Hybrid N-BEATS-Based Method for Equipment Assessment and System Risk Prediction in Urban Power Grids

To improve power system risk prediction under complex and extreme operating conditions, a hybrid N-BEATS-based framework is proposed for equipment assessment and loss of load expectation (LOLE) forecasting. The method integrates thermal circuit modeling, online thermal parameter identification via physics-informed neural networks (PINNs), Arrhenius–Weibull temperature–failure mapping, Monte Carlo system risk evaluation, and a hybrid spatiotemporal predictor combining N-BEATS and graph neural networks. Case studies on distribution transformers demonstrate improved thermal parameter identification and reduced LOLE forecasting error compared with benchmark methods. The remainder of the paper is organized as follows: Firstly, it presents thermal modeling and PINN identification; Secondly, it introduces the aging-failure and reliability evaluation; Thirdly, it describes the hybrid forecasting framework; Fourthly, it reports case studies; and finally, it concludes.