A Generalized Wind Turbine Anomaly Detection Method Based on Combined Probability Estimation Model

Anomaly detection based on the data collected from the supervisory control and data acquisition (SCADA) system is crucial to reduce the failure rate of wind turbines (WTs). The difficulty of this kind of methods is to dynamically identify the threshold for anomaly detection under changing operating conditions. In this paper, a generalized WT anomaly detection method based on the combined probability estimation model (CPEM) is proposed. The CPEM can estimate the conditional probability density function (PDF) of the target variable given changing conditions. Its generalization and accuracy are better than those of the independent probability estimation model because it combines the advantages of various kinds of probability estimation models through linear combination. By using the CPEM, the normal operating bounds under different operating conditions can be obtained, which dynamically form the thresholds for anomaly detection. Meanwhile, with respect to the thresholds, hypothesis testing (HT) is adopted to identify the anomaly by inspecting whether the observations exceed the thresholds at a given significance level, providing sound mathematical support for anomaly detection and making the detection results more reliable. The effectiveness of the proposed method is tested by using the actual data of WTs with known faults. The results illustrate that the proposed method can detect the abnormal operating state of the gearbox and generator much more early than the system fault alarm.