Research Of Processing Seismic Data On CUDA Architecture

In the field of seismic data processing, the increasing of data scale and the algorithm complexity has brought in huge challenges for the computing technology. Parallel processing was attracting the attention of geologists in the petroleum enterprises worldwide, thus how to quickly and efficiently parallel processing large scale seismic data has become a very important problem. Recently the technology of general purpose GPU has been already applied into many industry areas. Therefore in this paper, based on the CUDA (Compute Unified Device Architecture) platform, we do research on the parallel computing approaches and algorithms for the two important parts of the large scale seismic data processing module, including pre-stack seismic migration processing and coherent body of parallel processing technology.In order to solve the imaging problem for complex geological structure, and enhance the ratio of signal to noise and resolution of seismic data, in this paper, we studied the Kirchhoff pre-stack seismic migration, including Kirchhoff prestack time migration and Kirchhoff prestack depth migration. Since Kirchhoff prestack migration is one of the most time-consuming parts of seismic data process. In order to enhance the speeds of seismic data computing and displaying, based on CUDA, we redesigned the traditional Pre-Stack Kirchhoff depth Migration algorithm, including parallel computing ray travel times algorithm and parallel seismic imaging algorithm. Hereafter we also improved traditional Pre-Stack Kirchhoff time Migration algorithm, including CUDA-based Kirchhoff prestack time migration algorithm, CUDA-based wave equation algorithm, GPU and CPU communication algorithm. The experimental results showed that our new parallel algorithms of computing seismic data which are based on the CUDA platform can really greatly improve its performance.In addition, for the seismic coherence technology, it needs to run a long time by using traditional computing methods. In order to decrease the resolving cycle, in this paper, based on the CUDA platform we proposed a one-side rotation algorithm which improved the calculating method of the matrix feature value in the coherence C3 algorithm. Extensive experiments are conducted in a PC with Intel Core2Due CPU and NVIDIA GeForce 8800 GT graphic card. Without losing the accuracy of seismic imaging, CUDA-based coherence C3 algorithm has improved the processing speed 8 times faster than traditional CPU algorithms. Our research shows that processing seismic data based on CUDA platform is expected to offer a potential method for large-scale seismic data processing and the real-time of displaying.