Multivariable Power Curve Modeling and Condition Monitoring Based on Gaussian Process

The power curve can intuitively reflect the performance of the wind turbine. Modeling the power curve and monitoring its outcomes can promptly detect anomalies in wind turbine operations, thereby enhancing equipment reliability. This paper employs an adaptive Density-based Spatial Clustering of Applications with Noise (DBSCAN) method to eliminate outliers, utilizes adaptive feature selection to determine modeling variables, and constructs modeling samples by adding observation vectors to the process memory matrix. A multivariate power curve model is established based on Gaussian processes for power prediction. The technique employs a sliding window for residual analysis to track and identify unusual shifts in predicted residuals, activating alerts for anomalies when required. The sample outcomes demonstrate that the approach presented in this document is capable of developing a highly accurate power curve model, precisely pinpointing deviations in wind turbine performance, overseeing model malfunctions, and initiating fault notifications.