An Alert-Ambient Enrolled Deep Learning Model for Current Reliability Prediction of Weather Impacted Photovoltaic Inverter

An artificial neural network-based deep learning model is designed and evaluated in this work to predict a reliability score of a photovoltaic (PV) inverter impacted by irradiance, temperature, and relative humidity. In doing this, the proposed method counts the inverter's alert types, propagation, and behavioral relation with output power. In parallel with transient events over time-varying inverter stresses or inverter alerts, varying weather patterns affect the cumulative internal degradation of sub-components and hence associate inverters' reliability. Continuous time-domain inverter performance degradation prediction is worked on in this work, by developing an “Alert-Ambient” supervised deep learning model consisting of two stages of classification and regression, trained and tested by field-extracted data. Correlation results showed alerts are positively correlated with ambient temperature and negatively with humidity. The classification stage was able to predict the correct alert class from weather data with over 80 % accuracy. Next to it, a regression stage, triggered by weather patterns and classes, was able to predict inverter performance in the range of 0% to 100%. The trained and tested model is suitable to use in any digital twin representation of a PV inverter such that real-time ambient parameters can be fed to predict its output power or current availability or reliability.