Wind energy is one of the fastest growing renewable energy sources. According to a report issued by the U.S. Department of Energy (DOE), wind power installation in the United States increased by nearly a factor of 10 in the past decade, from 6350 megawatts (MW) in 2003 to 61,108 MW by the end of 2013 (DOE 2014). The DOE advocates working toward the goal that wind power accounts for 20% of the total electricity generated in the United States by 2030 (DOE 2008).
To manage wind turbines and to plan wind energy production, it is critical to assess wind power generation under a given weather profile. The so-called power curve plays a central role in this task. In the wind industry, the power curve measures the relationship between power output of a turbine and the wind speed. In this article, we estimate the power curve associated with individual turbines at both inland and offshore wind farms using turbine-specific power output data and environmental datameasured from ameteorological mast on the corresponding farm.
POWER curve is widely used in the wind power industry to estimate the output power generated given a a set of operating conditions. It is also widely used for analyzing the health of a turbine and the efficiency of a wind-farm . The widely used convention in the industry, the EIC binning method, considers the wind-speed as the predictor and estimates the power generated. The industry uses this as a ‘gold standard’, but many works have shown that wind speed alone is not sufficient for accurate estimations of wind power generated.
In the wind industry, a power curve refers to the functional relationship between the power output generated by a wind turbine and the wind speed at the time of power generation. Power curves are used in practice for a number of important tasks including predicting wind power production and assessing a turbine’s energy production efficiency. Nevertheless, actual wind power data indicate that the power output is affected by more than just wind speed. Several other environmental factors, such as wind direction, air density, humidity, turbulence intensity, and wind shears, have potential impact.
A plethora of environmental factors like wind direction, air density, temperature also play a major role in the power generation of a turbine. Also, the power readings from a wind-farm show that the variability of wind power for a given speed and operating conditions is very high, suggesting that there is room for improvement in predicting the power curve. The present implementation uses a ‘boosting’ technique to effectively capture the intricacies in the power curve by using incremental improvements of a tree-based model. In the present project, working as a group, we learnt a numerous things which were crucial to step up our knowledge base in machine learning. The course gave us important tools for analysis and prediction of data including basic methods like regression and classification to advanced methods like Gaussian process regression, tree based models, etc.
Using our understanding of methods in class, we employed numerous (almost all) methods to the present data and understood how each type of learning methods behave with the data. Also, we took home the modeling approaches required to model real-life data, which depends on a numerous factors than just a few obvious variables considered for studies. We learnt to use ML techniques like feature engineering, variable selection, parameter tuning and modeling techniques like grid-search to enhance our predictive models. We also wanted to explore the field of deep-learning, by looking at it’s wide deployment in the field of ML.
Although deep-learning (a neural net framework) was not our go-to model used finally due to low prediction accuracies than the boosting model we developed, we got an understanding on how black-box model works. After exploration of various models (which will be shown in the comparative analysis section in the sequel), we chose Boosting as our final model. Although, boosting is also viewed as a black-box technique by many, we explored the effects of variables within a boosting model by using partial dependency plots and supporting our arguments on why the boosting model is to be used.
- Report contains submitted results
- Python-IPYNB file contains all the code.
- PPT file contains presentation for this project.