Optimal Design of Gain-Flattened Raman Fiber Amplifiers Using a Hybrid Approach Combining Randomized Neural Networks and Differential Evolution Algorithm

Abstract

An efficient method based on a hybrid approach that combines extreme learning machine (ELM) technique and differential evolution (DE) algorithm is proposed to optimize the multipumped Raman fiber amplifier (RFA). The proposed method takes advantage of the fast learning speed and high generalization of the ELM as well as the strong global search capability of DE. From a novel perspective, we utilize ELM as a powerful learning tool to construct the nonlinear mapping between the pump parameters and gains of RFA. Instead of time-consuming integration of Raman coupled equations, the gains can be directly and accurately determined by the ELM model. To obtain a flat gain spectrum, DE algorithm is employed to find the optimal wavelengths and powers of pumps. The well-trained ELM model is incorporated into the evolution of DE to accelerate the search process. The results show that the designed RFAs with the optimized pump parameters achieve the desired gain performance and meanwhile maintain very low level of gain ripple. In comparison to other related methods, the proposed method significantly shortens the computation time and enhances the overall optimization efficiency, which offers potential for real-time adjustment and flexibility of RFA design.