Research On Forward Modeling Of Magnetic Field Excited By Underwater Pressure Wave Based On LSTM

Large scale underwater disturbances are usually caused by internal waves that exist widely in the ocean.The pressure waveguide in deep water causes large-scale and rapid diffusion of sea water.The moving conductive particles cut the earth’s magnetic field and generate induced current in sea water.The current vibrates mechanically with sound waves and generates induced magnetic field in the ocean.Due to the large distribution range and slow attenuation of the magnetic field,it is possible to observe it on the sea surface.In order to simulate this kind of acoustic and magnetic radiation process,we need to combine the sound field Helmholtz equation with the electromagnetic Helmholtz equation,and use a high-order finite element method:spectral element method(SEM)for numerical solution.A comprehensive study of the magnetic field signal can provide us with a reliable theoretical basis for in-depth understanding of the magnetic environment under the ocean background,the detection of magnetic anomalies of seabed resources,and the identification of underwater vehicles.However,the traditional multi physical simulation method is cumbersome in calculation,and also has the disadvantages of large amount of calculation and low efficiency.With the rapid development of artificial intelligence,neural network modeling method has been applied in many fields and greatly improved the computational efficiency.In this paper,by introducing a new neural network structure and corresponding training algorithm,we study the recursive neural network(RNN)and long-term and short-term memory neural network(LSTM)to simulate the forward modeling process of underwater pressure wave excited magnetic field.Based on RNN neural network,we set up RNN-AF for sound field and RNN-MF for magnetic field,and LSTM-AF and LSTM-MF neural networks based on LSTM neural network.In terms of the accuracy of electromagnetic forward modeling,the neural network algorithm and the traditional electromagnetic forward modeling algorithm are compared and analyzed.The results show that the electromagnetic forward modeling algorithm based on LSTM has stronger generalization and is suitable for dealing with the forward modeling problem of the magnetc field excited by underwater pressure wave.The accuracy can reach more than 98%.It has good robustness and accuracy,and significantly improves the calculation efficiency.T his method is fast and easy to apply.It provides a theoretical reference for predicting the magnetic field distribution under underwater pressure wave excitation.