Optimization of Amplitude Versus Offset Attributes for Lithology and Hydrocarbon Indicators Using Recurrent Neural Network

• Main Authors: Refael, R., Hermana, M., Hossain, T.M.
• Format: Article
• Published: Springer 2022
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85133653489&doi=10.1007%2fs11053-022-10103-1&partnerID=40&md5=c7fea4f3ac6fff788b4956861de24ce3; http://eprints.utp.edu.my/33387/

This article demonstrates the implementation of recurrent neural network (RNN) model in optimizing amplitude versus offset (AVO) attributes for indicating lithology and hydrocarbon zone on seismic data. Several drawbacks exist in the conventional implementation of AVO attributes for hydrocarbon exploration, including ambiguous AVO amplitude response as direct hydrocarbon indicators (DHIs), high cost of conventional seismic inversion and ambiguity from a non-absolute range of AVO scale of quality factor of P-wave (AVO SQp) and AVO scale of quality factor of S-wave (AVO SQs) attributes. Hence, this study aimed to optimize the application of AVO attributes by implementing an RNN model that solves nonlinear approximation with a faster, more economical, and reliable approach, supplied with well data as the data control. Sixteen features were extracted from seismic data as input with two targets of SQp and SQs from wells as output in the Angsi field. The model consisted of the improved algorithm of standard RNN called gated recurrent unit layers, followed by a simple RNN layer, and fully connected dense layers to predict SQp and SQs as lithology and hydrocarbon indicators, respectively. The model managed to predict the I-35 hydrocarbon reservoir zone anomalies, characterized by low SQp (sand-prone interval) and high SQs (hydrocarbon-bearing interval). The results indicated that the proposed RNN model performed efficiently as an alternative approach to bypass the conventional seismic inversion method and to optimize the application of AVO attributes in indicating lithology and hydrocarbon zone in seismic data. © 2022, International Association for Mathematical Geosciences.