Machine Learning Outlier Detection Algorithms for Enhancing Production Data Analysis of Shale Gas | Chapter 9 | Fundamental Research and Application of Physical Science Vol. 4

 Economically judging shale gas reservoirs, that have huge amounts of reserves, is challenging by way of the intricate driving methods. Decline Curve Analysis (DCA) has historically been regarded as the plainest approach for production prediction in rock gas reservoirs since it only demands production annals. Nevertheless, uncertainties persist in selecting a appropriate DCA model to match the production practice of shale gas wells. Moreover, the result data are typically boisterous due to the dynamic changes in block size employed to organize the bottom hole abounding pressure and the periodic shut-in employed to remove the befriended water. Various statistical and machine-education approaches have been used in the study of production predicting, reservoir property belief, and resource evaluation. However, many of these systems are not effective in detecting production styles and reservoir signals. The aim of this unit is to comperhensivly reviewe different machine intelligence algorithms for outlier detection and evaluate their efficacy in enhancing the value of production data for DCA. Out of these algorithms, five were regarded unsuitable since they removed entire sections of result data instead of only recognizing and eliminating sporadic data points. The remaining algorithms (seven) went through a rigorous evaluation, accompanying a presumption that 20% of the production dossier is composed of outliers. Further, eight distinct DCA models were intentional and implemented before and after erasing the noise to test their sensitivity to explosion. It was found that usually reconstructing the production data betters their goodness of fitting and reliability of prognosis. The clustered-based eccentricity factor, k-nearest neighbor, and bent-based outlier determinant algorithms were found expected effective in improving the dossier quality for DCA, whereas the theory of probability outlier selection and subspace oddity detection algorithms were least productive. Furthermore, certain DCA models, such as Arps, Duong, and Wang models, illustrated less sensitivity to the removed commotion, irrespective of the outlier relocation algorithm used. On the other hand, the capacity law exponential, logistic development, and stretched exponent result decline models exhibited greater sympathy to noise eradication, with their performance variable based on the employed exception removal algorithm. The stage discusses the optimal mixture of outlier detection algorithms and DCA models that take care of potentially mitigate the doubts associated with production predicting and reserve estimation in shale smoke reservoirs.

Author(s) Details:

Taha Yehia,
Department of Petroleum Engineering, Faculty of Engineering and Technology, Future University in Egypt (FUE), Cairo-11835, Egypt.

Ali Wahba,
Department of Petroleum Engineering, Faculty of Petroleum and Mining Engineering, Suez University, Suez-11252, Egypt.

Sondos Mostafa,
Department of Petroleum Engineering, Faculty of Petroleum and Mining Engineering, Suez University, Suez-11252, Egypt.

Omar Mahmoud,
Department of Petroleum Engineering, Faculty of Engineering and Technology, Future University in Egypt (FUE), Cairo-11835, Egypt.

Please see the link here: https://stm.bookpi.org/FRAPS-V4/article/view/10593