Wind turbine Cp,max and drivetrain-losses estimation using Gaussian process machine learning

E Hart,W E Leithead,J Feuchtwang

doi:10.1088/1742-6596/1037/3/032024

E Hart, W E Leithead + Show 1 more

Open Access

https://doi.org/10.1088/1742-6596/1037/3/032024

Copy DOI

Abstract

In this paper it is shown that measured data in a wind turbine, available to the controller, can be formulated into a polynomial regression problem in order to estimate the turbine’s maximum efficiency power coefficient, Cp,max, and drivetrain losses, assuming the latter can be well approximated as being linear. Gaussian process (GP) machine learning is used for the regression problem. These formulations are tested on data generated using the Supergen Exemplar 5 MW wind turbine model, with results indicating that this is a potential low cost method for detecting changes in aerodynamic efficiency and drivetrain losses. The GP approach is benchmarked against standard least-squares (LS) regression, with the GP shown to be the superior method in this case.

Highlights

As wind turbine assets grow in size and move further offshore, where access becomes more problematic and expensive, the need for increased reliability becomes even more pronounced
Rather than proposing a new measurement technique per se, we instead look at how existing data available to a wind turbine controller can be more fully utilised in order to try and detect changes in a turbine’s aerodynamic efficiency and drivetrain losses
The results presented here first demonstrate that Gaussian process (GP) is superior to LS in this case, with LS results being so scattered as to make them effectively useless

Summary

Introduction

As wind turbine assets grow in size and move further offshore, where access becomes more problematic and expensive, the need for increased reliability becomes even more pronounced. There is a need for a diverse range of monitoring techniques with which faults and behavioural changes in wind turbines can be quickly detected and appropriate steps taken. There have been a huge number of new measurement techniques developed in recent years, all with their associated costs. Rather than proposing a new measurement technique per se, we instead look at how existing data available to a wind turbine controller can be more fully utilised in order to try and detect changes in a turbine’s aerodynamic efficiency and drivetrain losses. Changes in the mechanical losses in the drivetrain can be indicative of damage or an imminent failure. Prior warning of any changes, and the tracking of such changes over time, is desireable to inform operation and maintenance (O&M) for these wind turbines

Results

Discussion

Conclusion