Soliton Molecule Dynamics Evolution Prediction Based on <i>LSTM</i> Neural Networks
Jiangyong He, Caiyun Li, Pan Wang, Congcong Liu, Yange Liu, Bo Liu, Dengke Xing, Zhi Wang
Abstract
In this article, we design a long short-term memory ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">LSTM</i> ) scheme, combined with dense networks, to realize soliton dynamics prediction in passively mode-locked fiber lasers. Based on the particle characteristic of soliton interaction, we propose to use the separation and relative phase between solitons as characteristic parameters to model and predict the dynamics. The network predicted soliton collision and soliton molecule dynamics accurately. This scheme of precoding physical information with subsequent dynamics prediction not only introduces new prospects for the laser self-optimization algorithm, but also brings new insights for the modeling of nonlinear systems and description of soliton interactions.