ACCELERATION OF SEISMIC REFLECTION DATA PROCESSING WITH CRS-STACK METHOD USING GPGPU-CUDA AND DIFFERENTIAL EVOLUTION

• Main Author: Maulana, Irfan
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/40465
• Institution: Institut Teknologi Bandung

Seismic reflection data processing with the Common Reflection Surface (CRS) method - Stack is currently felt less widespread, because the calculation process is slower compared to conventional methods. However, this CRS-Stack method is believed to be a method that can map the reflector tilted and to better regulate, while at the same time increasing the S / N ratio. The search for the attributes of CRS-Stack N, RN, and RNIP that have been adopted so far is to use the cascade method. If the attributes are searched directly, the best combination of CRS-Stack attributes is determined from all seismic data, so it requires a very long processing time. Previous research has shown that the use of parallel computing with CPU - OpenMP can improve CRS-Stack data processing performance with a direct method of 3.6 times faster than without parallel computing. However, faster seismic data processing is needed. Thus, the purpose of this study is to find alternative methods that can speed up the CRS-Stack data processing process compared to the previous methods. In this study, the same use of synthetic data needs to be done, so that the acceleration of processing seismic data from a method against other methods can be compared. The data processing acceleration method proposed in this study is GPGPU-CUDA parallel computing method and Differential Evolution optimization. The performance of these two methods is then compared to the CPU-OpenMP implementation proposed by previous researchers. The results showed that the performance speed of CRS calculations with the GPGPU-CUDA method increased 9 times and Differential Evolution increased 26 times faster than the CPU-OpenMP method. Improving the speed of data processing using the GPGPU-CUDA method is still not as good as the results of the Differential Evolution method. Although GPU uses more cores than CPU usage, the GPGPU-CUDA method has Input-Output problems between CPU and GPU, as well as using brute force for all combinations of parameters. Whereas in applying the Differential Evolution method, acceleration occurs because it uses the optimum value of a sample by a certain iteration. Thus, increasing the processing speed of reflection seismic data using the CRS-Stack method produced from this study is expected to increase its popularity, so that a better and more accurate description of the subsurface of the earth can be achieved with a much shorter time.