Improving water saturation estimation in a tight shaly sandstone reservoir using artificial neural network optimized by imperialist competitive algorithm – A case study

• Data preprocess techniques were implemented to improve the quality of used data.
• The ICA parameters have been carefully designed to optimize the neural network.
• This strategy effectively combines back propagation with ICA.
• The comparison of ICA-ANN with BP-ANN, Indonesian and improved Indonesian models was carried out.
• The results showed the superiority of ICA-ANN on convergence and solution.

Tight reservoir refers to reservoirs with low porosity and permeability. Estimating Petrophysical parameters of Tight Gas Sand (TGS) reservoirs is one of the most difficult tasks in reservoir characterization studies. These reservoirs usually produce from multiple layers with different and complex properties. Water saturation is an important petrophysical property representing the fraction of pore volume occupied by formation water that needs to be determined accurately when attempting to characterize hydrocarbon reservoirs. The exact determination of water saturation leads to a precise evaluation of initial hydrocarbon in place, which in turn provides valuable insight into future oil field development plans. In this paper, a model based on feed-forward - back propagation error Artificial Neural Network (ANN) optimized by Imperialist Competitive Algorithm (ICA) to predict water saturation in TGS reservoirs is proposed. ICA is employed to obtain the optimal contribution of ANN for a better water saturation prediction. Conventional well log data are used as input and water saturation data measured on core samples as output variables to the ANN model. In the current study, a number of 2200 data taken from 12 wells selected from a number of TGS basins are used to build a database. The performance of the proposed ICA-ANN model has been compared with the conventional petrophysical and ANN models. Based on cross validation measures, the results clearly show that the ICA-ANN model has outperformed the conventional methods in terms of effectiveness, robustness and compatibility.