Modeling and Prediction of Resistivity, Capillary Pressure and Relative Permeability Using Artificial Neural Network
Mustafa Ba alawi, Salem Gharbi, Mohamed Mahmoud
Abstract
Capillary pressure and relative permeability are essential measurements that are directly affecting multi-phase fluid flow in porous media. The difficulty of calculating them rises being constrained to core analysis in the laboratory with many challenges of mimicking reservoir conditions. This makes capillary pressure measurement process to be both time consuming and expensive. However, as resistivity is conveniently obtainable, it can be used to predict both capillary pressure and relative permeability given all relation to wetting phase saturation. Artificial intelligence methods have achieved promising results in modeling extremely complicated phenomena in oil and gas industry. This study aims to find a relation between all of resistivity, capillary pressure and relative permeability. Ultimately, we are going to generate a model by utilizing Artificial Neural Network (ANN) technique to predict both capillary pressure and relative permeability from resistivity. In addition, the implemented technique will be used to improve the data quality and to extend its resolution to thousands of data points. After that, as countless artificial neural network architectures can be created, the most efficient one will be evaluated given its performance and accuracy results. This paper presents the use of artificial neural network technique to both model and predict capillary pressure and relative permeability from resistivity attained from core analysis, and define core samples pore distribution systems. It was found artificial neural network architecture captures the complexity of the physics of the problem. Thus, it successfully fulfilled the required prediction objective. Additionally, compared to the traditional methods, this is proved to be accurate, fast, and significantly cost effective. Consequently, this process could replace the current traditional approaches. Finally, as artificial intelligence techniques are improving exponentially over time nowadays, this will increase the accuracy of the model predictability majorly.