Wind Farm Performance Monitoring with Exploratory Factor Analysis

Transcription

1 Wind Farm Performance Monitoring with Exploratory Factor Analysis Niko Mittelmeier, Katharina Neumann Analysis of Operating Wind Farms, EWEA Workshop, Malmö

2 Agenda Introduction About Exploratory Factor Analysis EFA applied to wind turbine data A monitoring demonstration Summary and Outlook 2

3 Introduction Objective: Robust detection of changes in turbine behaviour Massive amount of data is collected from each turbine Proposal: Advanced statistical models 3

4 Agenda Introduction About Exploratory Factor Analysis EFA applied to wind turbine data A monitoring demonstration Summary and Outlook 4

5 About Exploratory Factor Analysis Reduce observed variables to fundamental underlying unobserved variables 5

6 About Exploratory Factor Analysis X i = a 1 F 1 + a 2 F 2 + a p F p + e i X i : F 1 p : a 1 p : e i : ith observed variable common factors (underlying unobserved variable) loadings not explained by the common factors (error term) 6

7 About Exploratory Factor Analysis Develop covariance matrix from raw data No missing values are allowed Kaiser-Harris criteria to estimate the number of common factors (number of eigenvalues of cov. matrix > 0) Estimate loadings (are not unique) Rotate matrix (maximize one loading and minimize the others) Maximum likelihood Principal axis (generalized) weighted least square Minimum residual ( minimizes the square sum of the off diagonal ) 7

8 Agenda Introduction About Exploratory Factor Analysis EFA applied to wind turbine data A monitoring demonstration Summary and Outlook 8

9 EFA applied to wind turbine data Choose the variables X i to be observed: Pitch Power Torque Revolution speed Wind speed 9

10 EFA applied to wind turbine data F1 F2 10

11 EFA applied to wind turbine data Standardize Observations: X i = X i X i S i 11

12 EFA applied to wind turbine data minimum residuals (method) Oblique: correlation between the Factors has been allowed F1 F2 12

13 Agenda Introduction About Exploratory Factor Analysis EFA applied to wind turbine data A monitoring demonstration Summary and Outlook 13

14 A monitoring demonstration Turbine Data from different operational modes 14

15 A monitoring demonstration To cover the full turbine behaviour, at least five variables are necessary torque torque pitch pitch power power wind speed revolution speed Normal operation Mode 1 Mode 2 Wind speed revolution speed 15

16 A monitoring demonstration The correlation of the two common factors shows clearly a different behavior Normal operation Mode 1 Mode 2 16

17 EFA applied to wind turbine data Normal operation Mode 1 F1 F1 F2 F2 F1 F2 Mode 2 17

18 A monitoring demonstration Choosing a smaller sample from the data Normal operation Mode 1 Mode 2 No suspicious behaviour visible in this plot 18

19 A monitoring demonstration The correlation of the two common factors shows clearly a different behavior but less samples reduce the clarity Normal operation Mode 1 Mode 2 19

20 A monitoring demonstration Wind Farm Scan 20

21 Agenda Introduction About Exploratory Factor Analysis EFA applied to wind turbine data A monitoring demonstration Summary and Outlook 21

22 Summary and Outlook Summary: Turbine behaviour can be expressed with two common factors Exploratory factor analysis needs a certain sample size With 10 min data, the time to detect changes would be to long Higher frequency data (e.g. 30s data) needs more memory space Factors can be used as statistical summary of high frequency data Outlook: Check minimal sample size Check optimal data frequency 22

23 Thank you for your attention SENVION SE Mittelmeier Niko Wind Farm Performance Specialist SENVION SE All rights reserved. No part of this document may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photography, recording, or any information storage and retrieval system, without permission from SENVION SE.