A new integrated analytics approach for wind turbine fault detection using wavelet, RLS filter and random forest

Abstract

ABSTRACT Nowadays wind energy production is growing fast, and the cost of operation and maintenance is growing fast also. Most wind turbines (WTs) are equipped with supervisory control and data acquisition (SCADA) system for system control and logging data. Huge amounts of data acquired for SCADA systems can be used for condition monitoring and fault detection (CMDF) by applying data mining methods. However, the collected data are not used effectively. Few researches are about input and output data optimization and proper feature selection. In this paper, the proposed method regards SCADA data as data points. Wavelet analysis is applied to the input signal to make noise reduction and uses recursive least square (RLS) filter to reduce false alarm rate. On the basis of the general model-based CMDF approach, the random forest algorithm is used to find the best input features. With these methods applied, a more precise output is obtained as well and greatly reduces the false alarm rate as well. Experiments are given with real SCADA data to show the effectiveness of the proposed method.