Towards an improved feature-selection approach for oil-immersed transformer top-oil temperature calculation

Power transformers are necessary components for the reliable operation of the power grid. However, the increasing use of renewable energy technology with highly dynamic power generation has created new scenarios, which affect the lifetime of such devices. There exist standards that calculate the top-oil temperature, hottest-spot temperature and aging factor of transformers based on empirical models, such as IEC 600076-7. However, the accuracy of these models may be limited due to their steady-state nature. Although these formulations have been improved with machine-learning techniques through adaptation of experimental thermal equations to specific contexts by means of ad-hoc modelling, the systematic and heuristic analysis of the influence of different environmental and meteorological variables has not been addressed. In this context, this paper presents a novel systematic parameter-selection process to improve transformer top-oil temperature estimation, reducing the prediction error by half, as confirmed with the results. The proposed approach has the potential to deliver better health management of transformers through an intelligent feature selection process.