Ryan Tyree

Date of Graduation


Document Type


Degree Type



Statler College of Engineering and Mineral Resources


Petroleum and Natural Gas Engineering

Committee Chair

Shahab D Mohaghegh

Committee Co-Chair

Samuel Ameri

Committee Member

Joseph Frantz


This work introduces a process of using AI neural networks, for analyzing complex datasets, in order to achieve a higher prediction accuracy in regards to frac hits, at the individual stage level, in the Marcellus Shale when compared to traditional linear methods.;We examined 63 producing wells (parent) along with 79 completed wells (child) to determine the best predictors for accurate frac hit predictions. Our dataset consists of 959 records with 77 predictors and a single binary output (YES or NO) for a frac hit occurrence. Linear methods make analyzing these 77 predictors, along with their interactions, difficult. Neural networks, specifically backpropagation learning algorithm that was used, integrated with a fuzzy pattern recognition algorithm, allow end users to analyze a seemingly endless number of predictors at one time in order to produce a model with increased prediction accuracy over linear approaches. The four techniques discussed include accepting the null hypothesis, a method we refer to as the industry standard, a modified version of the industry standard, and backpropagation algorithms.;In this work we observed a 92.9% prediction accuracy when using a backpropagation neural network. Traditional approaches for the same dataset yield overall accuracies of 73.0%, 64.8%, and 82.8% for the three approaches that are discussed, respectively. Increased prediction accuracy is important because this allows the operator to make proactive data driven decisions for changes in completion design, well spacing, shutting in the parent well prior to the offset frac, or simply doing nothing. These decisions are better justified with increased prediction accuracy, potentially saving the operator valuable time and money.