The amount of total organic carbon (TOC) is one of the most important parameter in evaluating hydrocarbon source rock. This parameter is not only used for hydrocarbon geochemical studies but also plays an important role in evaluating the extension of hydrocarbon source rock. As the increase in TOC may indicate the presence of source rock, the depletion of TOC reveals no extension of source rock in a certain depth. Therefore the need for a powerful tool in this aspect is essential. One of the linear methods for solving such problem is artificial neural network, a biologically inspired computing method which has an ability to learn; self adjusted and are trained, capable of classification, image processing and different problem analysis, with an attempt to estimate.
This paper presents the features and framework for application of neural network in estimating TOC for hydrocarbon source rock in Binak oil field, Bushehr province, using well log data.
The results of this study reveal that Multi-Layer Perception (MLP) is the optimum network which was used for TOC estimation. MLP topology was a hidden layer with 6 nodes, back propagation momentum learning algorithm and tangent activation function. After training is completed, the estimated error calculated as 0.0013, and then the network performance was tested upon training and testing data. Ultimately the predicted TOC values were compared with the actual one which showed a reliable network performance (R=0.9956). Finally the sensitivity analysis was attempted on effective parameters and based on neutron porosity parameter (NPHI) found to be as the most sensitive, and the sonic travel time (DT), the least sensitive parameters in estimating TOC.