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

To achieve high-precision and high-efficiency laser pulse shaping in the front-end system of high-power laser facilities, this paper proposes a deep learning model aimed at calibrating the initial operational status of the pulse shaping closed-loop control system. The model swiftly establishes a nonlinear mapping between the designed optical waveforms and the shaping electrical signals generated by the arbitrary waveform generator (AWG). The proposed model employs a U-shaped structure integrated with residual connections as its core network. An attention mechanism comprising the Kolmogorov-Arnold Network (KAN) and the temporal convolution network (TCN) between the model's input and output layers. The dataset constructed based on the pulse waveforms collected by the front-end system of the SG-II facility is used to train and test the model. The RMSEs between the predicted AWG waveforms and the targets are less than 5%, as well as the input complex waveforms and the waveforms with a high contrast of 100:1. The RMSEs for waveforms of different shapes and contrasts are less than 3%. The method can rapidly invert AWG shaping electrical signals by optical waveforms with different shapes, pulse widths and contrasts. Based on the model, the output optical waveforms closely approximate the design target in the initial iteration of the closed-loop control system, which will improve accuracy and efficiency of the pulse shaping system.