Estimation of Total Organic Carbon from well logs and seismic sections via neural network and ant colony optimization approach: a case study from the Mansuri oil field, SW Iran

Abstract

In this paper, 2D seismic data and petrophysical logs of the Pabdeh Formation from four wells of the Mansuri oil field are utilized. ΔLog R method was used to generate a continuous TOC log from petrophysical data. The calculated TOC values by ΔLog R method, used for a multi-attribute seismic analysis. In this study, seismic inversion was performed based on neural networks algorithm and the resulting acoustic impedance was utilized as an important predictor attribute. Afterward, a probabilistic neural network was trained by using a set of predicting attributes derived from multiple regression. Subsequently, TOC was estimated by using seismic attributes with a correlation coefficient of 75%. In the next step of the study, the nonlinear ant colony optimization technique was utilized as an intelligent tool to generate a 2D TOC section from seismic attributes. Nonlinear ant colony optimization proposed an intelligently derived equation for which weight factors of each predictor seismic attribute in TOC estimation model were derived by using stochastic optimization. The results show that nonlinear ant colony equation (stochastic optimization) outperforms the probabilistic neural network model (gradient optimization).