Data Mining Techniques for Segmentation Analysis of Seismic Data

Daniel R.S. Moraes, Rogério P. Espíndola, Márcia K. Karam, Alexandre G. Evsukoff, and Nelson F. F. Ebecken
Federal University of Rio de Janeiro, Rio de Janeiro, Brazil

Increasing oil and gas production is crucial for the Brazilian development. In order to overcome the growing demand, it is necessary to reduce the exploration risks and costs, as well as the time necessary for the development of new ventures. The complexity of deep water exploration and exploitation has been challenging scientists and engineers to find new and efficient computer-based tools that bring together technology's state-of-the art. Interpretation of seismic data has played an important role in exploration phase. Usually there exists no a priori information about the geological structure and performing a systematic analysis of all the seismic data is time consuming. Recent technologies of data mining have opened a new frontier for oil & gas exploration, providing meaningful information at low cost, allowing experts to quickly scan through multiple volumes of data and combine information to get the optimal view of any geological feature of interest. Data clustering methods are important tools in data mining, since they enable the detection of hidden patterns not explicitly recorded in data. Using clustering methods on seismic attributes data, it is expected that different lithologies will present have different patterns that will enable the clustering algorithm to separate lithology patterns. In this work two clustering methods are compared based on clusters quality criteria: the Self Organizing Maps (SOM) neural networks and the Fuzzy C-Means (FCM) algorithm based on fuzzy sets theory. The analysis is performed over a 3D seismic survey of the Namorado field located in the Campos Basin.