A Sample-Enhanced Prediction Method for Insulation Life Assessment of Pumped Storage Units

Abstract

Pumped storage units, serving as core equipment for electrical energy storage, peak shaving, and frequency regulation in new power systems, play a critical role in ensuring the reliability of power supply and the efficiency of energy utilization. This study reviews and compares mainstream methods for insulation life prediction, including traditional aging models, intelligent algorithm-based predictions, and dynamic prediction methods based on online monitoring data. In scenarios with limited samples, the Bootstrap method is proposed to augment sample data and enhance the reliability of parameter estimation. The augmented samples are then utilized as the data source for insulation life prediction of polyimide insulation materials, considering varying temperature and individual differences. The research demonstrates that the prediction method, which updates model parameters using posterior distributions, achieves smaller prediction errors and holds significant value for engineering applications.