A Machine Learning Method for the Optimization Design of Laser Pulse in Fast Ignition Simulations

Abstract

High energy gain is essential for the energy production via laser fusion. In this paper, an efficient method combining the hydrodynamic simulations and the machine learning algorithms is proposed to optimize the laser pulse for fast ignition simulations. An analytical model between the energy gain and compressed plasma parameters is derived as the evaluate function for the optimizations. An implosion with a fusion gain more than 100 is achieved with a total laser energy about 730 kJ in the spherical fast ignition scheme or 300 kJ in the double-cone ignition (DCI) scheme in one-dimensional simulations. The implosion data generated during the course of optimization is found to be suitable for the training of a deep neural network (DNN) surrogate model. In the future, this DNN surrogate model could be transfer learned with experimental feedback and optimize the laser pulse with a higher accuracy.