MACHINE LEARNING APPLICATION TO SEISMIC DATA ANALYSIS

ABSTRACT

Seismic reflection is used to examine the interior of the earth through analysis of mechanical waves. It has found use by many industries especially the petroleum industry for the exploration of hydrocarbon. Oil companies deploy seismic reflection to help determine where to place wells for most effective exploration. To identify prospects and make drilling decision, oil operating companies have used 3D seismic since the 1980s. This is achieved by creating geologic models with the integration of seismic and sub-seismic data analysis. The process of developing geologic model is very laborious and time consuming. In a bid to solve the challenge of identifying hydrocarbon prospects using seismic data within stringent time deadlines, the application of deep learning algorithms for seismic interpretation tasks is being deployed. This process involves annotating dataset for training and testing of the model. However, the greatest challenge is that, there is paucity of large publicly available annotated dataset to carry out seismic facies classification and thus, the available dataset has to be utilized. This research has led us to training a fully convolutional neural network called U-Net in order to perform semantic segmentation of seismic data. The U-Net is a variant of CNN developed to accept very few datasets as input and still produce precise results, unlike other deep learning algorithms that requires very large dataset to produce acceptable results.