A Photovoltaic Inverter Attenuation Prediction Method based on Computer Simulation and Transformer Algorithm

Abstract

To address the difficulty of timely assessment of photovoltaic (PV) inverter degradation during power-station operation and maintenance, this paper proposes a degradation prediction method that combines mechanism-based simulation data generation with Transformer-based time-series forecasting. First, a PV inverter operation simulation model is developed that considers irradiance, ambient temperature, module temperature, dust accumulation, baseline module degradation, and shading and temperature effects. The model is used to generate long-term samples under multiple operating conditions. Second, daily aggregated features and the performance ratio (PR) are adopted as key health indicators, and a multivariate Transformer forecasting model is constructed to capture long-range dependencies among historical operating states, environmental disturbances, and degradation trends. Simulation results show that, for the 200 h ahead output-power prediction task, the proposed method achieves a mean relative error of 2.64% and a standard deviation of 1.18%. The predicted curve closely follows the trend of power fluctuations. These results indicate that combining simulation data with a Transformer model provides a low-cost, scalable technical approach for PV inverter condition assessment and preventive maintenance. Further validation using real power-station data is still required before engineering deployment.