A production economics analysis for quantifying the efficiency of wind turbines

Abstract

We quantify the productive efficiency of a wind turbine, using power output and environmental variable data, measured either at the turbine or at a meteorological mast near the turbine. The methods described can potentially help with decision makings in asset procurement, maintenance planning, or wind turbine control optimization. The current recommendation from the International Electrotechnical Commission regarding turbine performance evaluation is to use a power curve or power coefficient. What is commonly used in practice is the average performance power curve or power coefficient. When using the power curve to quantify productive efficiency, one crucial shortcoming is the lack of a common best performance benchmark, while the power coefficient approach uses an absolute efficiency measure that is not achievable. We introduce a new approach for efficiency quantification based upon production economics' concepts which provides estimates of a best performance benchmark. Our specific approach has two main components: (a) a best performance power curve is estimated and used together with the average performance curve to show how well a turbine has performed relative to its full potential; and (b) a covariate matching procedure is developed to control for environmental influences for the comparison of turbine performances over different periods. Through a simulation study, we demonstrate that the proposed efficiency is more sensitive to potential changes in the turbine. When analyzing multi-year wind turbine data, we observe that the turbine's efficiency is improving during the first 2 years of operation and then remains relatively constant during years 3 and 4.

A Comprehensive Review of Artificial Intelligence and Wind Energy

“…Decision making is oriented in two directions: planning (both short and long term) and variable assessment. Among planning tasks are to decide the best suited components [217,218], short or long term planning [159, 187], power prediction [219,220], route optimization [221,222], state definition [223,224], WT manufacturing [225], and risk assessment [226][227][228] and mitigation [229,230]. Variable assessment is aimed at defining the best suited variables to perform O&M tasks [231][232][233] and to reduce computational costs [211,234], as well as to predict fault apparition considering signal uncertainties [235][236][237] or failure probability [238,239].…”

Gaussian process aided function comparison using noisy scattered data

“…We apply our method to the four datasets as used by Hwangbo et al (2017), which also constitutes a large portion of Chapter 6 of Ding ( 2019), and we download the four datasets from the book website of Ding (2019). Each dataset corresponds to a different turbine.…”

“…Each of the four datasets comprises four years of data. We conduct a year to year comparison for each turbine, as done in Hwangbo et al (2017). For this reason, each turbine's dataset is divided into four annual datasets.…”

“…Power curves are used to characterize the performance of wind turbines (IEC, 2005). Hence, comparing power curves plays a critical role in turbine performance comparison and benchmarking, helping identify better designs and best practice (Hwangbo et al, 2017). One key feature of the datasets arising from wind turbines, however, is that the input conditions (e.g.…”

A Case Study of Space-time Performance Comparison of Wind Turbines on a Wind Farm

“…On this point, Chapter 5 in [3] provides an elaboration backed by numerical evidences. Another shortcoming of the existing performance evaluation methods is the followingalthough the power curve is a functional curve, or a functional response surface while considering multi-dimensional inputs, almost all existing methods reduce the functional curve into a scalar metric, be it the annual energy production (AEP) [1], or the power coefficient [7,8], or a recently proposed productive efficiency measure [9]. We believe that it would be more ideal to compare two functional curves directly without reducing them into a scalar metric because in comparing two functional curves one can identify the regions of difference in power production, which can lead to valuable clues behind performance changes and help figure out the root causes.…”