Fault-related Alarm Detection of a Wind Turbine SCADA System

Renewable energy sources and production facilities have increasing prevalence around the world. Thanks to technology, this type of energy production facilities use and produce more data day by day. Wind power turbines use SCADA (supervisory control and data acquisition) systems to monitor and control production, alarms, faults, etc. Data that is produced by SCADA systems are used for fault and alarm prediction. In this study, the SCADA system data of a wind farm located in the southeast of Turkey is used to predict the system's alarms. While some alarms are about faults, it could be that the others are about non-fault situations. This study aims to predict fault-related and non-fault-related alarms. SMOTE is used to balance unbalanced classes to increase the performance of the model. The proposed model consists of 4 main steps: The first step is data acquisition from SCADA data. The second step is data analysis and pre-processing. The third step of the model is feature selection, and the last step is classification study. SVC (support vector classifiers) and decision trees are compared for the classification step, and according to the performance results, the decision tree is selected to model prediction.