Artificial intelligence and machine learning in the era of digital transformer monitoring: Exciting developments at Hitachi Energy

The era of digitalization brings new challenges and new paradigms since transformer users and manufacturers alike are moving towards digital solutions. This transition requires new approaches, new architectures, and new ways of looking at data collection, storage, and assessment. Speed and reliability of actionable information become essential at a time when data is ubiquitous, loads are more complex, and energy production moves from traditional plants to distributed generation. This article intends to show some of the ongoing efforts at Hitachi Energy to address these and other demanding technical and economic issues. Our wind power forecast approach deals with the problem of uncertainty in upcoming power demand. We propose a machine learning model to predict power demand to improve the calculation of loadability and cooling / hotspot calculations. Similarly, our Bushing Tan δ and Capacitance Fault Detection solution uses the error of a model to detect problems with Tan δ and capacitance. Our Probabilistic Fault Tree describes an open-source approach that uses Bayesian networks to find the probability of failure of a specific transformer. Finally, we describe two publications made by our team regarding the use of synthetic data created using the Duval Pentagons to generate a model that diagnoses transformer faults; and a patent regarding the creation of an infrastructure that uses blockchain to anonymize users and provide them with information about their transformer fleet using artificial intelligence