Research on Intelligent Algorithm for Reliability Evaluation of UHVDC Transmission System

The reliability assessment of ultra-high voltage direct current (UHVDC) transmission systems is crucial for the safe operation of the power grid. Traditional methods have limited accuracy due to insufficient static modeling and multisource data utilization. This study proposes a hybrid framework of “traditional theoretical modeling+intelligent algorithm optimization”, which constructs an explicit reliability model of the system functional structure through a universal generating function (UGF). The LSTM-GCN (Long Short Term Memory Network Graph Convolutional Network) integrates the spatiotemporal characteristics of multi-source data such as electrical quantities, equipment status, and environmental parameters to dynamically modify the model parameters. Based on reinforcement learning (RL), the risk cost collaborative optimization of maintenance strategies is achieved. The experimental results show that the reliability evaluation accuracy of the framework on a certain ± 800kV UHVDC system dataset reaches $96.2 \%$, which is $18.7 \%$ higher than the traditional UGF+Markov model. The single evaluation time is reduced from 220 ms to 50 ms, the annual operating risk of the system is reduced by $12.6 \%$, and the maintenance cost is reduced by $9.5 \%$. The introduction of metadata management module and Bayesian regularization technology ensures the traceability and uncertainty quantification ability of the evaluation process. The research results provide a solution that combines physical interpretability and data-driven advantages for intelligent operation and maintenance of UHVDC systems under complex working conditions. Future research can further integrate multimodal monitoring data, introduce federated learning techniques, and optimize model lightweight deployment performance.