|
Article information
2015 , Volume 20, ¹ 1, p.25-37
Glinskikh V.N., Gorbatenko V.A.
Electromagnetic logging data inversion on GPU
The paper deals with the development of algorithms and software for processing of electrical logging data and interpretation on GPUs. The traditional scheme for solving the inverse problem of borehole geoelectrics is based on the targeted search of model parameters. It includes experimental and simulated data matching, and utilizes iterative solution of the direct problem. Ambiguous relationship between electromagnetic field characteristics and geological medium properties along with inaccuracy of measured data leads to imprecise determination of parameters in the resulting model. As a result, instead of an unique solution there are a great number of equivalent models that can be described by quasi-solution areas in the inverse problem. For creating such quasi-solution areas, we use a complete enumeration of model parameters. However, a sequential computing algorithm is utterly resource-intensive but inefficient. Substantial improvement of computational efficiency may be achieved with the help of a parallel algorithm on multiprocessors. We have implemented a parallel algorithm for the numerical solution of the electromagnetic logging of two-dimensional inverse problem on GPUs, which utilizes NVIDIA CUDA technology. Along with that, we have revealed various approaches to the optimization of the GPU parallel algorithm, based on detailed analysis of computational performance and efficiency for different memory types. The processing speed and performance of the optimized version of the parallel algorithm on NVIDIA GPUs with architectures of different generations (Kepler, Fermi, Tesla) are estimated. It is shown that implementation of this up-to-date GPU parallel algorithm speeds up calculations by a factor of 1400 and increases performance up to 530 Gflops in comparison with the identical sequential algorithm. In addition, we have performed the numerical inversion of noisy synthetic data and carried out the comparative analysis of the inverse problem quasi-solution areas in complex-structure geologic models. Using the developed algorithm the electrical conductivities of invasion zones and non-invaded beds are reconstructed, which includes the estimation of inversion errors.
[full text]
Keywords: parallel computations, GPU, NVIDIA CUDA technology, inverse problem, electromagnetic logging, electrical conductivity
Author(s):
Glinskikh Viacheslav Nikolaevich
PhD. , Associate Professor
Position: Head of Laboratory
Office: Trofimuk Institute of Petroleum Geology and Geophysics SB RAS
Address: 630090, Russia, Novosibirsk
Phone Office: (383) 3304505
E-mail: GlinskikhVN@ipgg.sbras.ru
Gorbatenko Vadim Alexandrovich
Position: Student
Office: Trofimuk Institute of Petroleum Geology and Geophysics SB RAS
Address: 630090, Russia, Novosibirsk
Phone Office: (383) 3304505
E-mail: GorbatenkoVA@ipgg.sbras.ru
References:
[1] Tabarovskiy, L.A., Epov, M.I., Sosunov, O.G. Otsenka razreshayushchey sposobnosti elektromagnitnykh metodov i podavlenie pomekh v sistemakh mnogokratnogo nablyudeniya (teoriya, algoritmy, programmy) [Estimating resolving power of electromagnetic methods and noise reduction in systems with multiple observations (theory, algorithms and programs)]. Novosibirsk: Preprint IGIG SO AN SSSR; 1985 (7):48. (In Russ.)
[2] Tabarovsky, L.A., Epov, M.I. Estimating resolution of electromagnetic methods. Geologiya i Geofizika. 2006; 47(5):568–578. (In Russ.)
[3] Epov, M.I., Glinskikh, V.N., Nikitenko, M.N., Sukhorukova, C.V. Fast solution of thetwo-dimensional high-frequency electromagnetic logging inverse problem in oil and gas wells. Geophysical Journal. 2012; 34(4):292–297 (In Russ.)
[4] CUDA C Programming Guide. Design Guide (PG-02829-001_v6.0, February 2014). NVIDIA CUDA Toolkit Documentation. 2014. Available at: http://docs.nvidia.com/cuda/pdf/CUDA_C_Programming_Guide.pdf (date of visit: 01.09.14).
[5] Boreskov, A.V., Kharlamov, A.A. Parallel'nye vychisleniya na GPU. Arkhitektura i programmnaya model' CUDA [Parallel Computations on GPU. CUDA Architecture and Software Model]. Ìoscow: Izdatel'stvo Moskovskogo universiteta; 2012: 332. (In Russ.)
[6] Glinskikh, V.N., Epov, M.I., Labutin, I.B. Simulation of electromagnetic logs using graphics processing unit. Computational Technologies. 2008; 13(6):50–60. (In Russ.)
[7] Labutin, I.B., Surodina, I.V. Development of parallel algorithms for solving logging problems on GPUs. Parallel computational technologies 2012: transactions of the international scientific conference. Chelyabinsk: YuUrGU, 2012:690–697. (In Russ.)
[8] Surodina, I.V., Epov, M.I. High-frequency induction data affected by biopolymer-based drilling fluids. Russian Geology and Geophysics. 2012; 53(8):817–822.
[9] Nechaev, O.V., Shurina, E.P., Fedoruk, M.P. Using the vector finite element method fornumerical solutions of the quasi-steady Maxwell’s equations. Computational Technologies. 2004; 9(5):73–81. (In Russ.)
[10] Eremin, V.N., Nechaev, O.V., Haberhauer, S. Parallel realization of mathematical modelling of electromagnetic logging processes using VIKIZ probe complex. Computational Technologies. 2007; 126:18–33. (In Russ.)
[11] Shurina, E.P., Shtabel, N.V. Analysis of vector finite element approximations of Maxwell’s equations in anisotropic media. Computational Technologies. 2013; 18(4):92–104 (In Russ.)
[12] Epov, M.I., Glinskikh, V.N. Electromagnetic Logging: Simulation and Inversion.Novosibirsk: GEO; 2005: 98. (In Russ.)
[13] Epov, M.I., Nikitenko, M.N. System for one-dimensional interpretation of high-frequency induction logging sounding data. Geology and Geophysics. 1993; 34(2):124–130. (In Russ.)
[14] Glinskikh, V.N., Nikitenko, M.N., Epov, M.I. Processing high-frequency electromagnetic logs from conducting formations: linearized 2D forward and inverse solutions with regard to eddy currents. Russian Geology and Geophysics. 2013; 54(12):1515–1521.
[15] NVIDIA Nsight Visual Studio Edition 4.0 User Guide. NVIDIA CUDA Technical Documentation. 2014. Available at: http://docs.nvidia.com/gameworks/index.html#developertools/desktop/ nsight_visual_studio_edition_user_guide.htm (date of visit: 01.09.14).
[16] Profiler User’s Guide (DU-05982-001_v6.0, February 2014). NVIDIA CUDA Technical Documentation. 2014. Available at: http://docs.nvidia.com/cuda/pdf/CUDA_Profiler_Users_Guide.pdf (date of visit: 01.09.14).
Bibliography link:
Glinskikh V.N., Gorbatenko V.A. Electromagnetic logging data inversion on GPU // Computational technologies. 2015. V. 20. ¹ 1. P. 25-37
|
