Autonomous decision-making of operational schedules for lead–bismuth fast reactors based on BOHB optimization

Abstract

This study proposes a hybrid autonomous decision-making framework for Lead-Bismuth Eutectic (LBE) Cooled Fast Reactors (LBEFRs), designed for deployment in remote and maritime environments where real-time human intervention is limited. To address challenges posed by uncertain missions, variable loads, and fault scenarios, the framework integrates Bayesian Optimization with Hyperband (BOHB) to jointly optimize discrete operational schedules and continuous control targets in a high-dimensional, mixed-integer space. A backpropagation neural network (BPNN) surrogate model is employed to approximate thermal–hydraulic behavior with minimal computational overhead, enabling real-time decision-making. The framework targets fault-tolerant conditions in which the reactor retains partial operability, aiming to maintain system functionality rather than replicate conventional safety responses. It is designed to complement, rather than replace, traditional safety systems in mission-critical scenarios. The framework’s performance is validated under two representative fault conditions: a steam line rupture and a turbine overspeed event, where it autonomously derives reconfiguration strategies that stabilize system parameters while satisfying safety constraints. Results demonstrate strong convergence, high accuracy, and robust adaptability, confirming the framework’s effectiveness for operational strategy optimization in LBEFRs and its potential for application in next-generation intelligent nuclear systems.