Intelligent optimization and neural network surrogate for DCI colliding plasmas

High-density and high-areal-density plasmas are the fundamental prerequisites for fusion ignition in inertial confinement fusion. To generate such plasma conditions via the double-cone ignition (DCI) scheme, a laser facility delivering 160 kJ nanosecond laser pulses is currently under construction in Shanghai. In this work, we propose a compression-stage optimization method for colliding plasmas that integrates two-dimensional radiation hydrodynamics simulations with a genetic algorithm (GA). We show that an isochoric colliding plasma with a peak density of 409 g/cm3 and an areal density of 1.63 g/cm2 can be achieved. We also identify an optimal implosion velocity range of about 251–283 km/s for the plasma jet-collision process for this 160 kJ facility, which can be described by a physically interpretable fit curve. To enable large-scale, high-fidelity predictions of colliding plasma behavior, we develop a neural network ensemble surrogate model trained on over 20 000 simulation datasets derived from the aforementioned GA optimization. This surrogate model takes laser pulse shapes and target geometric parameters as inputs to predict the key plasma parameters, including peak density, areal density, and implosion velocity. The model achieves predictions with an average relative error below 3% and accelerates the optimization process by more than two orders of magnitude compared to direct MULTI-2D simulations. This work establishes a solid foundation for the rapid design and parameter optimization of future DCI experiments.