Using 2W-PE method based on machine learning to accurately predict field strength distribution in flat-top obstacle environment

Abstract

A new two-way parabolic equation (2W-PE) method is proposed in this paper which uses machine learning to determine boundary condition equations, and accurately predicts the electromagnetic field strength distribution in the environment of media ground and obstacles. On the basis of previous researches, the obstacle area is decomposed according to the principle of domain decomposition, and the processes of solving 2W-PE in each subdomain are briefly explained. Because the direction and strength of the incident waves on the boundaries vary greatly with the propagation environment, it is difficult for us to determine boundary condition equations. The machine learning method is introduced here, and method of moments (MoM) is applied to generate sample data sets, thus, coefficients in boundary condition equations can be trained through the non-end-to-end neural network combined with backpropagation algorithm and gradient descent method. So that no matter how the environment changes, appropriate boundary conditions can be obtained by the well-trained neural network and help improving the accuracy of 2W-PE. Simulation results show that the accuracy of the new 2W-PE method based on machine learning is better than that of 2W-PE with traditional boundary conditions by comparing with the results of MoM, which also reflects the advantages of machine learning in radio wave propagation analysis.Therefore, this paper provides an innovative and reliable method for accurately predicting field distribution in flat-top obstacle environments.