Sonic Log Effect Assessment on Lithology Estimation by well Logs using Artificial Neural Network in One of the South Pars Field Reservoirs

Lithology Prediction is a fundamental stage in petroleum engineering and formation evaluation. This research is a type of artificial neural network modeling in order to use well bore logs in lithology prediction in a South Pars hydrocarbon reservoir. Here, two networks with three-layer back propagation (BP) method and Levenberg-Marquwardt algorithm have been used for lithology estimation. The network in the first stage, utilized gamma-ray, neutron, density and photoelectric effect (PEF) logs as inputs. On the other hand, the data of sonic log has been also added to the inputs and the results of the two stages have been compared in the second network. Considering the excessive coring expenses, this method can be used as a milestone in decreasing the coring expenses. In this paper, the following procedure is considered first, data four wells in South Pars field has been used. Second, the network has been trained in one of the reservoir wells (well C) in which core analysis data was available. Third, in another well (well D) which its data does not affect the training process, it has been tested. Forth, after approching to the desired level of confidence in network efficiency, it has been utilized to estimate the lithology in the two other wells (wells A and B). lithologies investigated interval consist of: Dolostone, Limestone, Dolomitic Limestone, Limy Dolostone, Anhydrite, Shale, Shaly Limestone and Shaly Dolostone. In the first case, the mean square error (MSE) for well A was 0.081 and for well B was 0.094. In the second case, the sonic log was added to other input, MSE has become 0.051 in well A and 0.063 in well B. Comparing two cases, it was revealed that the model accuracy has been improved significantly in the second case and sonic log data caused the estimated lithology become closer to the real case.