Enhanced Gaussian Process Regression for Diagnosing Wind Energy Conversion Systems

Fault detection and diagnosis techniques are increasingly important to ensure robust and resource efficient operation of Wind Energy Conversion (WEC) systems. In this context, this paper presents a Reduced Enhanced Gaussian Process Regression (REGPR)-based Random Forest (RF) technique (REGPR-RF) to identify and diagnose faults occurring in a nonlinear WEC systems. The proposed technique uses REGPR technique for features extraction and selection from raw sensor data. Then, these selected features are fed to RF classifier to reliably detect and classify faults. The use of REGPR to learn features avoid the dimension problems and improves the classification performance significantly with a small number of training data. The results obtained by REGPR-RF are compared to those obtained with other conventional classifiers (Support Vector Machines (SVM), Naive Bayes (NB),.). The results show that the developed REGPR-RF technique achieve higher accuracy (99.99%) with small data sets.