Risk Assessment, Optimization and Surrogate Modeling for the Water Level Control System

Maintaining optimal water levels in pressurizers is crucial to ensuring the stable operation of nuclear reactor primary circuits. This paper presents an integrated framework for system risk analysis and operational strategy optimization in water level control systems. We first develop a comprehensive control system model and employ the dynamic event tree (DET) methodology to investigate the temporal evolution of system risk during extended operations. Then we develop a multi-objective optimization algorithm to identify optimal operational strategies that minimize long-term system risk while balancing safety requirements with operational efficiency. To address the computational challenges of state-space explosion inherent in long-sequence DET analysis, we propose a neural network-based surrogate model that enables rapid risk prediction while maintaining accuracy. Experimental results demonstrate that our approach achieves significant computational efficiency gains in risk prediction while providing actionable insights for risk-informed decision-making. The neural network surrogate model reduces the majority of the prediction time compared to conventional DET methods without compromising prediction accuracy. This research offers a practical solution for real-time risk assessment and optimal control in complex nuclear power systems, with potential applications in other safety-critical industrial domains.

Keywords: Dynamic Event Tree, Multi-objective Optimization, Neural Network Surrogate, Risk Assessment.