An ML-based Strategy to Identify Insulation Degradation in High Voltage Capacitive Bushings

Real-time monitoring of the electric power system has become essential for the efficient operation of the Smart Grid. In this scenario, the power transformer is the core equipment of a substation, and the proper operation of this equipment is of great importance for the power supply. The component responsible for terminal insulation in the power transformer is the bushing. As the bushing's lifetime depends on many factors, including the power demands, the manufacturing method, and the received stresses, online monitoring systems of these components are increasingly being used. An efficient and reliable monitoring system for identifying bushing's problems can reduce maintenance costs. It is possible to reduce the number of shutdowns for inspections and offline tests, and the risks of accidents caused by transformer explosions. Online monitoring systems for capacitive bushings are susceptible to acquisition circuit inaccuracies, noises, and interferences. In addition, bushing behavior can change due to temperature and humidity conditions. These operational parameters can cause fluctuations in online monitoring measurements and represent a challenge for correctly identifying bushing anomalies or degradations. This paper evaluates different machine learning (ML) approaches to identify anomalies in capacitive bushings. We propose the proper selection of features and the most efficient ML strategy to detect anomalies. We based the study on measured data from power transformers under normal and anomalous operation conditions.