Probabilistic Safety Margin analysis method for SBLOCA in nuclear power plants based on a Multi-Level Hybrid Efficient Solution Strategy

To rapidly and accurately assess the Probabilistic Safety Margin (PSM) of nuclear power plants, this paper proposes a Probabilistic Safety Margin Analysis Method Based on a Multi-Level Hybrid Efficient Solution Strategy (PSMA-MLHESS) that addresses three major bottlenecks in traditional PSM analysis: branching explosion of accident sequences, large sample sizes for single sequences, and time-consuming multi-sequence simulations. The strategy achieves optimization through three approaches: First, it streamlines sequences by integrating expert knowledge with a hybrid particle swarm optimization algorithm via a three-step process—“pre-classification, limit state analysis, and probabilistic cutoff screening”. Second, it constructs single-sequence datasets by focusing on safety boundaries through adaptive sampling, thereby controlling volume while enhancing efficiency and precision. Third, it replaces thermal–hydraulic programs with a Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) hybrid surrogate model, accelerating similar sequence modeling through transfer learning. Validation using a typical dual-loop pressurized water reactor scenario — a small-break loss-of-coolant accident at 100% and 105% power — demonstrated that PSMA-MLHESS maintains the accuracy of traditional methods while achieving an order-of-magnitude improvement in analysis efficiency. This provides an efficient tool for PSM evaluation and decision-making in scenarios such as power plant power increase and life extension.