A comparison of edge detectors in the framework of wake pattern modeling for wind turbines

Abstract

To monitor wind turbine health, wind farm operators can take advantage of the historical SCADA (supervisory control and data acquisition) data to generate the wake pattern beforehand for each wind turbine, and then decide in real time whether observed reduction in power generation is due to wake or true faults. In our earlier efforts, we proposed an effective wake pattern modeling approach based on edge detector using Linear Prediction (LP) with entropy-thresholding, and smoothing using Empirical Mode Decomposition (EMD) on the wind speed difference plots. In this paper, we compare the LP based edge detector with two other predominant edge detec-tors, Sobel and Canny edge detectors, to quantitatively justify the appropriateness and effectiveness of the LP based edge detector in wind turbine wake pattern analysis. We generate a fused wake model for the turbine of interest with multiple neighboring turbines, and then analyze the wake effect on turbine power generation. With a fused wake pattern, we do not need to identify the individual source of wake any more. We expect that wakes cause reduced wind speed and hence reduced power generation, but we have also observed from the SCADA data that the wind turbines in wake zones tend to overreact when the wind speed is not yet close to the high-wind-shut-down threshold, which causes further power generation loss.