Session: 08-15-01: Wind and Water Power

Paper Number: 94173

94173 - A Robust Hybrid Machine Learning-Based Modeling Technique for Wind Power Production Estimates 

Quantifying power output from a wind turbine is essential to optimize mechanical and electrical performance of the turbine, as well as has implications in financial planning for any wind energy project. Wind power curves play important roles in wind power forecasting, wind turbine condition monitoring, estimation of wind energy potential and wind turbine selection. Power curves for wind turbines establish a relation between wind speed and electrical power output. The current standard IEC 61400-12-1:2017 specifies a procedure for measuring the power performance characteristics of a single wind turbine and applies to the testing of wind turbines of all types and sizes connected to the electrical power network. Because of uncertainties associated with weather conditions such as with wind, and the non-linear dependency of power on wind speed, the power curve combines the characteristic of the turbine together with the statistical features of the wind. However, systematic averaging errors can be introduced due to the above-mentioned nonlinearity, and therefore producing effective power curves from raw wind data is inherently challenging. For this reason, the use of non-linear learning methods to accurately develop wind power curves is gaining attention. In practice, it is also a challenging task to produce effective and reliable wind power curves from raw wind data due to the presence of outliers formed in unexpected conditions, e.g., wind curtailment and blade damage. In this paper, we present a comprehensive review of power curve modeling methods using machine learning techniques, including regression models, ANN-based models, classifier and clustering models, and adaptive neuro-fuzzy inference-based models. Most models rely on the fact that all outliers will be removed from the raw wind data. However, that is hardly ever the case and therefore designing robust models and robust objective functions can be two effective ways to obtain accurate power curves in the presence of outliers. We propose a robust non-parametric hybrid power curve modeling technique – a fuzzy clustering (FCM) method is first used to identify outliers in the dataset, followed by neural network (ANN) models that are trained using the outlier-free data, followed finally by Support Vector Machine (SVM) based regression model to obtain accurate power curves. The results show the proposed method is superior to probabilistic models that use error-functions to generate accurate power curves and that the proposed hybrid model can generate more accurate power output estimations in the presence of outliers compared to deterministic models such as integrated curve fitting models that are known to be robust.  

Presenting Author: Amit Banerjee Penn State University Harrisburg

Presenting Author Biography: Amit Banerjee is an Associate Professor of Mechanical Engineering at Pennsylvania State University Harrisburg. His research interests include artificial intelligence, computational intelligence in manufacturing, evolutionary computation, machine learning, robotics and automation, and application of computational technologies in engineering education. He teaches courses in robotics, manufacturing, design and mechanics. He has an M.Des in Product Design from the Indian Institute of Science, Bangalore and a Ph.D. in Mechanical Engineering from the New Jersey Institute of Technology. He has also been a postdoctoral research fellow at the Evolutionary Computational Systems Lab in the department of Computer Science and Engineering at the University of Nevada Reno.

Authors:

Amit Banerjee Penn State University Harrisburg
Issam Abu-Mahfouz Penn State University Harrisburg
Jianyan Tian Taiyuan University of Technology
Ahm Esfakur Rahman Penn State Univ Harrisburg