Intelligent recognition method of rock formation based on lithology recognition method is based on machine learning classification algorithm, which can classify and recognize logging data and logging map, and effectively learn and memorize the characteristics of rock formations in the reservoir.
The rock formation intelligent recognition method based on lithology recognition method is based on machine learning classification algorithm, which can classify and recognize the logging data and logging charts The method can compare four algorithms, namely, adsorbed rock body, decision tree, random forest, and SVM, and select the highest lithology recognition accuracy. To meet the recognition accuracy, efficiency and other requirements to provide an important basis for the automatic interpretation of logging data and computerized stratigraphic self-recognition is of great significance.
The TZA well is located in the Tarim Basin. Its location is Devonian Donghetang Formation, which contains three lithologies including fine sandstone.
Experimentally measured.
In order to explore the effect of different algorithms on the accuracy, the TZ4 wells were tested using a 50% training ratio and eight logs. The amount of training data on the total dataset has a large impact on the training accuracy. Five sets of scenarios were used to investigate the effect of different training ratios on the prediction results. A decision tree algorithm and eight log data were used to vary the ratio of training data.
Each case was tested 10 times and averaged to minimize error. Typical results based on different training set volumes found that by training only 30% of the total dataset, the accuracy can reach more than 80%. In addition to explore the effect of different logging parameters on accuracy, multiple sets of scenarios were analyzed using a decision tree algorithm and a 50% training ratio.
By adding the number of logging parameters to test the effect of parameters on the model, it can be obtained that through the increase of logging parameters, the accuracy of the model is improved between, and the final accuracy can reach more than 90%.
| # | number | name | type |
| 1 | XDA14000000 | Strategy Priority Research Program (Category A) of Chinese Academy of Sciences | |
| 2 | XDA14040000 | Key technologies for ultra-deep guided drilling | Strategy Priority Research Program (Category A) of Chinese Academy of Sciences |
This work is licensed under a
Creative
Commons Attribution 4.0 International License.
CSTR:11738.11.NCDC.XDA14.DB6395.2024
10.12072/ncdc.xda14.db6395.2024