目的 针对单迷道屏蔽结构中迷道角度连续变化对出口归一化剂量响应影响规律不明确的问题,研究角度参数与出口平均剂量响应之间的定量关系,为低剂量角度筛选和屏蔽结构优化提供依据。方法 基于蒙特卡罗N粒子程序（Monte Carlo N-Particle Transport Code,MCNP）建立单迷道光子输运模型,设置角度范围为-45°~45°、步长为0.1°,获得901组计算结果;定义随角度旋转的出口平行四边形感兴趣区域（Region of Interest,ROI）,提取出口平均剂量Dmean、相对误差和有效网格数量;在质量筛选后,以y=log10(Dmean)为响应变量,建立全局多项式、分段正余弦函数和机器学习预测模型。结果 筛选后获得407个最终拟合样本。出口平均剂量随角度变化呈现明显非线性和正负角度不对称特征,0°附近为低剂量区域。8阶全局多项式加权R2为0.836 300,交叉验证R2为0.828 613;分段正余弦模型整体非加权R2达到0.918 959,可描述-45°~0°区间的余弦式衰减规律和0°~45°区间的正弦式低剂量波动规律。Regularized XGBoost模型测试集R2为0.785 348,最终筛选样本全量R2为0.857 388,RepeatedKFold平均R2为0.835 008。结论 单迷道角度对出口归一化剂量响应具有显著影响。分段正余弦模型更适合描述正负角度区间的差异化响应,机器学习模型可用于候选角度的快速预测和低剂量角度初筛,但关键工况仍需通过MCNP复算进行安全校核。
Abstract
To clarify the effect of continuous maze-angle variation on the normalized outlet dose response in a single-maze shielding structure, the work aims to investigate the quantitative relationship between the angle parameter and the outlet mean dose response, so as to provide a basis for low-dose angle screening and shielding structure optimization. A single-maze photon transport model was established with the Monte Carlo N-Particle Transport Code (MCNP). The angle range was set from -45° to 45° with an interval of 0.1°, resulting in 901 calculation cases. A rotating parallelogram region of interest (ROI) was defined at the outlet to extract the outlet mean dose Dmean, relative error and number of effective mesh cells. After quality filtering, y=log10(Dmean) was taken as the response variable, and global polynomial fitting, piecewise sine-cosine fitting and machine learning prediction models were established. A total of 407 final fitting samples were obtained after filtering. The outlet mean dose showed obvious nonlinear variation and asymmetry between positive and negative angles, with a low-dose region near 0°. The weighted R2 of the eighth-order global polynomial was 0.836 300, and the cross-validation R2 was 0.828 613. The overall unweighted R2 of the piecewise sine-cosine model reached 0.918 959, indicating that it could describe the cosine-like attenuation in the -45° to 0° interval and the sine-like low-dose fluctuation in the 0° to 45° interval. The Regularized XGBoost model achieved a test R2 of 0.785 348, a full-sample R2 of 0.857 388 on the final filtered samples, and a mean RepeatedKFold R2 of 0.835 008. The single-maze angle has a significant effect on the normalized outlet dose response. The piecewise sine-cosine model is more suitable for describing the differentiated responses in the negative and positive angle intervals, and the machine learning model can be used for rapid prediction of candidate angles and preliminary screening of low-dose angles. However, key cases still need to be verified by MCNP recalculation for safety assessment.
关键词
单迷道 /
出口归一化剂量响应 /
蒙特卡罗N粒子程序（MCNP） /
分段正余弦拟合 /
正则化极端梯度提升（Regularized XGBoost） /
误差加权
Key words
single-maze /
normalized outlet dose response /
Monte Carlo N-Particle Transport Code (MCNP) /
piecewise sine-cosine fitting /
Regularized eXtreme Gradient Boosting (Regularized XGBoost) /
error weighting
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] International Atomic Energy Agency. Regulatory Control of the Safety of Ion Radiotherapy Facilities: IAEA-TECDOC-1891[R]. Vienna: International Atomic Energy Agency, 2019
[2] KULESZA J A, ADAMS T R, ARMSTRONG J C, et al.MCNP Code Version 6.3.1 Theory & User Manual: LA-UR-24-24602, Rev. 1[R]. Los Alamos, NM, USA: Los Alamos National Laboratory, 2024
[3] PARK H, PAGANETTI H, SCHUEMANN J, et al. Monte Carlo Methods for Device Simulations in Radiation Therapy[J]. Physics in Medicine & Biology, 2021, 66(18): 18TR01.
[4] SULIMAN I I, KHOUQEER G A, MAYHOUB F H.Photoneutrons and Gamma Capture Dose Rates at the Maze Entrance of Varian TrueBeam and Elekta Versa HD Medical Linear Accelerators[J]. Toxics, 2023, 11(1): 78.
[5] PIKES G, ROWSHANFARZAD P, DASS J, et al.Monte Carlo Simulations in the Optimisation of Linac Bunker Shielding[J]. Nuclear Engineering and Technology, 2025, 57(7): 103469.
[6] EGHDAM-ZAMIRI R, GHIASI H.Capture Γ-Ray Dose Equivalent at Double-Bend Maze Entrance; Monte Carlo Simulation and Analytical Methods and Measurements[J]. Frontiers in Biomedical Technologies, 2023, 10(1): 21-26.
[7] CHEN L F. Machine Learning-Assisted Optimization of Modular Neutron Shielding Based on Monte Carlo Simulations[EB/OL]. arXiv, 2025: arXiv:2504.17319. DOI: 10.48550/arXiv.2504.17319.
[8] ALGHAMDI A A A, ALJASSIR A Z, ALMUAYBID R O, et al. Photon Effective Dose Prediction Using XGBoost: A Machine Learning Approach for Radiological Protection[J]. Radiation Physics and Chemistry, 2025, 229: 112552.
[9] CHEN T Q, GUESTRIN C.XGBoost: A Scalable Tree Boosting System[C]//Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. San Francisco California USA. ACM, 2016: 785-794.
[10] 朱益楠, 林作康, 郁海燕, 等. 基于蒙特卡罗几何分裂减方差技巧的AB-BNCT治疗室屏蔽模拟分析[J]. 核技术, 2025, 48(1): 62-70.
ZHU Y N, LIN Z K, YU H Y, et al.Simulation Analysis of AB-BNCT Treatment Room Shielding Based on Monte Carlo Geometric Splitting Variance Reduction Technique[J]. Nuclear Techniques, 2025, 48(1): 62-70.
[11] 卢一冉, 吕焕文, 王雪晴. 角度相关的蒙特卡罗级联减方差方法研究[J]. 核技术, 2026, 49(3): 030602.
LU Y R, LYU H W, WANG X Q.Research on Angle-Dependent Monte Carlo Cascade Variance Reduction Method[J]. Nuclear Techniques, 2026, 49(3): 030602.
[12] YADAV P, RACHAMIN R, KONHEISER J, et al.Estimation of Weight Window Parameters Based on Recursive Monte Carlo Approach for Reactor Shielding Problems[J]. Annals of Nuclear Energy, 2024, 197: 110277.
[13] SATO T, HASHIMOTO S, DAMIÁN J I M, et al. Development of a Forward Monte Carlo Based Weight- Window Generator Using the History-Counter Function in PHITS[J]. Nuclear Instruments and Methods in Physics Research Section B: Beam Interactions with Materials and Atoms, 2024, 557: 165535.
[14] KULESZA J A, ADAMS T R, ARMSTRONG J C, et al.MCNP® Code Version 6.3.0 Theory & User Manual: LA-UR-22-30006, Rev. 1[R]. Los Alamos, NM, USA: Los Alamos National Laboratory, 2022.
[15] SOCIETY A N.Photon and Neutron Fluence-to-Dose Conversion Coefficients: ANSI/ANS-6.1.1-2020[S]. La Grange Park, IL, USA: American Nuclear Society, 2020
[16] PETOUSSI-HENSS N, BOLCH W E, ECKERMAN K F, et al.Conversion Coefficients for Radiological Protection Quantities for External Radiation Exposures[J]. Annals of the ICRP, 2010, 40(2/3/4/5): 1-257.
[17] BOZZATO D, FROESCHL R, OTHERS. Implementation of ICRP116 Fluence to Effective Dose Conversion Coefficients in a FLUKA User Routine:[R]. Geneva: CERN, 2024
[18] YEOM Y S, GRIFFIN K, HAN H, et al.Dose Conversion Coefficients for Neutron External Exposures with Five Postures: Walking, Sitting, Bending, Kneeling, and Squatting[J]. Radiation and Environmental Biophysics, 2021, 60(2): 317-328.
[19] PEDREGOSA F, VAROQUAUX G, GRAMFORT A, et al.Scikit-Learn: Machine Learning in Python[J]. Journal of Machine Learning Research, 2011, 12: 2825-2830.
[20] GEURTS P, ERNST D, WEHENKEL L.Extremely Randomized Trees[J]. Machine Learning, 2006, 63(1): 3-42.
[21] HUBER P J.Robust Estimation of a Location Parameter[J]. The Annals of Mathematical Statistics, 1964, 35(1): 73-101.
[22] ALGHAMDI A A A. Machine Learning for Predicting Neutron Effective Dose[J]. Applied Sciences, 2024, 14(13): 5740.
[23] CHEN L F.Deep Learning Optimization of Modular Neutron Beam Shutters via Monte Carlo Simulations[J]. Radiation Physics and Chemistry, 2026, 239: 113289.
基金
中核集团领创项目;上海市探索者项目（25TS1402500）;国家自然科学基金（U24B2025）;松山湖材料试验室开放课题基金（2023SLABFN09）
PDF(796 KB)
渝公网安备 50010702501716号 渝ICP备15012534号-2
