| Abstract |
CO2 geological storage technology is considered as an effective measure to reduce atmospheric CO2 concentration and is one of the key technologies to achieve the goal of carbon neutrality in China. Reservoir permeability is an important parameter to evaluate the permeability of reservoir, which restricts the prediction of CO2 injection capacity. High precision prediction of it is one of the main tasks of CO2 geological storage reservoir evaluation. The Upper Paleozoic reservoir in the Ordos Basin are favorable reservoirs for CO2 storage, but they are relatively dense and highly heterogeneous, making it difficult to accurately characterize their pore throat characteristics and seepage law. The current permeability models are mostly set up for conventional reservoirs, and there is a significant deviation between the predicted and measured permeability of tight sandstone reservoirs, resulting in poor application effectiveness. However, the reservoir permeability model constructed by commonly used machine learning algorithms such as BP neural network can not simultaneously balance prediction accuracy and generalization performance due to the disadvantages of each algorithm. How to dig the permeability information hidden in the conventional logging data using cutting-edge mathematical algorithms, and establishing the permeability model with high prediction accuracy and strong generalization performance is to lay the technical support for the whole evaluation system of CO2 geological storage. In this paper, the random forest algorithm is used to build a permeability prediction model for CO2 storage reservoir. The goodness of fit, mean absolute error and mean square error of the model on the training set are 0.7524, 0.1813 and 0.1411, respectively. The goodness of fit, mean absolute error and mean square error of the model on the test set are 0.7265, 0.1947 and 0.1389, respectively. Compared with the BP-ANN permeability model, the RF-based permeability model has higher prediction accuracy and stronger generalization performance in the training set and test set. A reservoir permeability model based on random forest algorithm provides a reference for permeability prediction of CO2 storage reservoir. |
| Authors |
Ping Lu , Zunsheng Jiao  , Jing-zhe Guo  , Jianhua Zhao , Jie Liu , Guilin Yang , Guihao Zhong , Furong Tan , Yong Bai , Yuyu Hao
|
| Journal Info |
Springer Nature | Proceedings of the International Field Exploration and Development Conference 2023 , pages: 533 - 547
|
| Publication Date |
1/1/2024 |
| ISSN |
1866-8755 |
Type |
book-chapter |
| Open Access |
closed
|
| DOI |
https://doi.org/10.1007/978-981-97-0268-8_41 |
| Keywords |
Geological Modeling (Score: 0.51569)
|
