Development of the HT-BP Neural Network System for the Identification of a Well-Test Interpretation Model
Wonmo Sung, Inhang Yoo, Sinyoung Ra, Heungjun Park · SPE Computer Applications · 1996
Summary The back propagation (BP) neural network approach has been the subject of recent focus because it can identify models for incomplete or distorted data without performing data preparation procedures. However, this approach uses only partial sets of data to reduce computing time and memory, and it may miss the points representing characteristics of the curve shape. Therefore, the resulted model may not be correct, forcing one to use sequential neural nets to find the correct model. We present the Hough Transform (HT) method combined with the BP neural network to improve this problem. With the aid of an HT, we can extract one simple pattern, including noisy and extraneous points, from the full-set data. A number of exercises also have been conducted for the published well-test data with the artificial intelligence neural network identification system (ANNIS) we developed. The results show that ANNIS is quite reliable, especially for the incomplete or distorted data. They also demonstrate that the modified Levenberg-Marquart interpretation model, also developed in this work, successfully estimates reservoir parameters.