AN APPLICATION OF PATTERN RECOGNITION TO AUTOMATIC DETERMINATION OF LITHOLOGY FROM WELL LOGS | ||||
| International Conference on Aerospace Sciences and Aviation Technology | ||||
| Article 60, Volume 2, A.S.A.T. CONFERENCE 21-23 April 1987 , CAIRO, April 1987, Page 1243-1254 PDF (2.49 MB) | ||||
| Document Type: Original Article | ||||
| DOI: 10.21608/asat.1987.26224 | ||||
 View on SCiNiTO | ||||
| Authors | ||||
| M. M. SYIAM1; A. SHAHIN2; F. SALAEH3 | ||||
| 1Graduated student. | ||||
| 2Associate professor, Det. of A/C Elec. Equip.& ARM., M.T.C, Cairo, Egypt. | ||||
| 3Professor, Dpt. of Electronics and Communications Engineering, Cairo University, Guiza, Egypt. | ||||
| Abstract | ||||
| Determination of lithology from well logs is the subject of interest of many geologists. The method introduced in this paper gives an automatic determination of lithology from well logs using a 1-nearest neighbor classifier. The 1-nearest neighbor rule (1-NNR) with editing and condensing techniques 4 used for the design of 1-NN classifier.For this study a training data set has been obtained from a key well in (ABU GARADIG) field. This well has a suitable suite of logs and a continuous core as a geological reference. First, the training data set is edited to obtain a homogeneous clusters of data that improves the performance of the 1-NNR. Basic condensing technique and the ordered condensing technique are applied on the edited data set to obtain a reference patterns for the 1-NN classifier. HOLD-OUT method and ROTATION method are used to estimate the performance of the 1-NN classifier. Both methods partition the edited data set into design data set and test data set. The 1-NN classifier is trained on design data set and tested on the test data set. The estimated recognition rate of the proposed classifier using all patterns in the design data set as a reference patterns is compared with ones using condensed data subset and ordered condensed data subset.The comparison have demonstrated that the 1-NN classifier using ordered condensed data subset as a reference patterns is the optimum one in this application. The designed classifier is tested on other several wells in the same field.The analysis of the experimental results demonstrates that all the output lithologies are in accordance with geological references. | ||||
|
Statistics Article View: 106 PDF Download: 219 |
||||
