Data-Driven Meteorological-Feature-Informed Wind Power Prediction with Machine Learning Decision Trees
Loading...
Links to Files
Permanent Link
Author/Creator
Author/Creator ORCID
Date
2020-01-01
Type of Work
Department
Mechanical Engineering
Program
Engineering, Mechanical
Citation of Original Publication
Rights
Access limited to the UMBC community. Item may possibly be obtained via Interlibrary Loan through a local library, pending author/copyright holder's permission.
This item may be protected under Title 17 of the U.S. Copyright Law. It is made available by UMBC for non-commercial research and education. For permission to publish or reproduce, please see http://aok.lib.umbc.edu/specoll/repro.php or contact Special Collections at speccoll(at)umbc.edu
This item may be protected under Title 17 of the U.S. Copyright Law. It is made available by UMBC for non-commercial research and education. For permission to publish or reproduce, please see http://aok.lib.umbc.edu/specoll/repro.php or contact Special Collections at speccoll(at)umbc.edu
Abstract
Uncertainties exist in wind resource assessments during wind farm pre-construction that create under-performance biases in turbine output power. While wind speed is the main factor in power output, other factors, such as lapse rate and considering wind variation throughout the rotor layer, also have an effect on the predicted power output. To mitigate these uncertainties and to improve the estimation of annual energy production of a wind project, this work uses a decision tree machine learning model to assess the effectiveness of hub height wind speed, rotor equivalent wind speed, and lapse rate as variables in power prediction. Data from a scanning Doppler lidar and a meteorological tower are used to train regression trees and predict the output of a two-megawatt wind turbine. To further address how wind speed is characterized throughout the rotor layer, the decision tree model is trained for four vertical wind profile classifications which showcase the need for multiple calculations of wind speed at various levels of the rotor layer. This approach uses the atmospheric data to correlate the power outputs to wind profiles and meteorological characteristics to be able to predict power responses according to physical patterns. Four sets of predictors are used to train the decision tree model and test its prediction capability: hub height wind speed, rotor equivalent wind speed, hub height wind speed combined with lapse rate, and rotor equivalent wind speed combined with lapse rate. Results indicate that when compared to traditional power curve methods, the decision tree combining rotor equivalent wind speed and lapse rate improves prediction accuracy by 22% for the given data set, while also proving to be the most effective method in power prediction for all classified vertical wind profile types. It was observed that models that incorporated lapse rate into predictions performed better than those without it, showcasing the importance of considering atmospheric criteria in wind power prediction analyses.