Electronic Thesis/Dissertation


Data-Driven Analytics For Oil and Gas Well Parameter Estimation Open Access

Downloadable Content

Download PDF

Oil and Gas Exploration and Production (E&P;) projects routinely experience cost overruns and delays, with about 64% of new projects exceeding their budgets and almost 75% completed behind schedule partly as a result of non-optimal project designs1. Owing to the highly complex and heterogeneous nature of most petroleum reservoirs, the design of drilling and completions projects are typically associated with a high degree of uncertainty regarding vital reservoir parameters such as permeability. Many researchers have tried to deterministically develop mathematical functions with a general application for determining permeability; however, it is difficult to build a universal model relating the mathematical behavior of measured variables2 due to the extremely complex heterogeneous nature of petroleum reservoirs because every formation has its unique petrophysical characteristics. No generalized function between variables in well-log data and permeability has been uncovered. Statistical analytical methodologies that leverage historical and empirical geophysical and petrophysical data are currently widely used to estimate permeability; however, these empirical assumptions invariably result in over- or underestimation. The application of deep learning (DL) neural networks algorithms to identify patterns in the well data can deliver much-needed insights faster and improve the accuracies of parameters predictions used in the design process. Artificial Neural Networks algorithms have advanced rapidly and are being deployed successfully in many industries including aerospace, pharmaceuticals, financial institutions, crime prevention, and social media (Facebook, Google, YouTube). For instance, in handwriting recognition, commercial banks use them to process checks and post offices use them to recognize addresses with an accuracy of over 99%.Currently, the application of machine learning techniques such as deep learning neural networks to optimize oil & gas projects is very limited in the oil and gas E&P; industry. Enormous amounts of data are routinely collected by oil & gas E&P; companies and held in siloed disciplines. Less than 2% of the data collected are actually utilized to perform advanced data analytics. Deep learning models could be used to integrate and analyze such data to gain competitive business intelligence. In this study, an ensemble of base learning algorithms was successfully trained and used to accurately predict reservoir permeability from well-log data for a shaly sandstone reservoir. Knowledge of such accurate permeability could be used to optimize completions designs such as hydraulic fracturing.

Author Language Keyword
Date created Type of Work Rights statement GW Unit Degree Advisor Committee Member(s) Persistent URL