Pulse characteristics prediction and optimization of passive mode-locked lasers based on Kolmogorov-Arnold network

Abstract

Passive mode-locked fiber lasers (PMLFLs) have great advantages in generating ultrashort pulses due to their unique parameter tunability. The pulse duration, energy and peak power can be controlled by parameter tuning. Here, Kolmogorov-Arnold network (KAN) algorithm is used to predict the pulse characteristics of PMLFLs, and several optimization algorithms are used to explore the optimal output and parameters' combination. Firstly, we obtain the dataset between different parameters and pulse characteristics by solving the PMLFL model. Then, KAN algorithm is used to interpret and predict the data set, and the function between input parameters and ultrashort pulse is obtained. We compare KAN to a series of traditional machine learning (ML) models, and KAN is more efficiently than traditional ML models. Then we use ensemble learning to design a multi-mechanism prediction model to enhance the robustness of the prediction model. Finally, we use particle swarm optimization (PSO) to optimize the KAN model, and compare the optimization results with other intelligent algorithms for the traditional ML model. The KAN prediction model and explicit function are helpful to optimize the parameter setting of optical communication system, laser processing system and laser medical system, and improve the communication quality, processing quality and treatment effect.