Skip to Main Content
The need for dynamic loading of power components in the deregulated electricity market demands reliable assessment models that should be able to predict the thermal behavior when the load exceeds the nameplate value. When assessing network load capability, the hot-spot temperature of the components is known to be the most critical factor. The knowledge of the evolution of the hot-spot temperature during overload conditions is essential to evaluate the loss of insulation life and to evaluate the consequent risks of both technical and economical nature. This paper discusses an innovative grey-box architecture for integrating physical knowledge modeling (a.k.a. white-box) with machine learning techniques (a.k.a. black-box). In particular, we focus on the problem of forecasting the hot-spot temperature of a mineral-oil-immersed transformer. We perform a set of experiments and we compare the predictions obtained by the grey-, white-, and black-box approaches.