학술논문
An integrated workflow of history matching and production prediction for fractured horizontal wells.
Document Type
Article
Author
Source
Subject
*NEWTON-Raphson method
*STOCHASTIC approximation
*APPROXIMATION algorithms
*PRODUCTION scheduling
*MANUFACTURING processes
*HORIZONTAL wells
*
*
*
*
*
Language
ISSN
1070-6631
Abstract
This paper presents a novel method for history matching and production prediction for fractured horizontal wells by combining the data space inversion method (DSI) with the embedded discrete fracture model (EDFM), referred to as DSI-EDFM. In this approach, several initial numerical models with varying reservoir geological and fracture geometry parameters, but identical production schedules, are generated through random sampling and then run using the EDFM. The DSI method is subsequently employed to process the production data, creating a proxy model that matches actual historical data and predicts production performance by solving a quadratic optimization problem. A key improvement over the original DSI method is introduced, providing and proving the conditions under which the optimization problem derived from DSI is a positive definite quadratic optimization problem. With these conditions, the optimal solution can be directly obtained using the Newton method without any iterations. Furthermore, it is identified that overfitting issues frequently arise when using the Newton method for DSI. However, the simultaneous perturbation stochastic approximation algorithm effectively mitigates this problem, allowing the proposed DSI-EDFM to handle real reservoirs and uncertainty parameters efficiently. Three numerical examples are implemented to validate the method, including depleting development, water flooding operations, and the flush stage of fractured horizontal wells. The results demonstrate that the proposed DSI-EDFM achieves high accuracy in conducting history matching and performance prediction for fractured horizontal wells, even under complex flow model conditions and with a limited number of initial models. Additionally, the accuracy improves as the number of initial models increases. [ABSTRACT FROM AUTHOR]