Development of Automated Neuro-Simulation Protocols for Pressure and Rate Transient Analysis Applications

Restricted (Penn State Only)
Author:
Zhang, Jian
Graduate Program:
Energy and Mineral Engineering
Degree:
Doctor of Philosophy
Document Type:
Dissertation
Date of Defense:
August 16, 2017
Committee Members:
  • Turgay Ertekin, Dissertation Advisor
  • Turgay Ertekin, Committee Chair
  • Sanjay Srinivasan, Committee Member
  • Li Li, Committee Member
  • Kultegin Aydin, Outside Member
Keywords:
  • Neuro-Simulation
  • Machine Learning
  • Neural Network
  • Numerical Reservoir Simulation
  • Pressure Transient Analysis
  • Rate Transient Analysis
Abstract:
Traditional rate and pressure transient analysis apply simplified analytical models to calculate the reservoir characteristics. The analytical models are based on ideal assumptions which are hardly satisfied in practice, and the analysis process also relies on well-trained human experts and their valuable experiences. The traditional rate and pressure transient analysis approaches demand extensive resources in terms of personnel and time. Instead of relying on analytical models and human experts, Artificial Neural Network (ANN) tools, as an analog of the neural systems in human brains, are developed for rate and pressure transient analysis. The ANN tools developed are proved successful in processing complex problems with much less cost. Different from analytical models, however, each ANN tool developed is only applicable in analyzing problems of certain types of reservoir and well conditions. If the problem is not within the specification or range of the parameters of the existing ANN tools, a new ANN tool is required to be developed from the scratch. The development process of the ANN tools is called the “Neuro-Simulation” protocol. It applies numerical reservoir simulators to generate randomly distributed data sets and train an ANN to analyze the transient data. This protocol relies on well trained and experienced researchers and commercial numerical reservoir simulator and ANN development software. In this study, an automated Neuro-Simulation protocol is developed to assist and accelerate the development process of ANN tools for rate and pressure transient analysis. The protocol has the capability of automating the major processes in the Neuro-Simulation protocol. The protocol is implemented into a comprehensive toolkit. In this toolkit, a comprehensive and generalized in-house numerical reservoir simulator is implemented. The rectangular and radial-cylindrical grid systems are implemented based on the framework, and the black oil, compositional and naturally fractured system fluid flow models are developed into the simulator. An ANN development tool is developed with multiple built-in activation functions and learning algorithms. The automated Neuro-Simulation protocol based on the communication and cooperation between the in-house numerical reservoir simulator and ANN development tool is established. The toolkit also contains a user-friendly GUI and mini PC to provide convenience. The toolkit is designed to be general, flexible and independent. A tight gas reservoir system with dual-lateral horizontal well is studied to illustrate the capability of the protocol and toolkit.