Precise hohlraum x-ray flux predictions at the 100 kJ-level laser facility with deep learning

Precise prediction of the hohlraum x-ray drive is essential for target design optimization in indirect-drive inertial confinement fusion. Conventional simulations, however, are often hampered by the hohlraum drive deficit, which limits their predictive power for new experimental configurations. This study focuses on the two-shock implosions of the 100 kJ-level laser facility, aiming to infer the radiation source constrained by multiple diagnostics. A joint inversion method for the radiation source and M-band x-ray power multipliers was developed, leading to the construction of a surrogate model that maps experimental design parameters to radiation source waveforms. Based on this model, a deep learning inference framework, PRISM, was established to achieve precise radiation source prediction. The model exhibits excellent predictive accuracy, and the bang-time prediction error in pre-shot simulations is reduced from about 500 ps to within 200 ps. Its extrapolation capability was validated using transmission grating spectrometer data, demonstrating good agreement between simulated and measured spectra. Shapley Additive Explanations revealed the relative contributions of experimental design parameters to each multiplier. Finally, we used residual-based analysis to quantify engineering accuracy. The analysis indicates improved engineering control of the 100 kJ-level laser facility in 2025, particularly in laser output stability, although pulse synchronization remain areas for refinement.