A fast and effective method for reliability assessment of PV inverter

• A GPRAL-based surrogate model predicts IGBT junction temperatures with RMSE=0.18 using 130 electro-thermal simulations. • An adaptive K-means clustering method simplifies mission profiles by 85 % while keeping damage errors below 4 %. • An integrated framework enables high-resolution lifetime mapping and identifies robust inverter design regions. Reliability-oriented design of photovoltaic (PV) inverters requires repeated evaluation of power semiconductor lifetime under operating conditions, a process that is often computationally prohibitive during early-stage parameter optimization. This study proposes an integrated framework that combines Gaussian Process Regression with Active Learning (GPRAL) for data-efficient surrogate model and an adaptive K-means clustering method for mission profile simplification. The GPRAL surrogate achieves high prediction accuracy (RMSE = 0.18) using only 130 electro-thermal simulations, enabling rapid screening of inverter design schemes. The clustering-based simplification reduces annual simulation days from 365 to approximately 43–44 while maintaining annual fatigue damage estimation errors below 4 %, resulting in over 85 % savings in computational time. The framework is validated on a 175 kW two-stage grid-connected PV inverter operating under two distinct climatic profiles—Datong (temperate) and Lhasa (plateau), through high-resolution reliability mapping and gradient sensitivity analysis across 700,000 design points. Results demonstrate the ability to identify robust parameter regions that meet 25-year lifetime targets and to reveal sensitivity hotspots. The proposed approach offers a scalable, accurate, and practical tool for reliability-constrained PV inverter design in diverse operating environments.