures are rained using multi-fidelity transfer learning. First, pre-trained on low-fidelity engineering model simulations, and second fine-tuned on high-fidelity Reynolds-averaged Navier-Stokes actuator wind farm (RANS-AWF) data. The models are evaluated on procedurally generated wind farm layouts spanning diverse farm sizes, turbine spacings, wind speeds, and ambient turbulence intensities. Both architectures achieve high prediction accuracy but exhibit complementary strengths: evaluated over the wake region (δw ≥ 10−3 m s−1) and averaged across both evaluation grids, the GNO achieves a lower RMSE (0.024 vs. 0.028 m s−1), while the ARU-Net attains a higher F1 score (0.98 vs. 0.91), reflecting its superior wake boundary capture. Transfer learning substantially benefits the ARU-Net, while the GNO shows only marginal improvement.

  • Open Access

    Validation of RANS-calibrated engineering models and ANN-based surrogate for wind farm flow simulation and layout optimization

    by Jens Peter Schøler, Jens Peter, Ernestas Simutis, van der Laan, M. Paul, Julian Quick, Pierre-Elouan Réthoré

    Accurate yet efficient wake modeling is essential for wind farm layout optimization (WFLO). Wind turbine wakes are disturbed regions of flow behind a wind turbine, characterized by lower mean wind speeds and higher turbulence, which reduce downstream power production and increase structural loading. This study compares an Artificial Neural Network (ANN)-based surrogate trained on Reynolds Averaged Navier-Stokes (RANS) data with two representative engineering wake models based on the TurbOPark and Super-Gaussian formulations. The work includes recalibration of the engineering models, a systematic flow simulation study across varying turbine counts and spacings, and WFLO benchmarks validated against RANS-based Annual Energy Production (AEP). Results show that the ANN surrogate achieves the lowest RMSE and MAPE across all scenarios in flow estimation, albeit at a higher computational cost. In WFLO, the TurbOPark-based model produced the highest RANS-validated AEP layouts, despite having lower predictive accuracy, suggesting that optimization complexity influences outcomes. Blockage modeling increased computational cost without improving accuracy.

  • Open Access

    Graph Neural Operator for windfarm wake flow

    by Jens Schøler, Jens Peter, Frederik Peder Weilmann Rasmussen, Julian Quick, Pierre-Eloaun Réthoré

    Wind farm flow simulations are computationally expensive. However, numerous simulations are often required in applications such as wind farm layout optimization or when considering multiple neighboring farms, which motivates the development of data-driven surrogate models. However, most existing approaches rely on classical superposition principles, which constrain their ability to capture nonlinear wake interactions. We propose a novel method that embeds a trainable and scalable superposition principle within a Graph Neural Operator (GNO) architecture. The model consists of two sequential Graph Neural Network (GNN) layers: the first encodes turbine–turbine interactions into a latent representation, while the second combines these latent turbine states to predict the wind speed at a desired location. The GNO is trained on a large dataset of simulated wind farms and achieves a low prediction error, with an RMSE of 0.353 ms−1 and a MAPE of 0.938 % on an unseen test dataset. The GNO accurately identifies regions of strong wake interaction, although the spatial extent of wakes is slightly underestimated in cases with pronounced wake effects. Overall, the proposed GNO represents a significant advancement in data-driven wind farm flow surrogates, introducing a new conceptual framework inspired by established engineering wake modeling principles.

  • Open Access

    Speeding up large-wind-farm layout optimization using gradients, parallelization, and a heuristic algorithm for the initial layout

    by Rafael Valotta Rodrigues, Mads Mølgaard Pedersen, Jens Peter Schøler, Julian Quick, Pierre-Elouan Réthoré

    As the use of wind energy expands worldwide, the wind energy industry is considering building larger clusters of turbines. Existing computational methods to design and optimize the layout of wind farms are well suited for medium-sized plants; however, these approaches need to be improved to ensure efficient scaling to large wind farms. This work investigates strategies for covering this gap, focusing on gradient-based (GB) approaches. We investigated the main bottlenecks of the problem, including the computational time per iteration, multi-start for GB optimization, and the number of iterations to achieve convergence. The open-source tools PyWake and TOPFARM were used to carry out the numerical experiments. The results show algorithmic differentiation (AD) as an effective strategy for reducing the time per iteration. The speedup reached by AD scales linearly with the number of wind turbines, reaching 75 times for a wind farm with 500 wind turbines. However, memory requirements may make AD unfeasible on personal computers or for larger farms. Moreover, flow case parallelization was found to reduce the time per iteration, but the speedup remains roughly constant with the number of wind turbines. Therefore, top-level parallelization of each multi-start was found to be a more efficient approach for GB optimization. The handling of spacing constraints was found to dominate the iteration time for large wind farms. In this study, we ran the optimizations without spacing constraints and observed that all wind turbines were separated by at least 1.4 D. The number of iterations until convergence was found to scale linearly with the number of wind turbines by a factor of 2.3, but further investigation is necessary for generalizations. Furthermore, we have found that initializing the layouts using a heuristic approach called Smart-Start (SMAST) significantly reduced the number of multi-starts during GB optimization. Running only one optimization for a wind farm with 279 turbines initialized with SMAST resulted in a higher final annual energy production (AEP) than 5000 optimizations initialized with random layouts. Finally, estimates for the total time reduction were made assuming that the trends found in this work for the time per iteration, number of iterations, and number of multi-starts hold for larger wind farms. One optimization of a wind farm with 500 wind turbines combining SMAST, AD, and flow case parallelization and without spacing constraints takes 15.6 h, whereas 5000 optimizations with random initial layouts, finite differences, spacing constraints, and top-level parallelization are expected to take around 300 years.