Performance of Artificial-Intelligence-based Modelling for Distributed Raman Amplification

In this work, we comparatively investigate five different regression techniques based on artificial intelligence models to emulate the physical model of distributed Raman amplifiers with a shaped continuous pump profile. The results indicate that an artificial neural network (ANN) structure including regularization layers, such as dropout and batch normalization, presented the best performance in terms of maximum error and mean squared error, with improvements of 46.19% and 29.63%, respectively, if compared to a state-of-the-art reference ANN.