Physics-Informed Neural Network for Fiber Amplifiers

Fiber amplifiers play a crucial role in high-power laser sources and are widely used in the field of precision machining. To model fiber amplifiers and understand underlying mathematical and physical mechanisms, the accurate solutions to the dynamic rate equations (DREs) and power transmission equations (PTEs) are required. Furthermore, the system parameter monitoring and estimation of fiber amplifiers in optical fiber communication systems is an urgent problem. However, the conventional numerical methods with inferior efficiency and increased computation complexity are not enough to predict the pulse amplification and infer the unknown parameters in the systems. This study proposes a method based on physics-informed neural network (PINN) to overcome these challenges by learning prior physical knowledge during network optimization. System parameter estimation can be performed by solving inverse problems which is essential for system detection and monitoring constantly. Moreover, the reasons for the differences generated by different methods are analyzed and discussed. This is the first combination of fiber amplifiers and PINN. Our work can bring new insights into the design, use and monitoring of fiber amplifiers.