Deep Learning for Condition Monitoring in High-Voltage Transformers

Deregulation, competition in power industry is growing, and there is an imperative need for cost effective asset management and improve the operation efficiency. We present the development of an Online Condition Monitoring System (OCMS) based on deep learning for high voltage transformers, which is an online monitoring system that provides real-time transformer condition based on historical data. The proposed system is based on predictive maintenance of normally operating transformers with real time operational data. Using a Recurrent Neural Network (RNN) framework so as to model spatio temporal dependencies in the data, we can detect degradation patterns earlier when critical failures are imminent. There are timely alerts offers from OCMS, which help us to identify issues accurately as well as to take informed maintenance decisions on basis of data. The system remains solely concerned with the normal behavior of the transformer and does not resort to fault-based diagnostics (for instance dissolved gas analysis), by making it more proactive and efficient. Comparison with a standard utility benchmark is made in regards to precision, online capability and cost effectiveness to demonstrate the system’s capability in improving up time and reliability in modern power networks.