1. Introduction
Wind turbine wakes can result in significant power losses within a wind farm, as they reduce the power available to the downwind turbines, as well as enhance the fluctuating loads experienced by these turbines [
1]. Understanding and predicting these wakes is crucial, especially during the planning and layout optimization phase of a wind farm. Computationally inexpensive analytical tools are widely popular in the wind energy community for this purpose, as they offer fast and reasonably accurate estimations of wind turbine wakes, which enables testing different layout configurations and wind conditions in a relatively short time. Given their paramount importance, a number of analytical models for wind turbine wakes have been proposed over the years.
Most attempts to analytically model wind turbine wakes assume an underlying flat homogeneous terrain, which implies a zero pressure gradient situation. Early attempts at analytical modeling of wind turbine wakes started with Jensen [
2], who applied mass conservation downwind of the turbine and assumed a top-hat distribution of the velocity deficit. Later, Frandsen et al. [
3] used mass and momentum conservation around a wind turbine to estimate the velocity deficit in the wake. Similar to Jensen [
2], they also assumed a top-hat distribution of the velocity deficit across the rotor cross-section. Based on the empirical evidence of a self-similar Gaussian distribution of velocity deficit in the turbine wake, Bastankhah and Porté-Agel [
4] proposed an analytical model for the wake velocity deficit derived from streamwise mass and momentum conservation. Their model has since been adapted to different scenarios, such as wakes of turbines under yawed conditions [
5], or the ones experiencing wind veer effects [
6]. Recent advances in the analytical modeling of wakes include the so-called super-Gaussian model, which transitions from a top-hat profile in the turbine near wake to a Gaussian profile in the far wake [
7], a model for the wake velocity and added turbulence intensity based on a combination of analytical and numerical studies [
8], analytical models for yawed turbine wakes [
9] and for added streamwise turbulence intensity in the wake [
10].
It is highly likely that wind turbines sited in complex flow conditions, such as heterogeneous surface roughness conditions or topographies, experience a pressure gradient imposed by the base flow. This pressure gradient can significantly affect the evolution of the turbine wake, such as the recovery of the wake center velocity deficit and the expansion of the wake. Most conventional wake modeling approaches, however, assume zero pressure gradient and a homogeneous base flow velocity. A practical approach to model wakes in topography, for instance, is to superpose the velocity deficit in the flat terrain on top of the topography. Although simple, this approach has been shown to work only for terrains with very gentle slopes [
11,
12]. Brogna et al. [
13] proposed a modified form of the Gaussian model [
4] to be used for their wind farm optimization study in topography. More recently, Farrell et al. [
14] presented a wind farm wake model for varying base flow velocity field. They also based their wake model on the Gaussian model [
4], while keeping the reference base flow velocity spatially variable. Hu et al. [
15] presented a genetic algorithm based approach for siting wind turbines in complex terrain. To account for wake effects, they used an adapted Jensen model and Gaussian model based on the Brogna et al. [
13] formulation. These approaches, however, do not explicitly account for the imposed pressure gradient, as the underlying models are derived under the assumption of a flat terrain. Recent years have seen an increased interest in data-driven approaches to the estimation of wake effects in wind farms on flat terrain (see e.g., [
16,
17,
18]). On complex terrain, the problem complexity increases further due to the dependence of the flow characteristics (such as the imposed pressure gradients) on site-specific terrain characteristics. This hinders the applicability of data-driven modeling to complex terrains due to limitations related to available data for training purposes.
The only existing models that account for the effect of an imposed pressure gradient on wakes are the ones proposed by Shamsoddin and Porté-Agel [
19,
20] and in turn successfully applied by them to study the wake of a wind turbine sited upstream of a hill [
21]. These models solve an ordinary differential equation (ODE) for the streamwise evolution of the maximum velocity deficit under pressure gradient, which is derived by applying streamwise momentum conservation in a control volume. A self-similar Gaussian profile of the wake velocity deficit is assumed, which has recently been verified by Dar et al. [
22] and Dar and Porté-Agel [
23] in different topographies. The invariance of the ratio between the maximum velocity deficit and wake width to the pressure gradient is used to close the system of equations, and to obtain the wake width under the pressure gradient situation. In order to obtain a numerical solution, a boundary condition is required to solve the ordinary differential equation for the streamwise evolution of the maximum velocity deficit. In their original work [
19,
20,
21], the surrounding base flow imposes a zero pressure gradient at the turbine location and becomes non-zero from a certain location downstream of the turbine. Therefore, the maximum velocity deficit at the first streamwise position is assumed to be the same with or without the imposed pressure gradient. While true for the above-described scenario or for the situations where the imposed pressure gradient at the turbine location is small enough, the assumption may not be valid for situations where there is significant imposed pressure gradient at the turbine location. One such example is a wind turbine sited close to the edge of an escarpment, where the pressure gradient induced by the escarpment is high closer to the edge and vanishes as we move further away from it. Dar and Porté-Agel [
23] applied the model of Shamsoddin and Porté-Agel [
20] to predict the wake velocity deficit of a turbine sited close to the edge of an escarpment. They observed that the model worked well for the escarpments with a sloped or a smooth leading-edge, but its performance degraded with the increase in the sharpness of the escarpment’s leading-edge.
The objective of the current work is to develop an analytical modeling framework that can be applied in situations where the turbine experiences an arbitrary pressure gradient imposed by the base flow. The new model develops on the one proposed by [
20], where Bernoulli’s equation is used to estimate a theoretical near-wake velocity under a non-zero imposed pressure gradient. This near-wake velocity is then used to obtain maximum wake velocity deficit at the start of the turbine far wake, where an ordinary differential equation is solved. The model is validated against the experimental data and compared with the results from two existing models [
4,
20]. The rest of the article is structured as follows: the analytical modeling framework is detailed in
Section 2; validation of the model against experimental data and comparison with other models is performed in
Section 3; finally, a summary of the work and concluding remarks are given in
Section 4.
3. Model Validation
Following the derivation, we aim to validate the model with experimental data. For this purpose, we use the experimental data from Dar and Porté-Agel [
23]. In their experiments, a miniature wind turbine (WiRE-01) is placed one rotor diameter downstream of the edge of an escarpment, where the shape of the escarpment is varied between a forward-facing step with different edge curvatures and a ramp-shaped escarpment.
Figure 2 shows the geometrical details of the escarpments used in the experiments, and the normalized base flow velocity at the turbine hub height. As can be seen, the variation in the base flow velocity is high closer to the turbine (
, where
D is the rotor diameter), and reaches almost a constant value about five rotor diameters downstream of the turbine. Different escarpment shapes also show differences in their base flow velocities, which indicates a difference in the imposed pressure gradient. The chosen experiments are well-suited to test the new model, as the imposed pressure gradient is higher closer to the edge of the escarpment (i.e., at the turbine location) and differs between the escarpments, which enables us to test the model under different pressure gradients.
Table 1 presents a description of the escarpments.
In order to apply the pressure gradient model, we need two main inputs: the base flow velocity under the pressure gradient and the characteristics of the turbine wake under the zero pressure gradient (
and
). For the maximum velocity deficit under ZPG
, we use Equation (
4), which requires the turbine thrust coefficient
and wake width
. From experiments [
23], the thrust coefficient of 0.8 is used, which does not change between the flat and escarpment cases [
5,
23]. To obtain the ZPG wake width, we use the linear growth of wake width in the far wake region [
1]:
where
is the wake growth rate in ZPG, and
is the initial wake width. The wake growth rate
can be related to the streamwise turbulence intensity (
) in the flow, where several linear relations between the streamwise turbulence intensity and the wake growth have been proposed in the literature [
28,
29]. Here, we use the relation proposed by Brugger et al. [
29], which states
, as it fits the wake growth rate found experimentally for the miniature wind turbine in flat terrain by Bastankhah and Porté-Agel [
5]. As the pressure gradient model does not explicitly relate the turbulence intensity change in ZPG and PG conditions, we take the rotor-averaged turbulence intensity in the base flow at the turbine location to compute the wake growth rate for the ZPG wake. This is performed in order to account for the change in the turbulence intensity between the zero and non-zero pressure gradient situations. The theoretical normalized wake width
value of
is used at the end of the near wake [
5]. Following [
5,
30], the end of the near wake is assumed to be the position where the theoretical and experimental velocity deficit maximum on the escarpments become equal. The near wake length obtained by this criterion is very similar to the one obtained from theoretical relations derived for flat terrain [
5,
31].
In order to use Equation (
13), we need to define position 4 in
Figure 1. Mathematically speaking, this position should be chosen such that Equation (
13) yields a real value. A choice of position 4 where Equation (
13) results in an imaginary number would indicate a breakdown of the theory, which could be similar to the situation of actuator discs with thrust coefficients above 1 in the classical one-dimensional momentum theory [
32]. Following [
26,
27], from a physical perspective, position 4 should correspond to a location where the pressure in the wake flow becomes equal to that in the base flow, and there is no mixing between the (outer) base and wake flow.
Figure 3 shows the contours of the normalized turbulence kinetic energy in the turbine wake for different escarpment cases. Behind the turbine top tip level, a region of high turbulence kinetic energy can be observed, which is relatively thin closer to the turbine but starts to expand in the vertical direction from a certain position downstream, corresponding to the position where tip vortices start to breakdown and the outer flow starts to mix with the wake flow. Therefore, position 4 should be chosen before the region of high turbulence kinetic energy starts to expand in the vertical direction. However, it should not be picked too close to the turbine to avoid influence of the pressure drop across the rotor.
A common approach in the literature [
3,
33,
34] is to assume one rotor diameter downstream of the turbine as the distance where pressure in the wake and base flow equalizes. This position also lies within the region where the turbulence kinetic energy does not start to grow for all the cases. Therefore, we choose one rotor diameter downstream of the turbine as a common assumption for position 4 in all cases. It is to be noted that the choice of position 4 used here might not be universal, and future work should investigate this. The above discussion comes from the one-dimensional momentum theory for actuator discs, and in reality, the structure of the turbine near wake is much more complex. As shown by [
5], the measured near-wake velocity deficit for the miniature turbine is higher than the theoretical one and varies instead of being a constant. This difference is attributed to several factors, including the wake of the nacelle and rotation of the wake. Although a simplified approximation, the theoretical near wake velocity provides useful information on the wake flow, such as the end of the near wake and a theoretical estimation for the velocity at the start of the far wake [
5,
14,
30].
Once all the required inputs for the pressure gradient model have been obtained, we compute the maximum velocity deficit under pressure gradient using Equation (
7) with the new boundary condition given by Equation (
14) and wake width using Equation (
6). In addition to the new model, we also test the pressure gradient model by Shamsoddin and Porté-Agel [
20], and the Gaussian model by Bastankhah and Porté-Agel [
4].
A comparison of the maximum velocity deficit normalized by the hub height velocity between the experiments and the analytical models is shown in
Figure 4 (left panels). The new pressure gradient model is represented by ‘PG-New’, whereas the pressure gradient model by Shamsoddin and Porté-Agel [
20] is named ‘PG-SPA’, and the zero pressure gradient model (Gaussian model) by Bastankhah and Porté-Agel [
4] is named ‘ZPG’. The imposed pressure gradient depends on two factors: the shape of the escarpment, as a sharper edge would induce a higher pressure gradient, and the distance from the escarpment leading edge, as the pressure gradient would reduce with the increase in the distance from the escarpment edge. As a result, the differences between the different models compared here are also dependent on the same two factors. In general, the new pressure gradient model predicts the maximum velocity deficit reasonably well for all escarpments, as it accounts for the imposed pressure gradient at the turbine location. The PG-SPA model performs well for the ramp-shaped escarpment, as the imposed pressured gradient at the turbine location is lowest in this case. For the forward facing step cases, however, its performance degrades with the increase in the sharpness of the escarpment edge, where it works for the FFS-III case at distances greater than five rotor diameters, but underestimates the maximum velocity deficit for the other two FFS escarpments. This is due to the fact that the imposed pressure gradient is higher at the turbine location than in the far wake, and the PG-SPA model does not account for it, thereby underestimating the maximum velocity deficit. The zero pressure gradient model also underestimates the maximum velocity deficit for almost all the cases as it cannot account for the contribution of the pressure gradient to the velocity deficit.
The escarpments impose an adverse pressure gradient on the flow, which is known to slow down the recovery of the turbine wake compared to that under the zero pressure gradient [
20,
24]. This explains why the models that do not account for the imposed pressure gradient at the turbine location underestimate the maximum velocity deficit. It can also be noted that for the two forward-facing step escarpments with relatively sharper edges (FFS-I and FFS-II), the PG-SPA and ZPG models show very similar values of the maximum velocity deficit. This is due to the fact that in the mentioned cases, the base flow velocity at the start of the far wake is almost the same with and without the escarpment. In other words, these escarpments not only induce the highest pressure gradient closer to the escarpment edge, but they also show the fastest decay in the induced pressure gradient with downstream distance. Therefore, at around four rotor diameters downstream of the turbine (five rotor diameters from the escarpment edge), the pressure gradient induced by the escarpments in the FFS-I and FFS-II cases is almost zero; as the PG-SPA model does not account for the imposed pressure gradient at the turbine location, it yields values similar to the ZPG model.
Following the maximum velocity deficit, the equivalent wake width obtained from the analytical models is compared with the experimentally obtained one in
Figure 4 (right panels). The ZPG wake width is smaller than the experimental equivalent wake width. This is to be expected, as an adverse pressure gradient results in a larger wake width compared to the zero pressure gradient one [
20,
24]. The PG-SPA underestimates the wake width for the FFS-I and FFS-II cases, but works well for the rest of the cases. The wake width obtained from the new pressure gradient model is observed to agree well with the experimental data for all the cases.
A comparison of the normalized velocity deficit profiles between the analytical models and experiments is shown in
Figure 5. The velocity deficit profiles obtained from the new pressure gradient model are observed to agree well with the experimentally obtained profiles for all escarpment cases. As shown by Dar and Porté-Agel [
23], the wake width in the lateral and vertical directions can vary depending on the escarpment shape. However, as mentioned earlier, in the current modeling approach, we solved the problem for an equivalent wake width (
). Comparing the experimental and (new) analytical velocity deficit profiles in the lateral and vertical direction shows that this approach works well. The PG-SPA and ZPG models, on the other hand, yield underestimated velocity deficit profiles for the most part. The PG-SPA model underestimates the velocity deficit profiles for the FFS-I and FFS-II cases, whereas it shows reasonable agreement for FFS-III case for downstream distances greater than five rotor diameters. For the ramp-shaped escarpment, it shows good agreement for all downstream distances. The ZPG model gives reasonable results at a downstream distance greater than five rotor diameters in the case of the ramp-shaped escarpment, which can be related to the fact that the effect of the pressure gradient is lowest for the ramp-shaped escarpment at high downstream distances. In general, we can say that the new pressure gradient model can successfully predict the velocity deficit in the turbine wake for all escarpment cases and outperforms the other two models tested in the study.