Status calibration of a pulse shaping system for the high power laser facility based on deep learning

Area of Science:

• Optics and Photonics
• Machine Learning Applications
• High-Energy Physics

Background:

• High-precision laser pulse shaping is critical for high-power laser facilities.
• Current closed-loop control systems require efficient calibration methods.

Purpose of the Study:

• To develop a deep learning model for calibrating the initial operational status of laser pulse shaping systems.
• To establish a nonlinear mapping between optical waveforms and arbitrary waveform generator (AWG) signals.

Main Methods:

• A U-shaped deep learning network with residual connections was employed.
• An attention mechanism integrating Kolmogorov-Arnold Network (KAN) and Temporal Convolutional Network (TCN) was utilized.
• The model was trained and tested using data from the SG-II facility's front-end system.

Main Results:

• Root-mean-square errors (RMSEs) between predicted and target AWG waveforms were below 5%.
• RMSEs for complex waveforms with high contrast (100:1) were less than 5%.
• RMSEs for waveforms of varying shapes and contrasts were below 3%.

Conclusions:

• The proposed deep learning method enables rapid inversion of AWG signals for diverse optical waveforms.
• The model significantly improves the accuracy and efficiency of laser pulse shaping in the initial control loop iteration.