Development of an Artificial Neural Network for Pressure and Rate Transient Analysis of Horizontal Wells Completed in Dry, Wet and Condensate Gas Reservoirs of Naturally Fractured Formations

Open Access
Gaw, Hussain
Graduate Program:
Energy and Mineral Engineering
Master of Science
Document Type:
Master Thesis
Date of Defense:
June 04, 2014
Committee Members:
  • Turgay Ertekin, Thesis Advisor
  • ANN
  • Pressure transient
  • Rate transient
  • Naturally fractured formations
In order to meet the increasing demand for natural gas, it has become important to increase production. Drilling horizontal wells in naturally fractured gas reservoirs can greatly help in achieving the desired high gas production. Reservoir simulation is used to history match production profiles in order to predict important reservoir characteristics. Nevertheless, the use of commercial simulators is time consuming. Therefore, the search for alternatively fast means to predict reservoir properties promoted the use of artificial neural networks. These networks have grown in popularity because of their ability to solve non-linear relationship problems, generate accurate analysis and predict results from large number of historical data. Artificial neural network (ANN) is a mathematical model, which tries to mimic the structure and functionality of a human biological network in acquiring, storing and using experimental knowledge. ANN predicts target outputs when given a set of input and it can be trained until optimum results are reached. The main objective of this study is to develop artificial neural networks that will perform pressure and rate transient analysis for dry, wet and condensate gas reservoirs with fixed composition. The network is given three main variables: production profiles, well parameters and reservoir characteristics, where each variable can be predicted in the presence of the other two. Three separate artificial neural networks were trained for each of the three models. These artificial neural network showed good results and were able to predict the desired outputs with an average error less than 10%.