Welcome to the PSAM 18 Abstract Status page.
Abstract KW244Full Paper + Presentation
Resilient Autonomous Operation for SMR Abnormal Conditions based on the Multi-Agent Reinforcement Learning and Abstraction Hierarchy
Small modular reactors (SMRs) are emerging as a strategic energy solution as existing grid infrastructure struggles to meet rapidly growing electricity demand. Their load-following capability complements intermittent renewables such as solar and wind, while providing stable, carbon-free baseload generation. These characteristics make SMRs well-suited for distributed deployment in response to surging demand from sectors such as AI data centers.
Responding to abnormal conditions is one of the most demanding tasks for operators. Large-scale plants such as APR1400 require operators to manage over 80 abnormal operating procedures, each covering multiple event types. Compounding this burden, abnormal operation is the most frequent operating regime and occurs in diverse, unpredictable forms, making consistent response inherently difficult.
These challenges are further amplified in the SMR context, where a single control room manages multiple modules with as few as three operators. This reduced staffing is compounded by the monitoring complexity that passive safety systems introduce, making high-level automation essential to ensure both safety and operational reliability.
This study proposes a novel framework for abnormal operation automation, called resilience operation for SMRs, developed through four sequential stages. First, Abstraction Hierarchy (AH) modeling decomposes a reference SMR, named integral Pressurized Water Reactor (iPWR) developed by the Tecnatom, into five functional-physical levels with quantitative formulations for each node. Second, a Hierarchical Multi-Agent Reinforcement Learning (H-MARL) architecture is designed by mapping each AH node to an independent RL agent in a manager-worker hierarchy. Third, a parallel simulation platform is constructed using multiple iPWR simulator instances to enable scalable training. Fourth, training is conducted with randomized fault parameters across selected abnormal scenarios to promote robust generalization. The result shows that the suggested framework successfully controls abnormal situations without any operator’s intervention.
← Check another abstractResponding to abnormal conditions is one of the most demanding tasks for operators. Large-scale plants such as APR1400 require operators to manage over 80 abnormal operating procedures, each covering multiple event types. Compounding this burden, abnormal operation is the most frequent operating regime and occurs in diverse, unpredictable forms, making consistent response inherently difficult.
These challenges are further amplified in the SMR context, where a single control room manages multiple modules with as few as three operators. This reduced staffing is compounded by the monitoring complexity that passive safety systems introduce, making high-level automation essential to ensure both safety and operational reliability.
This study proposes a novel framework for abnormal operation automation, called resilience operation for SMRs, developed through four sequential stages. First, Abstraction Hierarchy (AH) modeling decomposes a reference SMR, named integral Pressurized Water Reactor (iPWR) developed by the Tecnatom, into five functional-physical levels with quantitative formulations for each node. Second, a Hierarchical Multi-Agent Reinforcement Learning (H-MARL) architecture is designed by mapping each AH node to an independent RL agent in a manager-worker hierarchy. Third, a parallel simulation platform is constructed using multiple iPWR simulator instances to enable scalable training. Fourth, training is conducted with randomized fault parameters across selected abnormal scenarios to promote robust generalization. The result shows that the suggested framework successfully controls abnormal situations without any operator’s intervention.