Search

This paper considers the one-step targeted maximum likelihood estimation methodology for multi-dimensional causal parameters in general survival and competing risk settings where event times take place on the positive real line and are subject to right censoring. We focus on effects of baseline treatment decisions possibly confounded by pretreatment covariates, but remark that our work generalizes to settings with time-varying treatment regimes and time-dependent confounding. We point out two overall contributions of our work. First, our methods can be used to obtain simultaneous inference for treatment effects on multiple absolute risks in competing risk settings. Second, our methods can be used to achieve inference for the full survival curve, or a full absolute risk curve, across time. The one-step targeted maximum likelihood procedure is based on a one-dimensional universal least favourable submodel for each cause-specific hazard that we implement in recursive steps along a corresponding nonuniversal multivariate least favourable submodel. Our empirical study demonstrates the practical use of the methods., his paper considers the one-step targeted maximum likelihood estimation methodology for multi-dimensional causal parameters in general survival and competing risk settings where event times take place on the positive real line and are subject to right censoring. We focus on effects of baseline treatment decisions possibly confounded by pretreatment covariates, but remark that our work generalizes to settings with time-varying treatment regimes and time-dependent confounding. We point out two overall contributions of our work. First, our methods can be used to obtain simultaneous inference for treatment effects on multiple absolute risks in competing risk settings. Second, our methods can be used to achieve inference for the full survival curve, or a full absolute risk curve, across time. The one-step targeted maximum likelihood procedure is based on a one-dimensional universal least favourable submodel for each cause-specific hazard that we implement in recursive steps along a corresponding nonuniversal multivariate least favourable submodel. Our empirical study demonstrates the practical use of the methods.

Wind farm flow models allow for the analysis of wind power plant wake effects. There are several mathematical models available to describe this complex flow, and some of them are implemented across multiple code bases. This manuscript presents a framework for verification and validation of these models of different complexities. As part of this work, an API was developed to interface the different flow tools using the WindIO wind plant schema. Then, verification techniques are applied to the embedded tools as a test to make sure there are no mistakes in the codes, and across different flow models. This reveals similarities and differences in stochastic trends given realistic measurement uncertainty. Validation is performed using a database of large eddy simulations to quantify model agreement with high fidelity data and develop a predictor of what this implies for future simulations.

Artificial Neural Networks (ANNs) are being applied as a faster alternative to Computational Fluid Dynamics (CFD) for wind turbine engineering wake models. Unfortunately, ANNs can fail to generalize if the data is insufficient. Physics-Informed Neural Networks (PINNs) can improve convergence while lowering the required data amounts. This paper investigates the PINN methodology systematically by considering varying amounts of data and physics collocation points. This work considers the rotationally symmetric Reynolds Averaged Navier-Stokes (RANS) formulation. Initially, a baseline fully data-driven ANN is studied to determine a suitable network size. Then, multiple PINN-based wake surrogates are trained with continuity and momentum conservation knowledge, varying amounts of data, and physics collocation points. It was found that including physics information under the best circumstances could improve accuracy by 18% at the cost of increasing the training time by a factor of 116. The findings imply that physics information can improve neural network based wake surrogates.

Simulating conventionally neutral boundary layers (CNBLs) with the unsteady Reynolds-Averaged Navier-Stokes (URANS) technique is investigated in this paper using a modified two-equation linear eddy viscosity turbulence model. For CNBLs over a flat and uniform surface, as typically used as the inflow to wind farm simulations, the governing equations of URANS can be solved with a one-dimensional solver, which makes the simulation of a typical CNBL five to six orders of magnitude faster than with large-eddy simulation (LES) approaches. However, URANS on the other hand requires more modelling than LES, and its accuracy is heavily dependent on the turbulence model employed. Through a cross-code study of a CNBL case with data from five different LES codes, it is found that the length-scale limiter of the employed turbulence model should be removed to correctly predict the atmospheric boundary layer (ABL) height evolution and the qualitative shape of various atmospheric profiles. A parametric study of simulations with varying initial ABL height further demonstrates the prediction capabilities of URANS, although a comparison with LES data shows that modelling of turbulence anisotropy and near-surface turbulence could be improved.

This work aims to alleviate the memory requirements of the recent wake engineering model described in Criado Risco et al. [1]. The original model relies on a RANS-based look-up table of three-dimensional velocity deficit and added turbulence intensity fields computed for a stand-alone turbine under a wide variety of conditions. The objective is to develop an alternative to the model of Criado Risco et al. [1], particularly in terms of added turbulence intensity, for which little research has been carried out to date. To achieve this, a one-dimensional analytical expression is fitted to the look-up table and generalized to higher dimensions. The turbulence intensity model is then coupled to a velocity deficit model and implemented in PyWake, an open-source wake engineering software. Overall, the new turbulence intensity model is found to provide a reliable description of the RANS look-up table data while reducing by half the memory requirements of the original model. This conclusion is extended to multiple wake situations, for which this work also establishes a direct link between the adequate superposition method and the definition chosen to describe the added turbulence intensity in the wake.

Energy losses due to wind farm clustering and wind farm interaction are rarely well represented in the wind farm design process because of the lack of fast models that can accurately account for neighboring wind farm wakes. A recently developed solution is the actuator wind farm (AWF) model, which is a Reynolds-averaged Navier-stokes (RANS) based wind farm parametrization that models a wind farm as a distributed thrust force and applies a global wind farm thrust coefficient controller. We propose an improved version of the AWF model, where each turbine employs a local thrust force controller and uses turbine thrust and power coefficients as input to better handle inhomogeneous inflow conditions. The proposed AWF model shows improved performance compared to the original AWF model in terms of predicted wind turbine power of a downstream wind farm that operates in a partial wake of an upstream wind farm, without significantly increasing the computational effort. However, the annual energy production (AEP) wake losses of a large wind farm cluster are nearly unaffected by using local or global control and input because the largest impact is found near the cut-in wind speed, which does not contribute much to the AEP wake losses.

Wind farm efficiency is influenced by atmospheric turbulence and wake interactions from preceding turbines. Optimal performance necessitates effective control strategies, encompassing collective/individual pitch (and/or torque) control, yaw control, and innovative techniques, significantly boosting energy capture. Wake effects are crucial, impacting downstream turbines and reducing overall energy extraction. For this purpose, the study of wind farm flow control (WFFC) holds significant relevance in this context. In this study, a Deep Neural Network predicts downstream flow features of a wind turbine under diverse control scenarios, including varying thrust coefficient and yaw control. A feed-forward neural network models the deficit and added turbulent intensity of a single wake, trained using Computational Fluid Dynamics (CFD) simulations as reference data. Linear superposition establishes the wind farm flow field, allowing examination of downstream effects. Another feedforward neural network encompasses wind farm physics such as blockage and wake recovery in and around wind turbine arrays which are overlooked by the single wake model. The methodology employed in this study yields results that are more time-efficient compared to traditional CFD models while maintaining a higher level of accuracy (ideally) than the engineering models, especially when implemented for WFFC without calibration.

Reynolds-averaged Navier-Stokes (RANS) simulations of a single wind turbine wake in the neutral atmospheric pressure-driven boundary layer (PDBL) are conducted and compared to RANS simulations with inflow based on the more traditional log-law. The latter is valid in the neutral atmospheric surface layer (ASL), while the PDBL is a better representation of the whole atmospheric boundary layer (ABL). It is found that the wake results of the two types of simulations become more similar for increasing ABL height to rotor diameter ratio. In fact, the ASL is shown to be a special asymptotic case of the PDBL. The RANS simulations are also compared to a large-eddy simulation (LES) PDBL case, where it is found that both the ASL and PDBL RANS simulations compare well with the reference LES data in the wake region, while the RANS PDBL compares better with the data in the upper region of the domain., QC 20230722

Rhenium disulfide, a member of the transition metal dichalcogenide family of semiconducting materials, is unique among 2D van der Waals materials due to its anisotropy and, albeit weak, interlayer interactions, confining excitons within single atomic layers and leading to monolayer-like excitonic properties even in bulk crystals. While recent work has established the existence of two stacking modes in bulk, AA and AB, the influence of the different interlayer coupling on the excitonic properties has been poorly explored. Here, we use polarization-dependent optical measurements to elucidate the nature of excitons in AA and AB-stacked rhenium disulfide to obtain insight into the effect of interlayer interactions. We combine polarization-dependent Raman with low-temperature photoluminescence and reflection spectroscopy to show that, while the similar polarization dependence of both stacking orders indicates similar excitonic alignments within the crystal planes, differences in peak width, position, and degree of anisotropy reveal a different degree of interlayer coupling. DFT calculations confirm the very similar band structure of the two stacking orders while revealing a change of the spin-split states at the top of the valence band to possibly underlie their different exciton binding energies. These results suggest that the excitonic properties are largely determined by in-plane interactions, however, strongly modified by the interlayer coupling. These modifications are stronger than those in other 2D semiconductors, making ReS2 an excellent platform for investigating stacking as a tuning parameter for 2D materials. Furthermore, the optical anisotropy makes this material an interesting candidate for polarization-sensitive applications such as photodetectors and polarimetry., ChemE/Opto-electronic Materials

The increase in the number of installed offshore wind farms has led to clustering, where wind farm interaction can cause energy losses. The Danish government is planning an Energy Island in the North Sea consisting of ten 1 GW wind farms. The initial wind farm layout design from a consultant company (COWI) has reported a 5% annual energy production (AEP) wake loss based on a low-fidelity engineering wake model that is known to underestimate wind farm interaction. The present work employs higher-fidelity wake models based on Reynolds-averaged Navier-Stokes (RANS) and a mesoscale model, which predict higher AEP losses due to wind farm interaction; namely, between 8.6-10.1%. In addition, we investigate how a microscale model as RANS can be employed to simulate a large wind farm cluster efficiently, and we compare its results with mesoscale model simulations.

Inside wind farms, wake effects are the primary source of turbine interactions, and as such, they constitute one of the most important aspects of wind farm operations. The two most widespread methods for calculating the wind farm wake flow are computational fluid dynamics (CFD) methods and engineering models. Both methods have drawbacks; CFD methods can be very accurate but are computationally expensive. Vice versa, engineering models sacrifice accuracy by simplifying the physics, thereby improving computational efficiency. In cases where many evaluations of the flow are needed, this trade-off is a hindrance. One such case is wind farm layout optimization problems. It has been shown that the estimation of wake flows can be improved by surrogate modeling. Recently, Artificial Neural Networks (ANN) have been demonstrated to predict accurate wakes over a mesh. In this work, a new mesh-free ANN-based wake model is proposed. This new model can predict the flow everywhere in the domain, and as it employs smooth activation functions, it is suitable for gradient-based optimization. Two ANNs were trained with data generated by Reynolds-Averaged Navier-Stokes with Actuator Disc simulations for several yaw angles. The first ANN predicts streamwise wake velocity induction/deficit, the second ANN predicts added wake turbulence intensity. Both ANNs predict with a low error.

A new wake surrogate model based on Reynolds-averaged Navier-Stokes (RANS) single rotor simulations is presented. The model relies on a series of three-dimensional pre-calculated deficit and added turbulence intensity flow fields, stored in a look-up table (LUT) as a function of the thrust coefficient and the ambient turbulence intensity. For any combination of these parameters, the flow around a wind turbine can be predicted by linearly interpolating within the look-up table. Furthermore, the resulting three-dimensional flow fields from different turbine sources can be superposed linearly to calculate the total wind farm flow. The model is implemented in PyWake and benchmarked against other, commonly employed engineering wake models, namely the Gaussian-Bastankhah, the N. O. Jensen and the Zong models, where RANS wind farm simulations are used as reference. In both full and partial wake cases, the surrogate model achieves a higher accuracy than any other model. Besides providing an accuracy comparable to a full RANS solution, the model can compute a flow case in the order of 1 s on a single processor. The main disadvantage is that the generation of the look-up table is time consuming, computationally expensive and can be memory demanding (especially if more inputs, such as the yaw misalignment angle, stability, etc. are added). Nevertheless, generating the LUT only has to be done once per wind turbine type.

Offshore wind farms are more commonly installed in wind farm clusters, where wind farm interaction can lead to energy losses; hence, there is a need for numerical models that can properly simulate wind farm interaction. This work proposes a Reynolds-averaged Navier-Stokes (RANS) method to efficiently simulate the effect of neighboring wind farms on wind farm power and annual energy production. First, a novel steady-state atmospheric inflow is proposed. This inflow model is well suited for RANS simulations of large wind farms because it does not lead to the development of nonphysical wind farm wakes. Second, a RANS-based wind farm parametrization is introduced, the actuator wind farm (AWF) model, which represents the wind farm as a forest canopy and allows to use of coarser grids compared to modeling all wind turbines as actuator disks (ADs). When the horizontal resolution of the RANS-AWF model is increased, the model results approach the results of the RANS-AD model. A double wind farm case is simulated with RANS to show that replacing an upstream wind farm with an AWF model only causes a deviation less than 1 % in terms of wind farm power of the downstream wind farm. Most importantly, a reduction in CPU hours of 74.4 % is achieved, provided that the AWF inputs are known, namely, wind farm thrust and power coefficients. The reduction in CPU hours is further reduced when all wind farms are represented by AWF models; namely 89.3 % and 99.9 %, for the double wind farm case and for a wind farm cluster case consisting of three wind farms, respectively. For the double wind farm case, the RANS models predict different wind speed flow fields compared to output from simulations performed with the mesoscale Weather Research and Forecasting model (WRF), but the models are in agreement with the inflow wind speed of the downstream wind farm. The double wind farm case is also simulated with the TurbOPark engineering wake model. Similar wake shapes are reproduced by TurbOPark but

Understanding wind turbine wake recovery is important for developing models of wind turbine interaction employed in the design of energy-efficient wind farm layouts. Wake recovery is often assumed or explained to be a shear-driven process; however, this is generally not accurate. In this work we show that wind turbine wakes recover mainly due to the divergence (lateral and vertical gradients) of Reynolds shear stresses, which transport momentum from the freestream towards the wake center. The wake recovery mechanisms are illustrated using a simple analytic model and results of large-eddy simulation.

Wind farm Annual Energy Production (AEP) calculations are required to design energy efficient wind farm layouts. We investigate methods that can reduce the computational effort of AEP calculations using Reynolds-averaged Navier-Stokes simulations of an idealized atmospheric wind farm setup. In addition, we introduce a generalized wind turbine model that compares well with wind turbine aerodynamic data covering a large range of wind turbine sizes. We apply the general wind turbine model to reduce the computational effort of the AEP calculations by decreasing the number of independent wind speed flow cases. Furthermore, we apply Reynolds-number similarity to compute the wind speed flow cases faster and we show how wind farm layout mirror- and rotational-symmetry can reduce the number of independent wind direction flow cases.

Numerical wind resource modelling across scales from the mesoscale to the turbine scale is of increasing interest due to the expansion of offshore wind energy. Offshore wind farm wakes can last several tens of kilometres downstream and thus affect the wind resources of a large area. So far, scale-specific models have been developed but it remains unclear how well the different model types can represent intra-farm wakes, farm-to-farm wakes as well as the wake recovery behind a farm. Thus, in the present analysis the simulation of a set of wind farm models of different complexity, fidelity, scale and computational costs are compared among each other and with SCADA data. In particular, two mesoscale wind farm parameterizations implemented in the mesoscale Weather Research and Forecasting model (WRF), the Explicit Wake Parameterization (EWP) and the Wind Farm Parameterization (FIT), two different high-resolution RANS simulations using PyWakeEllipSys equipped with an actuator disk model, and three rapid engineering wake models from the PyWake suite are selected. The models are applied to the Nysted and Rødsand II wind farms, which are located in the Fehmarn Belt in the Baltic Sea. Based on the performed simulations, we can conclude that both WRF + FIT (BIAS = 0.52 m s−1) and WRF + EWP (BIAS = 0.73 m s−1) compare well with wind farm affected mast measurements. Compared with the RANS simulations, baseline intra-farm variability, i.e. the wind speed deficit in between turbines, can be captured reasonably well with WRF + FIT using a resolution of 2 km, a typical resolution of mesoscale models for wind energy applications, while WRF + EWP underestimates wind speed deficits. However, both parameterizations can be used to estimate median wind resource reduction caused by an upstream farm. All considered engineering wake models from the PyWake suite simulate peak intra-farm wakes comparable to the high fidelity RANS simulations. However, they cons