The Influence of PVTf on Machine Learning Estimation of IGBT Junction Temperature

The literature indicates that the PV inverter is the element of a photovoltaic plant most prone to failure. For this reason, power electronics converters are generally responsible for most photovoltaic projects’ operating and maintenance costs. This article describes the influence of PVTf (power, voltage, ambient temperature, and frequency) on the IGBT junction temperature. We use correlation coefficients and machine learning techniques on the dataset in (Tomislav et al. in IEEE Transactions on Power Electronics 34:7161–7171, 2019). We perform feature selection and the impacts of removing one or more inputs on model training. Data analysis indicates that input power is the most important attribute in the studied dataset and decision tree model presented the best results for the regression problem when compared with MLP and linear regressor. The tool took good advantage of the most important features at the beginning and only considered the less important ones for fine-tuning in the most distant nodes.