Machine learning for reactor power monitoring with limited labeled data - research findings on classification with label-starved multisensor transfer learning

Aug 3, 2026

Upload by Miles

Real-time reactor power monitoring underpins safety, security, operations, and maintenance, yet machine-learning performance remains underexplored when labels are scarce. This work evaluates the feasibility of classifying nuclear reactor power levels using multisource data with limited labeled examples. Low-resolution multisensors were deployed at four facilities (two large research reactors and two TRIGA units). In transfer-learning experiments, one reactor dataset provided the source while another served as the target. Although supervised models achieved strong classification, self-learning and transfer learning did not improve target accuracy.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages10

Reading Time25 min

AccessFree to download

Introduction

Motivation and applications of reactor power monitoring

Machine learning approaches and related work

Security contexts and prior sensing methods

UC Berkeley UC Berkeley Previously Published Works Title Machine learning for reactor power monitoring with limited labeled data Permalink [https://escholarship.org/uc/item/295787zv](https://escholarship.org/uc/item/295787zv) Authors Stewart, CL Goldblum, BL Abbott, RGet al. Publication Date 2025-04-01 DOI 10.1016/j.nima.2025.170285 Copyright Information This work is made available under the terms of a Creative Commons Attribution License, available at [https://creativecommons.org/licenses/by/4.0/](https://creativecommons.org/licenses/by/4.0/) Peer reviewed [eScholarship.org](eScholarship.org) Powered by the California Digital Library University of California Nuclear Instruments and Methods in Physics Research A 1073 (2025) 170285 | Full Length Article
Machine learning for reactor power monitoring with limited labeled data C.L. Stewart a, B.L. Goldblum a,b ,∗, R.G. Abbott c, L. Appleby c, B.J. Borghetti d, V. Hollingsheada, J.H. Whetzel c
a Department of Nuclear Engineering, University of California, Berkeley, California 94720, USA b Nuclear Science Division, Lawrence Berkeley National Laboratory, Berkeley, CA, 94720, USAc Sandia National Laboratories, Albuquerque, NM, 87123, USA
d Air Force Institute of Technology, Department of Engineering Physics, Wright-Patterson Air Force Base, OH 45433, USA | | | | | --- | --- | --- | --- | | A R T I C L E I N F O | | A B S T R A C T | | | Keywords:
Classification
Nuclear reactor power monitoring Transfer learning
Self-learning
Multisensor arrays | | Real-time reactor power monitoring is critical for a variety of nuclear applications, spanning safety, security, operations, and maintenance. While machine learning methods have shown promise in monitoring reactor power levels, there is limited research on their efficacy in label-starved environments. The goal of this work is to assess the feasibility of classifying nuclear reactor power level using multisource data in scenarios with limited labels. Data were collected using low-resolution multisensors at four nuclear reactor facilities: two large research reactors and two TRIGA reactors. Within each pair, one reactor dataset served as the source and the other as the target in a transfer learning paradigm. Twenty-three supervised models were trained on labeled sequences of magnetic field and acceleration data from each of the target sites. Self-learning and transfer learning methods were applied to the top performing models to assess their classification performance with increasing amounts of labeled data. While reactor power level classification was achieved with a Matthews Correlation Coefficient of up to 0.739 ± 0.003 and 0.622 ± 0.009 with only 400 sequences per power state for the large research reactor and TRIGA target sites, respectively, self-learning and transfer learning leveraging source site data did not improve target classification performance. These findings suggest that alternative methods, such as higher sensitivity sensors, digital twins, or the use of physics-informed models, are required to enable high-performance classification in machine learning approaches to reactor monitoring with a dearth of target ground truth. | | 1. Introduction Real-time monitoring of the power level at which a nuclear reactor operates is crucial for both industrial and security applications. In industrial nuclear programs, reactor power monitoring can be used to ensure safe operations, optimize reactor performance, and mitigate against nuclear accidents [1]. For security applications, reactor power level history can be used in conjunction with operational parameters to evaluate plutonium production in the reactor core [2]. Traditionally, neutron detectors are placed within the core, just above the source, or outside the reactor vessel to monitor reactor power levels. More recently, machine learning techniques have emerged that enable classification of reactor power level with

FAQs

为什么在核反应堆功率监测中需要研究“有限标注数据”的机器学习方法？

实时功率监测对安全、安保、运行与维护至关重要，但在标注稀缺场景下，现有研究对机器学习方法的有效性了解不足。本文针对这种“label-starved”环境评估分类可行性。

实验如何收集数据并构建迁移学习设置？

研究在四个核反应堆设施部署低分辨率多传感器，包含两个大型研究反应堆和两个TRIGA。每个成对设施中，一处作为源数据，另一处作为目标数据，形成迁移学习范式。

自学习与迁移学习是否提升了目标站点的分类表现？

在测试中，虽然反应堆功率水平分类达到了较高的指标，但利用源站数据的自学习与迁移学习并未提升目标站点的分类性能。

Aug 3, 2026

Upload by Miles