CONVOLUTIONAL NEURAL NETWORKS AS A METHOD TO SOLVE ESTIMATION PROBLEM OF ACOUSTIC WAVE PROPAGATION IN POROELASTIC MEDIA
Ключевые слова:
Neural network, Convolutional neural network, acoustic wave propagation, predicting models, supervised learning, activation function, PyTorch.Аннотация
This article concerns the problem of the acoustic wave propagation in the 3 layered half-space. The
first and third layers are assumed to be solid, whereas the second is assumed to be poroelastic. This article considers
results of finding acoustic wave propagation with regard to its depth and time (further called solution to the forward
problem) [1-4] and tries to estimate the initial physical properties of aforementioned three layers. The aim of this
article is to create a convolutional neural network that estimates said properties, namely speed of sound in each layer
and porosity of the second layer. Model was built using “PyTorch”, open-source machine learning library. In order to
evaluate the initial coefficients of acoustic wave propagation the convolutional neural networks were used. During
the procedure, 3 convolutional hidden layers and 2 fully connected linear hidden layers were used. The data for data
characterization was simulated by iteratively solving forward problem of acoustic wave propagation described by
Stokes equation and continuity equation with given initial values of the acoustic model.
