Enhanced interpretation of fault from 3D seismic data using seismic data conditioning and artificial neural network approach in liuzan area of the Nanpu sag, Bohai Bay Basin, China

Abstract:

In order to improve the quality of fault image, in this study, we adopted seismic data conditioning and seismic multi-attribute for calculating a new hybrid attribute through a supervised multilayer perceptron neural network in the Liuzan area located in the northeastern part of of the Nanpu Sag, Bohai Bay Basin, China. Firstly we conditioned original seismic data by using the dip-steering cube calculated from the original seismic data. Then, we calculated seismic attributes such as similarity, polar dip, curvature, laplacian, RMS and REF from the conditioned data that can effectively improve the image of fault features. Thirdly, we picked a set of “picks” (known as the training data sets) of seismic line 2090 from the seismic data volume which represents the presence or absence of faults. Fourthly, we applied the supervised multilayer perceptron neural network to train over the selected set of seismic attributes calculated at the fault and non-fault positions. The neural network consists of 16, 8, and 2 nodes in the input layer, hidden layer and output layers, respectively. Finally, we obtained a new fault probability cube containing data values ranging from 0 to 1, where 0 and 1 represent the lowest and highest probability of the presence of the fault. The strong unwanted (noisy) information was effectively suppressed from the original seismic data in the first phase of the data conditioning. Moreover, we observed that REF provided the highest contribution followed by polar dip, similarity longwindow parallel and similarity shortwindow parallel. The normalized root-mean-square error values for both test and train data produce a minimum value between 0.77 and 0.86. A minimum misclassification percentage of 16.91-23.2% is obtained between the train and test data sets. This research provides an effective way of fault imaging from 3D seismic data. Therefore, the efficacy and accuracy of fault imaging via a combination of seismic multi-attribute are more acceptable by interpreters.