Application of machine learning for adaptive subtraction of multiple reflected waves

Abstract

This work is devoted to the development and testing of an algorithm for adaptive subtraction of multiple reflected waves using a convolutional neural network. The algorithm is one of the main steps in the method of suppression of multiple reflected waves based on the separation of wave forms in the Radon region. The paper considers the formulation of a problem for a neural network, the preparation of training and test data sets and the testing of the algorithm. Using a convolutional neural network allows to automate and speed up the adaptive subtraction procedure. The algorithm was tested on synthetic data. Testing shows the effective adaptation of multiple waves, as well as the importance of correctly constructing a model of multiples.