Numerical Weather Prediction and Artificial Neural Network Coupling for Wind Energy Forecast

In the past two decades, wind energy has been under fast development worldwide. The dramatic increase of wind power penetration in electricity production has posed a big challenge to grid integration due to the high uncertainty of wind power. Accurate real-time forecasts of wind farm power outputs can help to mitigate the problem. Among the various techniques developed for wind power forecasting, the hybridization of numerical weather prediction (NWP) and machine learning (ML) techniques such as artificial neural networks (ANNs) are attracting many researchers world-wide nowadays, because it has the potential to yield more accurate forecasts. In this paper, two hybrid NWP and ANN models for wind power forecasting over a highly complex terrain are proposed. The developed models have a fine temporal resolution and a sufficiently large prediction horizon (>6 h ahead). Model 1 directly forecasts the energy production of each wind turbine. Model 2 forecasts first the wind speed, then converts it to the power using a fitted power curve. Effects of various modeling options (selection of inputs, network structures, etc.) on the model performance are investigated. Performances of different models are evaluated based on four normalized error measures. Statistical results of model predictions are presented with discussions. Python was utilized for task automation and machine learning. The end result is a fully working library for wind power predictions and a set of tools for running the models in forecast mode. It is shown that the proposed models are able to yield accurate wind farm power forecasts at a site with high terrain and flow complexities. Especially, for Model 2, the normalized Mean Absolute Error and Root Mean Squared Error are obtained as 8.76% and 13.03%, respectively, lower than the errors reported by other models in the same category.

The complex winds of the Alps: an unseen asset for the energy transition

Jérôme François Sylvain Dujardin

Wind energy has played a major role in the development of our society since 5000 BC when the first sail boats traveled the Nile River. Wind technology continued to advance with the invention of wind-powered pumps and mills in China and the Middle East as early as 200 BC, and by the late 19th century the first electricity-generating wind turbines were built. Today the effects of human driven climate change are pushing the world to transition from a fossil fuel dominated energy diet to a menu composed of a variety of technologies powered by renewable energies, including wind power. Our consumption habits and the historical design of our electricity grids were based on relatively few, centralized, and controllable power plants that use various fossil or nuclear fuels to generate power on demand to match real time use. The weather driven nature and low energy density of renewable sources like solar, wind and hydro lead to a more distributed and fluctuating electricity generation, which results in multiple problems: In Switzerland, under the foreseen future where photovoltaic (PV) panels are predominantly installed in urban areas, the nuclear phase-out will cause large deficits of electricity in winter. This thesis contributes to the solution of these important challenges in two ways: first by presenting a model that combines a variety of renewable energy sources to form a reliable and efficient electrical system, and secondly by improving the assessment of wind power potential in the Alps. The first two parts of this thesis are concerned with the simulation and optimization of the Swiss electrical system, in which the nuclear reactors are replaced by a combination of solar and wind energy sources. A spatially explicit approach is required to consider the spatio-temporal variability in renewable sources. Storage hydropower is a remarkable asset for the energy transition, and thus occupies a central role in the manner the system is modeled and operated (Part I). A novel algorithm explores how the whole system reacts under various combinations of PV and wind energies (Part II). By simultaneously optimizing the generation mix and location of PV panels and wind turbines, while considering the electrical grid and land use, this algorithm showed that a large number of carefully located wind turbines and a small number of PV panels located in the Alps can provide a much more balanced system than the conventional, urban PV scenario. Part III of this thesis addresses the difficulty of wind power potential assessment in mountainous terrain. On such terrain, high horizontal resolution is required in the numerical models that simulate the air flows. This constraint, and the need for long time series, prohibit the simulation of large domains and thus the identification of the best locations for wind turbines in the entire Swiss Alps. Downscaling models for wind are often used to obtain high resolution and long time series, but perform poorly on such terrain. In order to improve these models, a new method based on deep learning is presented. It downscales the outputs of the operational numerical model COSMO-1 used for the Swiss weather forecast, which has a resolution of 1.1 km, to a 50-meter resolution. The validation at 60 measurement stations shows that wind speeds and directions are more accurate than COSMO-1 and present many of the expected orographic effects like deflection, sheltering and ridge acceleration.

EPFL2021

Simulation of atmospheric flows over complex terrain for wind power potential assessment

When assessing the economic viability of a wind farm, the estimation of the on-site wind power potential is perhaps the most important step. The most common way of evaluating the wind power potential of an area of interest consists of making on-site measurements for a period of one year. In order to take account of the inter-annual variation of wind speed, the one year of data are normally correlated with data recorded at a reference site where long-term data (typically > 10 years) are available. A correlation analysis is formulated for the concurrent data sets at the reference and prediction sites. This correlation is then used to transform the long-term wind speed at the reference site to the long-term wind speed that would have been expected at the prediction site had long-term measurements been made at this site. An alternative approach is also used, which consists of establishing site-to-site relationships using a numerical model to simulate meteorological situations which are typical for the area of interest. These relationships are then used to transpose the known long-term wind statistics of the reference site to the prediction site. Such an approach is applied in this work to the region of Chasseral & Mt-Crosin. The wind data available for a period of 16 years at Chasseral are transposed to the Mt-Crosin site where they are then compared to the data measured at the location of the installed wind farm. Over complex terrain, the linearised models traditionally used for wind power potential assessment fail to reproduce accurate wind fields. Therefore, to be applied to mountainous terrain such as that found in Switzerland, the approach relying on numerical simulation requires the development and validation of a numerical tool capable of simulating wind fields over complex topography. As the numerical model would have to deal with relatively steep slopes requiring a fine horizontal (50-100 m) and vertical resolution (∼5-10 m in the lowest levels), a fluid dynamics model was used which solves the complete set of Navier-Stokes equations with κ-ε turbulence closure. The standard version of the model used (CFX4) is modified in a novel way to extend its field of application so that atmospheric phenomena could be simulated which are typical of the meso-scale. The modified version solves the flow equations with the anelastic approximation (deep Boussinesq) and assuming a background rotation of the wind field (with the high altitude wind field following the geostrophic approximation). In the first part of this work, the numerical model is validated. The results obtained in this phase show that for meteorological situations for which the wind at the ground is coupled to the high altitude wind, the numerical model is able to satisfactorily reproduce: the flow in the surface layer, reproducing the effects associated with the ground roughness, roughness change, or heat flux through the ground; the flow in the Ekman layer together with the interaction between the free flow thermal stability conditions and the boundary layer; the linear and non-linear effects associated with the perturbation induced by a mountain in a stably stratified flow. In the second stage of this work, an extension of the standard Measure-Correlate-Predict method is presented to calculate the wind speed distribution at the prediction site, from transposition relationships and from the wind statistics at the reference site. The validity of the underlying assumptions is confirmed using concurrent data sets that were collected at both the reference and prediction sites. To evaluate the accuracy that can be achieved with the transposition assumptions, a back-prediction is performed using the transposition relationships obtained using the observations. Different types of transposition relationships have been investigated. Finally, the transposition methodology is applied to calculate the wind speed conditions at Mt-Crosin from the Chassera1 data, using the transposition relationships calculated by the numerical model for a range of meteorological situations typical for the area considered. The Mt-Crosin to Chassera1 sector wind speed ratios calculated by the numerical model tend to slightly underestimate those observed. The mean wind speeds obtained from the transposition are underestimated by 7% to 18% at the three measuring mast locations on Mt-Crosin. The yearly energy output that can be produced by a wind turbine in these conditions is underestimated by 8% to 36%. For a further period, the actual energy production of the three installed wind turbines has been compared with the model prediction at hub height, which showed that the transposition results underestimate the actual yearly production by 22% to 24%. From the transposition of the long-term data at Chassera1 (16 years), with the relationships obtained by the numerical model, a wind power potential of between 470 MWh/year (Côte Est) and 596 MWh/year (Côte Nord) is predicted using the characteristics of a Vestas-V44 wind turbine. From the work presented here, it appears that for well-exposed sites such as those located along the Jura Crest, the methods developed are able to give a wind power potential prediction with a similar accuracy as a one year measurement campaign performed on site.

EPFL1998

Potential and uncertainty of wind energy in the Swiss Alps

Albertus Christiaan Kruyt

Switzerland has committed itself to an ambitious energy strategy. It aims to replace the exist- ing nuclear generation capacity with predominantly indigenous renewable resources. Wind power could play a significant role in this transition, yet the wind resource in the mountainous terrain that makes up most of the country is poorly understood. There are indications that this resource could be significant, but studies undertaken so far acknowledge large uncertain- ties. This is because the complex topography of the mountains influences the flow patterns significantly, and these can become partially decoupled from the synoptic flow aloft. This thesis aims to improve the understanding of the wind resource in highly complex terrain, and thereby contribute to a well informed energy transition in Switzerland. We start out by investigating the characteristics of the Swiss wind resource based on data from two meteorological measurement networks. From the pair-wise correlation between stations, it is concluded that wind farms across the country can be combined to produce a stable power output. It is also shown that elevation plays an important role in the wind resource, with the likelihood of sustained low wind speeds decreasing as a function of elevation, while mean speeds tend to increase with elevation. Next, a state of the art Numerical Weather Prediction model is assessed in its ability to simulate wind speeds over the Alps, and is shown to improve drastically upon existing mean wind speed estimates. This same model is then used to calculate the wind turbine capacity that is required to produce significant amounts of wind power, and it is found that the required capacity can be significantly reduced by allowing for wind turbines to be built at high elevations. In the last part of this thesis, smaller areas of the Alpine domain are simulated at high resolu- tions, to investigate the effect of increased model resolution on the accuracy and height of resource assessments. While it is found that optimal model parameterization is dependent on weather and terrain, strong indications of higher wind power potential are found with high resolution models compared to a model at lower resolution. This is explained by the fact that high resolutions are required to properly resolve the complex topography, which has a significant influence on the flow patterns and therefore, on the potential energy production.

EPFL2019