Выпуск: 1.5 , Том: 18 , Год издания: 2018

Место издания: Albena

Geophysical exploration, the necessary part of oil and gas exploration, generates more and more data, subject of processing. The most powerful super computer clusters are used by business and academic institutions. However, often it is necessary to have evaluation of the measured data shortly after measurements, even in the field. Modern computer architectures, namely Graphic Processing Units (GPUs) and Field Programmable Gates Arrays (FPGAs) provide a good basis for PC-based fast data processing, to have, say, supercomputer on the table. Here, we present several examples of code execution acceleration for seismic data processing. Seismic data is characterized by multidimensionality, large size and irregular structure. For optimal representation of this data one need to preprocess them by decomposing the data using appropriate basis. With NVIDIA CUDA technology for programming on GPU we implemented a fast algorithm of forward and inverse 3D wave-packet transform. The code was optimized based on physical device characteristics and structure of the algorithm. We obtained speed-up ~45 for one GPU and analyzed scalability for several GPUs. The program was tested on synthetic seismic data for their compression, de-noising and regularization. We also consider a seismic salt stringer image. The interpretation in the shadow zone beneath the stringer has complications due to that the diffracted and transmitted wavefields destructively interfere causing poor image. For simulating the real image, we evaluate seismic wavefields in the shadow zone by combining the Transmission-Propagation-Diffraction Operator Theory and the Tip-Wave Superposition Method (TPDOP & TWSM). This mathematical model has a layer with two flat boundaries, one of which has a dense coin-shaped addition reminding an anhydrite disk. We used GPU-cluster to accelerate modeling and give an estimated time of wavefields simulation for stringer model.