Non-stationary data reorganization for weighted wind turbine icing monitoring with Gaussian mixture model

Abstract

The ice accretion on blades is a serious problem affecting the working performance of the wind turbine. Due to the non-stationary characteristic of the wind turbine, the fault information of icing can usually be buried in the normal change of wind power generation process, which makes the icing monitoring a challenging task. In this paper, a novel data reorganization for weighted wind turbine icing monitoring method is developed to handle this problem. Specifically, a novel condition driven analysis concept to replace the traditional time-driven methods is proposed for the non-stationary wind power generation process. The data can be divided into multiple wind speed slices (WSSs), i.e., condition slices, by cutting indicator variable into multiple equal intervals, where each condition slice can characterize certain characteristics for the concerned operating condition. Subsequently, by evaluating the similarity of WSSs, different wind speed modes (WSMs) are revealed by a step-wise sequential condition partition algorithm. In this way, the process characteristics can be similar within the same WSM while quite different for different WSMs. Further, the distribution of each WSM is then evaluated by Gaussian mixture model and weighted monitoring method is designed by weighting the monitoring statistics with the posterior probabilities of each monitored sample belonging to different Gaussian components. Two real data sets are used to validate the effectiveness of the proposed method. In comparison with other methods, comprehensive results illustrate that the proposed method can transform the non-stationary process into different WSMs, and closely describe the process variations, based on which the accurate icing monitoring results can be achieved.