Machine-learning guided optimization of laser pulses for direct-drive implosions

Fuyuan Wu,Qi Zhang,Xiaohu Yang,Hao Liu,Xiaohui Yuan,Jie Zhang,Jian Zheng,Zhengdong Liu,Jiayong Zhong,Zhe Zhang,Yutong Li,Yanyun Ma

doi:10.1017/hpl.2022.4

Abstract

Abstract The optimization of laser pulse shapes is of great importance and a major challenge for laser direct-drive implosions. In this paper, we propose an efficient intelligent method to perform laser pulse optimization via hydrodynamic simulations guided by the genetic algorithm and random forest algorithm. Compared to manual optimizations, the machine-learning guided method is able to efficiently improve the areal density by a factor of 63% and reduce the in-flight-aspect ratio by a factor of 30% at the same time. A relationship between the maximum areal density and ion temperature is also achieved by the analysis of the big simulation dataset. This design method has been successfully demonstrated by the 2021 summer double-cone ignition experiments conducted at the SG-II upgrade laser facility and has great prospects for the design of other inertial fusion experiments.

Highlights

Laser-driven fusion [1,2,3] is a promising approach for the realization of controlled fusion energy
We propose a machine learning method in the combination of genetic algorithm and random forest algorithm to guide hydrodynamic simulations for the design of the drive laser pulse in the double-cone ignition (DCI) scheme
3.1 One-dimensional optimization guided by the genetic algorithm Figure 2 shows the target implosion trajectory and the laser pulse shape obtained by manual optimization and machine learning optimization

Summary

Introduction

Laser-driven fusion [1,2,3] is a promising approach for the realization of controlled fusion energy. Many laser-driven fusion schemes have been proposed in past decades, including the indirect-drive central ignition [4,5,6], the direct-drive central ignition [7,8,9], the hybrid-drive ignition. This peer-reviewed article has been accepted for publication but not yet copyedited or typeset, and so may be subject to change during the production process. In the DCI scheme, the conventional central ignition process is replaced by four progressive processes: quasi-isentropic compression, acceleration, collisional preheating and magnetic field assisted fast ignition. The need of fast ignition laser energy can be relaxed to an affordable level of the current PW laser technology with the head-on collisional preheating and the magnetic field collimated fast electrons

Methods

Results

Conclusion