Tech ID: 20-046
Inventors: Daniel Vassallo, Raghavendra Krishnamurthy, Harindra Fernando
Date Added: June 25, 2020
A neural network technology used for vertical extrapolation of wind speeds.
Wind power output has increased by over 400% in the past 10 years, and currently accounts for 25% of total renewable power capacity. Utility-scale wind energy is among the cheapest of clean energy technologies, but the unpredictability and variability of wind makes it imperative to investors that they harvest the maximum energy production from each turbine installed. To do so, energy investors must accurately predict the average wind speed at different wind turbine heights and across the wind field. Current parameterizations that are used to model and map wind energy have proven to inadequate and have in many cases led to unacceptable financial losses. Imperfect parameterizations lead to inaccurate flow predictions and large modeling uncertainties which are detrimental to the economic operation of wind farms. Recent studies have found that neural networks combined with lidar data can more accurately extrapolate wind speed, but it is unclear if these methods work well for more complex terrains. State of the art modeling methods are static, and a model created for one site will likely perform poorly at another site. There exists a need for a reliable method that can model chaotic systems and be used to produce high-accuracy wind speed forecasts in various terrains.
Researchers at the University of Notre Dame have developed a machine learning method to improve wind speed extrapolation utilizing artificial neural networks (ANNs) with atmospheric variables as inputs. ANNs consist of a multilayered network of nodes that compute an output from a set of meteorological inputs, thus enabling customized models for specific sites and more accurate extrapolation of expected wind speeds. Three inputs, turbine intensity, normalized current wind speed, and normalized previous wind speed most reliably improved ANN accuracy - up to 53% better than that obtained by simple power law extrapolation. ANNs are skilled at pattern and bias recognition, have the capability to interpret chaotic, nonlinear systems and are able to help models find trends that determine atmospheric occurrences. Feature extraction and selection and sufficient data not only increases ANN wind speed extrapolation accuracy but can also improve current industry standards.
- Improved correlation with wind speed
- Decrease in total extrapolation error and variability
- Feed-forward control of wind turbines and wind farms
- Prediction of yaw misalignment
- Optimizing wake steering approaches
Technology Readiness Level
TRL 3 – Experimental Proof of Concept
Decreasing Wind Speed Extrapolation Error via Domain-Specific Feature Extraction and Selection. doi:10.5194/wes-2019-58