We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
基于机器学习模型的斜井坍塌压力预测方法.
- Authors
马天寿; 张东洋; 杨 赟; 陈颖杰
- Abstract
Collapse pressure is an important basic parameter for optimizing drilling fluid density and maintaining wellbore stability, and plays an important role in ensuring the safe and efficient drilling of oil and gas wells. The traditional collapse pressure prediction methods have a complicated calculation process and low prediction accuracy. In order to solve these problems, this paper establishes a machine learning prediction method of collapse pressure for inclined wells by using four machine learning models such as random forest and polynomial regression. Then, the training samples are generated by using random parameter sampling and the traditional analytic model. In addition, an optimal model is selected, and the number of training samples, the structure of neural networks, and the hyperparameters of the model are optimized. Finally, the reliability and accuracy of this prediction method is verified by taking Well Z-1 as an example. And the following research results are obtained. First, the optimized multilayer perceptron model has the best prediction performance, and presents a better prediction capacity in the verification set and test set. Second, compared with the logging interpretation results, the collapse pressure in the vertical, inclined and horizontal intervals of Well Z-1 predicted by this model has an average absolute error lower than 0.007 3 g/cm³, an root mean square error lower than 0.013 8 g/cm³, an average absolute percentage error lower than 0.771 1%, and a coefficient of determination higher than 0.950 5, indicating that this model can accurately predict the collapse pressure profiles of different well intervals. Third, the maximum relative error of the hemispherical projection of the collapse pressure at three depths of Well Z-1 is lower than -1.97% and the determination coefficient is higher than 0.987 6, indicating that this model can accurately predict the collapse pressure of an inclined well at any depth. In conclusion, this method can accurately predict the collapse pressure of any inclined well within an given parameter range, and can capture the change laws of collapse pressure with well inclination and orientation, which provides an important support for maintaining the wellbore stability of inclined and horizontal wells and ensuring the safe and efficient development of oil and gas.
- Subjects
HORIZONTAL wells; MACHINE learning; GAS well drilling; OIL well drilling; STANDARD deviations; DRILLING fluids
- Publication
Natural Gas Industry, 2023, Vol 43, Issue 9, p119
- ISSN
1000-0976
- Publication type
Article
- DOI
10.3787/j.issn.1000-0976.2023.09.012