An ABC-SVM based Fault Prognosis of Wind Turbines using SCADA Data

Abstract

In modern world, Fault prognosis play an essential role in the safe and reliable operation of wind turbines (WT). An efficient fault prognosis method will help in the earlier identification of WT faults and failures, thereby reducing WT maintenance costs and improving their operating time. WT's are often controlled by Supervisory Control and Data Acquisition (SCADA), which, apart from controlling, provides very rich data pertaining to WT's working parameters which could be used for fault prognosis of WT's. Most of the WT prognosis systems make use of Support Vector Machines (SVMs) in conjunction with SCADA data to predetermine the faults that might occur in the turbines in the near future. In these models, proper selection of SVM parameters is essential for precise fault classification. In this work, an optimised methodology using Artificial Bee Colony Optimisation (ABC) is proposed to find an optimal penalty factor and kernel function parameter that guarantees better classification accuracy for the SVM model. Based on the real time SCADA data availed from a wind farm, it is observed that the proposed ABC-SVM fault prognosis model has a quick convergence rate and good accuracy compared to the other GA-SVM, PSO-SVM, and ACO-SVM based models.