Decreasing wind speed extrapolation error via domain-specific feature extraction and selection

Abstract

Abstract. Model uncertainty is a significant challenge in the wind energy industry and can lead to mischaracterization of millions of dollars' worth of wind resources. Machine learning methods, notably deep artificial neural networks (ANNs), are capable of modeling turbulent and chaotic systems and offer a promising tool to produce high-accuracy wind speed forecasts and extrapolations. This paper uses data collected by profiling Doppler lidars over three field campaigns to investigate the efficacy of using ANNs for wind speed vertical extrapolation in a variety of terrains, and it quantifies the role of domain knowledge in ANN extrapolation accuracy. A series of 11 meteorological parameters (features) are used as ANN inputs, and the resulting output accuracy is compared with that of both standard log-law and power-law extrapolations. It is found that extracted nondimensional inputs, namely turbulence intensity, current wind speed, and previous wind speed, are the features that most reliably improve the ANN's accuracy, providing up to a 65 % and 52 % increase in extrapolation accuracy over log-law and power-law predictions, respectively. The volume of input data is also deemed important for achieving robust results. One test case is analyzed in depth using dimensional and nondimensional features, showing that the feature nondimensionalization drastically improves network accuracy and robustness for sparsely sampled atmospheric cases.

New methods to improve the vertical extrapolation of near-surface offshore wind speeds

“…This performance held even when a model was trained at one measurement site and tested at others up to 100 km away, i.e., through a round-robin approach. In the offshore environment, Vassallo et al (2020) used a deep neural network to extrapolate near-surface winds in offshore California during a 1-month period, and they also found improvement relative to conventional techniques; however, the time period was short, and a round-robin approach was not applied.…”

Utilizing physics-based input features within a machine learning model to predict wind speed forecasting error

“…The forecast skill of observation-based methods generally reduces with forecast lead time within 1 h, and numerical models have higher skill in forecasting larger lead times (> 3 h; Haupt et al, 2014). Statistical learning methods have proved to be particularly effective from about 30 min to approximately 3 h ahead (Mellit, 2008;Wang et al, 2012;Yang et al, 2012;Morf, 2014), and roughly this time frame is thus the focus for this study. The shortest forecast predicts wind speeds 10 min ahead, roughly within the turbulent spectral band (Van der Hoven, 1957).…”

“…Recently, a handful of investigations have begun to determine which variables may be most useful for these models. Vassallo et al (2020) showed that invoking turbulence intensity (TI) can vastly improve vertical wind speed extrapo-lation accuracy. Similarly, Li et al (2019) showed that TI improves wind speed forecasting on multiple timescales, while Optis and Perr-Sauer (2019) showed that both atmospheric stability and turbulence levels are important indicators for wind power forecasting.…”

Atmospheric Boundary Layer Wind Profile Estimation Using Neural Networks Applied to Lidar Measurements

“…For example, Vassallo, Krishnamurthy and Fernando [23] suggest that the atmospheric stability plays a key role in the estimation of wind speed. Between the different meteorological parameters that influence the atmospheric stability, it can be found the temperature and radiation, which are directly related to the day of the year and time of measurement.…”

“…Although there are previous studies that use artificial neural networks (ANNs) to interpolate wind profiles, they present several differences with the methodology proposed in this one. Vassallo, Krishnamurthy and Fernando [23] measurements at multiple heights, not just the lowest, for doing the interpolation. So, for example, in order to estimate the wind at 120 m, they used the values observed at 100, 80, 60 and 40 m AGL (above ground level).…”