Research Papers

Prediction of Status Patterns of Wind Turbines: A Data-Mining Approach

[+] Author and Article Information
Andrew Kusiak, Anoop Verma

Department of Mechanical and Industrial Engineering, University of Iowa, 3131 Seamans Center, Iowa City, IA 52242-1527andrew-kusiak@uiowa.edu

J. Sol. Energy Eng 133(1), 011008 (Jan 28, 2011) (10 pages) doi:10.1115/1.4003188 History: Received February 14, 2010; Revised November 10, 2010; Published January 28, 2011; Online January 28, 2011

This paper presents the application of data-mining techniques for identification and prediction of status patterns in wind turbines. Early prediction of status patterns benefits turbine maintenance by indicating the deterioration of components. An association rule mining algorithm is used to identify frequent status patterns of turbine components and systems that are in turn predicted using historical wind turbine data. The status patterns are predicted at six time periods spaced at 10 min intervals. The prediction models are generated by five data-mining algorithms. The random forest algorithm has produced the best prediction results. The prediction results are used to develop a component performance monitoring scheme.

Copyright © 2011 by American Society of Mechanical Engineers
Your Session has timed out. Please sign back in to continue.



Grahic Jump Location
Figure 1

Component degradation curve

Grahic Jump Location
Figure 2

Example status descriptions of four categories

Grahic Jump Location
Figure 3

Category 1 status data for 100 turbines: (a) histogram and five probability density functions and (b) probability-probability (P-P) plot

Grahic Jump Location
Figure 4

Framework for the prediction of status patterns

Grahic Jump Location
Figure 5

Flow chat for identification of status patterns

Grahic Jump Location
Figure 6

Frequency distribution of identified status patterns: Turbines (a) 1–20, (b) 21–40, (c) 41–60, (d) 60–80, and (e) 81–100

Grahic Jump Location
Figure 7

Frequency plot of all status patterns identified in 100 turbines

Grahic Jump Location
Figure 8

Description of the data set generation

Grahic Jump Location
Figure 9

Data sampling steps

Grahic Jump Location
Figure 10

Parameters selected for prediction of status patterns

Grahic Jump Location
Figure 11

The confusion matrix for performance evaluation of algorithms

Grahic Jump Location
Figure 12

Prediction accuracy of five data-mining algorithms for different weight values

Grahic Jump Location
Figure 13

Comparison of the actual values and the values predicted by the random forest algorithm at time period t+30

Grahic Jump Location
Figure 14

The actual and the predicted status pattern 274=>275=>276: (a) t+10 and (b) t+60

Grahic Jump Location
Figure 15

Distribution of alarm signals of the status pattern 274=>275=>276




Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging and repositioning the boxes below.

Related Journal Articles
Related eBook Content
Topic Collections

Sorry! You do not have access to this content. For assistance or to subscribe, please contact us:

  • TELEPHONE: 1-800-843-2763 (Toll-free in the USA)
  • EMAIL: asmedigitalcollection@asme.org
Sign In