人工神经网络在HL-2A装置汤姆逊散射数据处理中的应用

刘春华; 侯智培; 王瑜琴; 冯震; 夏凡; 黄渊

doi:10.11884/HPLPB201931.180206

人工神经网络在HL-2A装置汤姆逊散射数据处理中的应用

doi: 10.11884/HPLPB201931.180206

核工业西南物理研究院聚变科学所, 成都 610041

基金项目:

国家自然科学基金面上项目 11775072

国家自然科学基金面上项目 11875022

国家重点研发计划项目 2018YFE0301102

详细信息

作者简介:
刘春华(1981－), 女，博士，副研究员，从事高温等离子体激光诊断及物理实验研究；liuchunhua@swip.ac.cn

中图分类号: O536
计量
- 文章访问数: 1153
- HTML全文浏览量: 330
- PDF下载量: 95
- 被引次数: 0
出版历程
- 收稿日期: 2018-07-28
- 修回日期: 2019-01-28
- 刊出日期: 2019-02-15

Artificial neural network approach applied to data processing of Thomson scattering on HL-2A

Center for Fusion Science, Southwestern Institute of Physics, Chengdu 610041, China

摘要

摘要: 人工神经网络是一种强大的非线性数据分析算法，其中的感知器神经网络第一次被用于处理HL-2A装置上汤姆逊散射系统的电子温度数据。采用输入层、隐藏层和输出层等三层神经网络结构，输入层为标定数据或测量数据，隐藏层使用sigmoid函数作为传递函数，输出层为电子温度值。从数据处理结果可以看出，该计算方法与传统的χ²最小值方法计算的结果吻合，能够得到可靠的电子温度数据。而且由于计算温度时采用矩阵计算，计算速度比使用χ²最小值法提高20倍以上，为将来利用汤姆逊散射测量的电子温度数据实现等离子体剖面实时反馈控制提供了可能。
- 汤姆逊散射 /
- 神经网络 /
- 数据处理 /
- 电子温度 /
- HL-2A托卡马克
Abstract: Artificial neural network(NN) as a powerful nonlinear data processing method, has been successfully applied to process electron temperature for Thomson scattering system on HL-2A. A type of perception is chosen. The NN has three layers: input layer, hidden layer, and output layer. Calibration data or measured data are the input layer, hidden layer uses sigmoid function as transfer function, and output layer is electron temperature. The calculation results fit well with that results calculated by traditional minimization chi-square method. And its calculation speed, about 1 ms per shot and per spatial point, is about 20 times faster than the minimization chi-square method. Therefore, it is possible for real time feed-back control plasma discharge by electron temperature measured by Thomson scattering on HL-2M, ITER or CFTER.
- Thomson scattering /
- neural network /
- data processing /
- electron temperature /
- HL-2A Tokamak

HTML全文

图 1 散射角为90°时不同电子温度下的散射谱强度与波长之间的关系

Figure 1. Thomson scattering spectrum intensity distributions in different electron temperature in the case that scattered angle equals to 90°

下载: 全尺寸图片幻灯片

图 2 标定得到的HL-2A激光散射系统芯部空间点对应的多色仪各通道的信号强度值与电子温度之间的关系

Figure 2. Calibration results of a polychromator in plasma core region of HL-2A

下载: 全尺寸图片幻灯片

图 3 应用到计算激光散射信号获得电子温度的神经网络结构

Figure 3. Neural network structure applied to Thomson scattering system for electron temperature calculation

下载: 全尺寸图片幻灯片

图 4 用于计算电子温度时训练神经网络的框图结构，及神经网络验证和计算电子温度过程

Figure 4. Flow block of training neural network procedure used for electron temperature calculation, and verification procedure and electron temperature calculating procedure

下载: 全尺寸图片幻灯片

图 5 不同隐藏层节点数和训练次数的比较结果

Figure 5. Comparison of errors calculated with different hidden nodes and training times

下载: 全尺寸图片幻灯片

图 6 同样训练次数情况下，不同隐藏层节点数时训练输出与理论值的比较情况

Figure 6. Comparison of training results of different hidden nodes with theoretical values under same training times

下载: 全尺寸图片幻灯片

图 7 分别利用神经网络方法(NN)、χ²最小值法(min χ²)和电子回旋辐射法(ECE)计算得到的电子温度随时间的变化

Figure 7. Results of T_e calculated by neural network (NN), minimization chi-square method (min χ²) and electron cyclotron emission (ECE)

下载: 全尺寸图片幻灯片

图 8 训练次数为1×10⁵和隐藏层节点数为54时的误差情况

Figure 8. Error results with 1×10⁵ training times and 54 nodes

下载: 全尺寸图片幻灯片

参考文献(27)

[1]	刘春华, 黄渊, 冯震, 等. HL-2A装置YAG激光汤姆逊散射测量电子温度的初步结果[J]. 强激光与粒子束, 2008, 20(7): 1119-1124. http://www.hplpb.com.cn/article/id/3572 Liu Chunhua, Huang Yuan, Feng Zhen, et al. Primary results of measuring plasma electron temperature with YAG laser Thomson scattering system on the HL-2A Tokamak. High Power Laser and Particle Beams, 2008, 20(7): 1119-1124 http://www.hplpb.com.cn/article/id/3572
[2]	慈佳祥, 黄渊, 唐昌建, 等. HL-2A装置Nd: YAG激光汤姆逊散射多道测量系统[J]. 强激光与粒子束, 2009, 21(3): 359-364. http://www.hplpb.com.cn/article/id/3931 Ci Jiaxiang, Huang Yuan, Tang Changjian, et al. Measurement system of multipoint Thomson scattering on HL-2A Tokamak. High Power Laser and Particle Beams, 2009, 21(3): 359-364 http://www.hplpb.com.cn/article/id/3931
[3]	Liu C H, Huang Y, Wang Y Q, et al. The progress in development of edge tangential Thomson scattering system on HL-2A[J]. Review of Scientific Instruments, 2016, 87: 11E555. doi: 10.1063/1.4963200
[4]	Liu C H, Wang Y Q, Feng Z, et al. Improvement in data processing of Thomson scattering diagnostic on HL-2A tokamak[J]. Journal of Instrumentation, 2015, 10: C12026. doi: 10.1088/1748-0221/10/12/C12026
[5]	Liu Chunhua, Huang Yuan, Feng Zhen, et al. Electron density measurement by Raman calibration of the HuanLiuqi-2A Thomson scattering system[J]. Plasma and Fusion Research: Regular Articles, 2014, 9: 1402042. doi: 10.1585/pfr.9.1402042
[6]	Rietman E A, Beachy M. A study on failure prediction in a plasma reactor[J]. IEEE Trans Semiconductor Manufacturing, 1998, 11(4): 670-680. doi: 10.1109/66.728564
[7]	Matthew S P. Interpretation of machine-learning-based disruption models for plasma control[J]. Plasma Physics and Controlled Fusion, 2017, 59: 085001. doi: 10.1088/1361-6587/aa72a3
[8]	Cannas B, Cau F, Fanni A, et al. Automatic disruption classification at JET: comparison of different pattern recognition techniques[J]. Nuclear Fusion, 2006, 46(7): 699-708. doi: 10.1088/0029-5515/46/7/002
[9]	Cannas B, Vries P C, Fanni A, et al. Automatic disruption classification in JET with the ITER-like wall[J]. Plasma Physics and Controlled Fusion, 2015, 57: 125003. doi: 10.1088/0741-3335/57/12/125003
[10]	Cannas B, Fanni A, Murari A, et al. Automatic disruption classification based on manifold learning for real-time applications on JET[J]. Nuclear Fusion, 2013, 53: 093023. doi: 10.1088/0029-5515/53/9/093023
[11]	Cannas B, Fanni A, Pautasso G, et al. An adaptive real-time disruption predictor for ASDEX Upgrade[J]. Nuclear Fusion, 2010, 50: 075004. doi: 10.1088/0029-5515/50/7/075004
[12]	Morabito F C, Versaci M, Pautasso G. Fuzzy-neural approaches to the prediction of disruptions in ASDEX Upgrade[J]. Nuclear Fusion, 2001, 41(11): 1715-1724. doi: 10.1088/0029-5515/41/11/321
[13]	Pautasso G. Vries P C, Humphreys D, et al. The ITER disruption mitigation trigger: developing its preliminary design[J]. Nuclear Fusion, 2018, 58: 036011. doi: 10.1088/1741-4326/aaa137
[14]	Wang S Y, Chen Z Y, Huang D W, et al. Prediction of density limit disruptions on the J-TEXT tokamak[J]. Plasma Physics and Controlled Fusion, 2016, 58: 055014. doi: 10.1088/0741-3335/58/5/055014
[15]	Sengupta A, Ranjan P. Forecasting disruptions in the ADITYA tokamak using neural networks[J]. Nuclear Fusion, 2000, 40(12): 1993. doi: 10.1088/0029-5515/40/12/304
[16]	Bishop C M, Roach C M, Hellermann M G. Automatic analysis of JET charge exchange spectra using neural networks[J]. Plasma Physics and Controlled Fusion, 1993, 35: 765-773. doi: 10.1088/0741-3335/35/6/010
[17]	Clayton D J, Tritz K, Stutman D, et al. Electron temperature profile reconstructions from multi-energy SXR measurements using neural networks[J]. Plasma Physics and Controlled Fusion, 2013, 55: 095015. doi: 10.1088/0741-3335/55/9/095015
[18]	Aguiam D E, Silva A, Bobkov V, et al. Implementation of the new multichannel X-mode edge density profile reflectometer for the ICRF antenna on ASDEX Upgrade[J]. Review of Scientific Instruments, 2016, 87: 11E722. doi: 10.1063/1.4961558
[19]	Vega J, Murari A, Gonzalez S. A universal support vector machines based method for automatic event location in waveforms and video-movies: Applications to massive nuclear fusion databases[J]. Review of Scientific Instruments, 2010, 81: 10E118. doi: 10.1063/1.3478689
[20]	Farias G, Dormido-Canto S, Vega J, et al. Image classification by using a reduced set of features in the TJ-Ⅱ Thomson scattering diagnostic[J]. Fusion Engineering and Design, 2018, 129: 99-103. doi: 10.1016/j.fusengdes.2018.02.081
[21]	Fujii K, Yamada I, Hasuo M. Machine learning of noise in LHD Thomson scattering system[J]. Fusion Science and Technology, 2018, 74(1/2): 57-64.
[22]	Lee S H, Lee J H, Yamada I, et al. Development of a neural network technique for KSTAR Thomson scattering diagnostics[J]. Review of Scientific Instruments, 2016, 87: 11E533. doi: 10.1063/1.4961079
[23]	Greenfield C M, Campbell G L, Carlstrom T N, et al. Real-time digital control, data acquisition, and analysis system for the DⅢ-D multi-pulse Thomson scattering diagnostic[J]. Rev Sci Instr, 1990, 61(10): 3286-3288. doi: 10.1063/1.1141615
[24]	Narihara K, Yamauchi K, Minami T, et al. Construction of a 100-Hz-repetition-rate 28-channel Thomson scattering system for the JIPPT-ⅡU Tokamak[J]. Japanese Journal of Applied Physics, 1996, 35: 266-273. doi: 10.1143/JJAP.35.266
[25]	姚可, 黄渊, 冯震, 等. 最大似然法在HL-2A汤姆逊散射中的初步应用[J]. 核聚变与等离子体物理, 2010, 30(3): 203-207. https://www.cnki.com.cn/Article/CJFDTOTAL-HJBY201003003.htm Yao Ke, Huang Yuan, Feng Zhen, et al. Preliminary application of maximum likelihood method in HL-2A Thomson scattering system. Nuclear Fusion and Plasma Physics, 2010, 30(3): 203-207 https://www.cnki.com.cn/Article/CJFDTOTAL-HJBY201003003.htm
[26]	Fischer R, Dinklage A, Pasch E. Bayesian modelling of fusion diagnostics[J]. Plasma Physics and Controlled Fusion, 2003, 45(7): 1095-1111.
[27]	Beurskens M N A, Barth C J, Chu C C, et al. Error analysis of Rijnhuizen Tokamak project Thomson scattering data[J]. Review of Scientific Instruments, 1999, 70: 1999-2011.