Method and algorithm of diagnostics modeling for a power transformer based on machine learning

Shaikhullin Artur Zufarovich, Nizamiev Marat Firdinatovich

The emerging technology of forecasting and condition management (PUS) has recently attracted a lot of attention from scientists and industries. The need to increase the availability of equipment and reduce maintenance costs is the driving force behind the development and integration of forecasting and condition management systems. PUS models depend on smart sensors and data generated by sensors. In this article, machine learning-based methods for the development of PUS models based on sensor data for performing fault diagnostics of transformer systems in an intelligent network are proposed. In particular, an algorithm is used to optimize the neural network of back propagation (OP) in order to build high-performance models of fault diagnosis. The models were developed using sensor data called dissolved gas data in the oil of a power transformer. The results obtained demonstrate that the developed algorithm for optimizing the parameters of the neural network is effective and useful; and models based on machine learning have significantly improved the performance and accuracy of diagnostics /fault detection for the power transformer PUS.

Keywords:machine learning; neural network; predicting and controlling the state of a power transformer; fault diagnosis.