Fast Predicting the Complex Nonlinear Dynamics of Mode‐Locked Fiber Laser by a Recurrent Neural Network with Prior Information Feeding

As an imperative method of investigating the internal mechanism of femtosecond lasers, traditional femtosecond laser modeling relies on the split‐step Fourier method (SSFM) to iteratively resolve the nonlinear Schrödinger equation suffering from the large computation complexity. To realize inverse design and optimization of femtosecond lasers, numerous simulations of mode‐locked fiber lasers with different cavity settings are required, further highlighting the time‐consuming problem induced by the large computation complexity. Here, a recurrent neural network is proposed to realize fast and accurate femtosecond mode‐locked fiber laser modeling. The generalization over different cavity settings is achieved via the proposed prior information feeding method. With the acceleration of GPU, the mean time of the artificial intelligence (AI) model inferring 500 roundtrips is less than 0.1 s, which is ≈146 times faster than the SSFM running on a CPU. The proposed AI‐enabled method is promising to become a standard approach to femtosecond laser modeling. A recurrent neural network is used to realize fast and accurate femtosecond mode‐locked fiber lasers modeling. In particular, to achieve generalization over various cavity settings, a prior information feeding method is proposed by which the prior cavity settings can be appropriately fed to artificial intelligence (AI). On an identical hardware platform, AI is 6 times faster than the traditional method.