Using Historical Activity Data for RTU Fault Prediction with Machine Learning and Deep Learning

Remote terminal unit (RTU) is the core device of the monitor system in a utility company. A dedicated team of engineers and technicians need to work together to ensure their smooth operation with minimal interruptions. A predictive model can provide a list of RTUs with highest likelihood to fault and will help to prioritize the efforts of maintenance and repair. This paper introduces multiple machine learning and deep learning methods to produce such results based on historical RTU activity data. It shares experience in such practices as experimental design, sampling, feature generation, feature selection and model evaluation.