Denoising application for electron spectrometer in laser-driven ion acceleration using a Simulation-supervised Learning based CDAE

Miyatake, T.; Shiokawa, Keiichiro; Sakaki, H.; Dover, N. P.; Nishiuchi, Mamiko; Lowe, Hazel; Kondo, Kotaro; Kon, Akira; Kobayashi, Masaki; Kondo, Kiminori; Watanabe, Yukinobu

doi:10.1016/j.nima.2021.165227

Cited by 2 publications

(2 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…因此, 有必要寻找新的研究参量映射规律的手段. 近几年内, 依赖于数据学习的神经网络算法 [25,26] 被逐渐应用在科研领域, 例如托卡马克装置中磁场的控制优化 [27] 、惯性约束聚变实验结果对实验参数的敏感度研究 [28] 、电子加速能谱的自动识别与分类 [29] 、高辐射环境下测得的电子能谱图的去噪分析 [30] 、预测粒子加速器内粒子束纵向相空间演化趋势 [31] 等. 在众多类型的神经网络算法中, 卷积神经网络(convolutional neural network, CNN)主要面向二维矩阵数据的处理, 在图像处理和识别等领域应用广泛 [30] .…”

unclassified

“…在众多类型的神经网络算法中, 卷积神经网络(convolutional neural network, CNN)主要面向二维矩阵数据的处理, 在图像处理和识别等领域应用广泛 [30] . 自组织映射神经网络(self-organizing map, SOM)的训练过程为无监督的竞争学习, 适用于高维数据的聚类分析 [29] . 全连接构型人工神经网络(fully connected neural network, FCNN)可以建立起数个变量关于单一映射量的连续模型, 适用于数据回归类问题的建模分析.…”

unclassified

See 1 more Smart Citation

Modeling of ion accelerated by borehole radiation pressure based on deep learning

Zhang,

Wang,

et al. 2023

Acta Phys. Sin.

View full text Add to dashboard Cite

Laser-driven ion acceleration has potential applications in high energy density matter, ion beam-driven fast ignition, beam target neutron source and warm dense matter heating and etc. Ultrashort relativistic lasers interacting with solid targets can generate ion beams with energies up to several hundreds of MeV, and the quality of the ion beams strongly depends on the interaction parameters of the laser and the targets. Developments in deep learning can provide new methods in the analysis of relationship between parameters in physics systems, which can significantly reduce the computational and experimental cost. In this paper, a continuous mapping model of ion peak and cutoff energies is developed based on a fully connected neural network(FCNN). In the model, the dataset is composed of nearly 400 sets of particle simulations of laser-driven solid targets, and the input parameters are laser intensity, target density, target thickness and ion mass. The model obtains the parameter analysis results in a large range of values with sparser parameter taking values, which greatly reduces the computational effort of sweeping the parameters in a large range of multi-dimensional parameters. Based on the results of this model mapping, the correction formula for the ion peak energy over ion mass is obtained. Furthermore, the ratio of ion cutoff energy and peak energy of each set of particle simulation is calculated. Repeating the same training process of ion peak energy and cutoff energy, the continuous mapping model of energy ratio is developed. According to the energy ratio model mapping results, the quantitative description of the relationship between ion cutoff energy and peak energy is realized, and the fitting formula for the cutoff energy of the Hole-Boring Radiation Pressure Acceleration (HB-RPA) mechanism is obtained, which can provide an important reference for the laser-driven ion acceleration experiments design.

show abstract