TY - GEN

T1 - Simulated Annealing Algorithm for Model Reconstruction of the Four-Layer Medium with Elliptical Inclusion in the Third Layer

AU - Prokhorov, Dmitry

AU - Reshetova, Galina

AU - Bratchikov, Denis

N1 - Conference code: 24

PY - 2024

Y1 - 2024

N2 - The paper presents an approach for reconstructing the properties of two-dimensional viscoelastic medium with defined geometry using the simulated annealing algorithm. The inverse problem solution requires a lot of computational resources because the direct seismic modeling is performed at each iteration. The staggered grid finite-difference scheme is implemented using CUDA technology to speed up the solution of the direct problem by parallelization. The choice of the simulated annealing method for solving inverse problem is due to the method’s ability to avoid local minima of the target functional. However, the simulated annealing method needs a good coverage of the model space by realizations of random probing vectors. It leads to enormous computation time in the case of a four-layer medium with elliptical inclusion in the third layer, which has 37 parameters. Therefore, the sequential reconstruction of model parameters, where the simulated annealing algorithm searches for the parameters in 1 or 2-D subspace, is introduced. Nevertheless, the attenuation properties of medium were not reconstructed by simulated annealing. For their study, the deep convolutional neural network is used.

AB - The paper presents an approach for reconstructing the properties of two-dimensional viscoelastic medium with defined geometry using the simulated annealing algorithm. The inverse problem solution requires a lot of computational resources because the direct seismic modeling is performed at each iteration. The staggered grid finite-difference scheme is implemented using CUDA technology to speed up the solution of the direct problem by parallelization. The choice of the simulated annealing method for solving inverse problem is due to the method’s ability to avoid local minima of the target functional. However, the simulated annealing method needs a good coverage of the model space by realizations of random probing vectors. It leads to enormous computation time in the case of a four-layer medium with elliptical inclusion in the third layer, which has 37 parameters. Therefore, the sequential reconstruction of model parameters, where the simulated annealing algorithm searches for the parameters in 1 or 2-D subspace, is introduced. Nevertheless, the attenuation properties of medium were not reconstructed by simulated annealing. For their study, the deep convolutional neural network is used.

KW - seismic modeling

KW - simulated annealing

KW - viscoelastic medium

UR - https://www.scopus.com/record/display.uri?eid=2-s2.0-85200722318&origin=inward&txGid=9891399a30457b117d69f958715630d6

UR - https://www.mendeley.com/catalogue/bcf9a249-3df8-33f4-88ca-b29453b777b9/

U2 - 10.1007/978-3-031-65238-7_23

DO - 10.1007/978-3-031-65238-7_23

M3 - Conference contribution

SN - 9783031652370

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 334

EP - 351

BT - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

PB - Springer Science and Business Media Deutschland GmbH

T2 - 24th International Conference on Computational Science and Its Applications

Y2 - 1 July 2024 through 4 July 2024

ER -