Statistical and AI-based reliability assessment of photovoltaic inverters using statistical modeling and machine learning

This study presents a methodology for assessing the reliability of a photovoltaic (PV) inverter by combining classical statistical approaches and machine learning algorithms. The analysis is based on real SCADA data from a PV system and includes measured quantities such as voltages, currents, active power, temperature, cos(φ), frequency and registered alarm states. Statistical profiling of each numerical characteristic was performed by comparing probability distributions - normal, exponential and Weibull. To select the best distribution, the information criteria AIC and BIC were applied, as well as fit criteria such as Pearson's χ² and the Anderson–Darling test. Additionally, a reliability analysis was performed by modelling the time intervals between consecutive failures with a Weibull distribution, calculating the parameters shape (β), scale (η) and mean time to failure (MTTF), visualized by survival and failure functions.