A hybrid stochastic-gradient optimization to estimating total organic carbon from petrophysical data: A case study from the Ahwaz oilfield, SW Iran

Highlights

• To hybridize stochastic and gradient optimization in BP-ANN structure.

• To minimize or avoid the risk of getting stuck in local minima when using BP-ANN.

• Higher performance of the ACOR model over other optimization techniques.

• High accuracy results and low computational time for TOC estimation from logs.

Abstract

One of the most important geochemical data in petroleum exploration is total organic carbon (TOC) which is used to evaluate the hydrocarbon generation potential of source rocks. To measure this parameter, expensive and time-consuming geochemical experiments are carried out on few cutting or core samples. In this study, stochastic optimization algorithms (ant colony and genetic programming) were hybridized with gradient optimization in a back propagation neural network structure to estimate TOC from petrophysical logs. The methodology is illustrated by using a case study from four wells of the Ahwaz oilfield. The results show that the hybrid ant colony-back propagation neural network model (ACO_R-BP) provides better results compared to the other intelligent models used. MSE and R² of the ACO_R-BP model in testing samples are 0.0051 and 0.952, respectively. This level of accuracy along with the fast speed of the algorithm is highly desirable for the estimation of the TOC parameter. The findings of this research demonstrate that employing ant colony optimization to initialize weights and biases of neural networks minimizes or avoids the risk of getting stuck in local minima. The methodology introduced in this study has a good performance and can be used to synthesize geochemical logs for the other wells of the Ahwaz oilfield.