Application of Machine Learning to Construct Solitons of Generalized Nonlinear Schrödinger Equation
摘要
This paper presents an application of the PINN method to solving the non-integrable soliton generalized nonlinear Schrödinger equation, with evaluations on the solution accuracy of the influence of introducing in the PINN algorithm the conservation laws. An effect of choice of different activation functions is also analyzed. Despite the fact that obtained results demonstrate the best accuracy for the model with the SIN activation function without a use of conservation laws, experimental results show that they may be useful in cases of non-optimal choices of activation functions or collocation points.