Assessment of Pore Size Distribution Using Image Analysis and Artificial Neural Network Modeling

Assessment of Pore Size Distribution Using Image Analysis and Artificial Neural Network Modeling

Calculation properties of reservoir and predicted production status in different sections of the reservoir and wells, is major task for petroleum engineer. With a better understanding of the factors of production, a more accurate assessment can be carried out for development of the oil fields. Because the carbonate reservoirs are generally heterogeneous; their description and evaluation require the use of special methods and techniques. Identification the different types of porosity in a reservoir from different types of pores, is a parameter application for the reservoir description. Pore space image analysis is based on image analysis and categorizing them using computer programs. Information using this method is obtained very fast. The reason for using this technique is authentic evaluation of the pore size distribution using digital images for predicting reservoir quality and its performance. Since the separation algorithm should be able to identify pore spaces in terms of geometry. Thus, the selected parameters should be a good indicator of the geometry of space. After studying the extracted parameters from the pores and by trial and error neural network; the optimal parameters, were selected to separate pore spaces. Hence, the selected features were obtained, including eccentricity, elongation, solidity, rectangularity, roundness and equivalent diameter/major diameter, respectively. Then, after normalizing the optimum parameters, we use them as the input for the neural network, in which the target data are a variety of pore spaces in the rock. The neural network was trained and the best estimate of the network was obtained with 10 hidden neurons and correlation coefficient of 0.82936 and mean square error (MSE) of 0.035. Indeed; a model for the recognition of pore space within thin sections images using a neural network is introduced in MATLAB. As a result, the estimates for porosity types were obtained with MSE of 0.0789 between the actual data and the network data Finally, to determine the accuracy of the algorithms used to classify different types of pore spaces, their accuracies were individually examined. Accordingly, we could classify and separate a variety of pore spaces, including vuggy, inter-particle, complete moldic, intra-particle, and biomoldic pores, with the accuracy of 100%, 91.87%, 83.33%, 77.77% and 66.11%, respectively