Prediction model of K2CsSb photocathode reflectivity based on LSTM
-
摘要: 针对目前K2CsSb光阴极制备过程中无法预判光阴极生长状态的问题，提出一种基于长短期记忆（LSTM）循环神经网络的K2CsSb光阴极反射率预测模型。一维原始反射率数据集经过清洗、筛选、序列化等预处理手段后重构为二维数据输入模型。为充分利用反射率数据在时序上高度相关的特性，采用双层LSTM网络提取特征，预测结果通过全连接层输出，以均方误差（MSE）作为模型预测效果的评判标准。实验结果表明，该模型的网络结构合理且在不同数据集下的表现良好，预测准确率可达99.21%。该模型可运用在K2CsSb光阴极的制作过程中，通过反射率预测值反馈调节工艺参数以趋近目标走势，对提高光阴极性能具有促进作用。Abstract: Aiming at the problem that the growth state of K2CsSb photocathode cannot be predicted in the current preparation process of K2CsSb photocathode, a prediction model of K2CsSb photocathode reflectivity based on long short-term memory (LSTM) recurrent neural network was proposed. The one-dimensional original reflectivity data set was reconstructed into a two-dimensional data input model after cleaning, screening, serialization and other preprocessing methods. In order to make full use of the highly correlated characteristics of reflectivity data in time series, this model used a double-layer LSTM network to extract features, the prediction results were output through the fully connected layer, and the mean square error (MSE) was used as the evaluation standard for the prediction effect of the model. The experimental results show that the network structure of the model is reasonable and performs well in different data sets, and the prediction accuracy rate can reach 99.21%. The proposed model can be used in the fabrication process of K2CsSb photocathode, and the process parameters can be adjusted by feedback of the reflectivity prediction value to approach the target trend, which can promote the performance of the photocathode.
-
Key words:
- bialkali photocathode /
- long short-term memory /
- deep learning /
- reflectance
-
图 7 序列长度与准确率和训练时间趋势图
Fig. 7 Trend chart of sequence length, accuracy and training time
图 8 LSTM网络不同层数下的损失值对比
Fig. 8 Comparison of loss values under different layers of LSTM network
图 9 不同数据集下模型预测效果对比
Fig. 9 Comparison of model prediction effects under different datasets
表 1 网络组成和参数量
Table 1 Composition of network structure and the number of parameters
网络层 输出维度 参数量 输入层 (None, 5, 5) 0 LSTM (None, 5, 10) 640 随机失活层 (None, 5, 10) 0 LSTM (None, 10) 840 随机失活层 (None, 10) 0 输出层 (None, 1) 11 
下载: 导出CSV
表 2 实验模型参数
Table 2 Parameters of experimental model
主要参数 参数值 损失函数 MSE 学习率 0.0003 优化器 Adam 序列长度 9 网络层数 2 迭代次数 100 批处理大小/条 32
下载: 导出CSV
表 3 序列长度对准确率和训练时间的影响
Table 3 Effect of sequence length on accuracy and training time
序列
长度每轮训练
时间/s准确率/% 序列
长度每轮训练
时间/s准确率/% 1 2.13 98.42 11 15.21 98.45 3 4.36 98.63 13 15.46 98.26 5 5.12 99.21 15 15.79 98.91 7 5.63 98.82 17 15.86 98.99 9 9.62 99.01 19 15.98 98.43
下载: 导出CSV
表 4 网络层数对准确率和训练时间的影响
Table 4 Effect of network layers on accuracy and training time
BP神经网络 LSTM网络 网络
层数每轮训练
时间/s准确率/% 网络
层数每轮训练
时间/s准确率/% 1 2.12 90.89 1 3.26 98.79 2 4.52 94.45 2 5.12 99.21 3 5.89 95.69 3 6.65 98.93
下载: 导出CSV
-
[1] HIROYUKI S. Review of photo-sensor R&D for future water Cherenkov detectors: Report of the 12th International Workshop on Next generation Nucleon Decay and Neutrino Detectors[R]. Zurich, Switzerland: NNN11, 2010. [2] 常本康. 大面积MCP-PMT K2CsSb光电阴极理论与测控技术研究[J]. 红外技术,2013,35(8):455-462.CHANG Benkang. Theory and control technology of large area MCP-PMT K2 CsSb photocathode[J]. Infrared Technology,2013,35(8):455-462. [3] SMEDLEY J, RAO T, WANG Erdong. K2CsSb cathode development[J]. AIP Conference Proceedings,2009,1149(1):1062-1066. [4] DING Z, GAOWEI M, SINSHEIMER J, et al. In-situ synchrotron X-ray characterization of K2CsSb photocathode grown by ternary co-evaporation[J]. Journal of Applied Physics,2017,121(5):055305. doi: 10.1063/1.4975113 [5] ZHANG Fan, LI Xiaoping, LI Xiaoshen. Development of preparation systems with K2 CsSb photocathodes and study on the preparation process[J]. Chinese Physics Letters,2019,36(2):022901. doi: 10.1088/0256-307X/36/2/022901 [6] GHOSH C, VARMA B P. Preparation and study of properties of a few alkali antimonide photocathodes[J]. Journal of Applied Physics,1978,49(8):4549-4553. doi: 10.1063/1.325465 [7] 谢华木, 王尔东. 作为加速器电子源的高量子效率K2CsSb光阴极制备工艺研究[J]. 真空,2017,54(1):63-66.XIE Huamu, WANG Erdong. Research on fabrication recipe of high quantum efficiency K2CsSb photocathode[J]. Vacuum,2017,54(1):63-66. [8] 刘如彪. 高性能光电阴极工艺监控技术研究[D]. 南京: 南京理工大学, 2008.LIU Rubiao. Research on monitor technique of technology about high performance photocathode[D]. Nanjing: Nanjing University of Science and Technology, 2008. [9] 李晓峰, 田金生, 钱芸生, 等. 多碱阴极光学反射率监控[J]. 红外技术,1997,19(4):39-42.LI Xiaofeng, TIAN Jinsheng, QIAN Yunsheng, et al. Reflectance monitoring on multialkali photocathode layer growth[J]. Infrared Technology,1997,19(4):39-42. [10] 孙建宁, 司曙光, 王兴超, 等. 一种利用反射率理论模型指导K2CsSb光电阴极的制备方法[J]. 红外技术,2017,39(12):1087-1091.SUN Jianning, SI Shuguang, WANG Xingchao, et al. Preparation method of K2CsSb photocathode using the reflectance theory model[J]. Infrared Technology,2017,39(12):1087-1091. [11] 陶兆民, 许汴生. 半透明Sb-K-Cs光电阴极的研制[J]. 中国科学技术大学学报,1966(1):90-92.TAO Zhaomin, XU Biansheng. Preparation of Translucent Sb-K-Cs Photocathode[J]. Journal of University of Science and Technology of China,1966(1):90-92. [12] JIN Muchun, CHANG Benkang, CHENG Hongchang, et al. Research on quantum efficiency of transmission-mode InGaAs photocathode[J]. Optik,2014,125(10):2395-2399. doi: 10.1016/j.ijleo.2013.10.086 [13] MOTTA D, SCHÖNERT S. Optical properties of bialkali photocathodes[J]. Nuclear Instruments and Methods in Physics Research Section A:Accelerators, Spectrometers, Detectors and Associated Equipment,2005,539(1/2):217-235. [14] 田远洋, 徐显涛, 彭安帮, 等. 训练数据量对LSTM网络学习性能影响分析[J]. 水文,2022,42(1):29-34. doi: 10.19797/j.cnki.1000-0852.20210256TIAN Yuanyang, XU Xiantao, PENG Anbang, et al. Effects of training data on the study performance of LSTM network[J]. Journal of China Hydrology,2022,42(1):29-34. doi: 10.19797/j.cnki.1000-0852.20210256 [15] HOCHREITER S, SCHMIDHUBER J. Long short-term memory[J]. Neural Computation,1997,9(8):1735-1780. doi: 10.1162/neco.1997.9.8.1735 [16] 葛靖, 刘子龙. 基于CNN和LSTM的睡眠呼吸暂停检测算法[J]. 电子科技,2021,34(2):21-26. doi: 10.16180/j.cnki.issn1007-7820.2021.02.004GE Jing, LIU Zilong. The algorithm based on CNN and LSTM for sleep apnea syndrome detection[J]. Electronic Science and Technology,2021,34(2):21-26. doi: 10.16180/j.cnki.issn1007-7820.2021.02.004 -
点击查看大图
计量
- 文章访问数: 110
- HTML全文浏览量: 52
- PDF下载量: 15
- 被引次数: 0
陕公网安备 61011302001501号 