Improving permeability estimation using neural networks and NMR logs: Case study

This dissertation presents three methods for evaluating formation permeability using log data in the CTR field, Saudi Arabia. The first method used the concept of hydraulic flow units (HFU's). The second method derived formation permeability using neural networks and log data. The third method estimated permeability using calibrated nuclear magnetic resonance (NMR) permeability. For calibrating the NMR permeability, the alternating conditional expectation algorithm (ACE) was used.;Formation permeability controls the strategies for designing well completion, well stimulation and reservoir management. Formation permeability can be evaluated from core analysis, well test analysis, and well log analysis. However, the most economic method of evaluating formation permeability on a foot-by-foot basis is from well log analysis. The petroleum industry needs more accurate and reliable methods to estimate permeability from well log data.;We developed a new, unbiased clustering technique that was used in the application of the hydraulic flow unit method. The application of the HFU method to the CTR field data showed much-improved permeability estimation as compared to the conventional permeability estimation from porosity alone. Software was developed during this research to perform unbiased clustering of core data, carry out sensitivity runs on the formation data to determine the formation optimal number of HFU's, and provide the reservoir permeability-porosity relationship.;We used neural networks in this research to reconstruct bad porosity logs, generate synthetic nuclear magnetic resonance effective porosity and permeability, and estimate formation permeability using well log data. Powerful and valuable tools, efficiently implemented, neural networks enhanced formation analysis. Neural networks when implemented efficiently can enhance the formation analysis. Guidelines for selecting and processing data used for training neural networks were developed during this research.;NMR permeability was calibrated against core data in the CTR field using nonparametric regression. The nonparametric regression technique was based on variable transformation to generate regression relations between dependent and independent variables. This technique is called alternating conditional expectations (ACE), which is a general and computationally efficient algorithm for deriving optimal nonparametric transformation of variables.;The permeability evaluation methods were applied to five wells in the CTR field, Saudi Arabia. The permeability estimates from the three methods above were compared with the core permeability, resulting in good agreement. The methodology developed in this dissertation can be applied to virtually any formation to estimate formation permeability from well log data.