| 本文已被：浏览次 下载次 |
码上扫一扫！ |
|
|
|
| A modeling paradigm for unconventional reservoir parameters based on an interpretable neural network |
|
ZHANG Fengjiao1,2,3,4, DENG Shaogui1,2,3,4, CHEN Yan5, GAO Beibei1,2,3,4
|
|
(1.State Key Laboratory of Deep Oil and Gas, Qingdao 266580, China;2.Engineering Research Center of Deep Oil & Gas Exploration Technology Equipment, Qingdao 266580, China;3.Shandong Provincial Key Laboratory of Deep Oil and Gas, Qingdao 266580, China;4.Key Laboratory of Deep Oil and Gas, Ministry of Education, Qingdao 266580, China;5.PetroChina Qinghai Oilfield Company, Dunhuang 736202, China)
|
| Abstract: |
| Machine learning plays an increasingly important role in modeling unconventional reservoir parameters. However, its "black box" structure often results in a lack of interpretability for the prediction results.To address this problem, an interpretable neural network(INN)-based modeling paradigm is proposed for reservoir parameters.The model is based on a generalized network structure, and the interpretability is demonstrated to some extent by visualizing the sub-network basis functions and their coefficients. The brittleness index modeling of the shale oil reservoir from Qing1 Member of Qingshankou Formation in Songliao Basin is used for model validation. The results show that the proposed INN model is not only interpretable, but also outperforms traditional machine learning models such as ELM, SVM, and BP with the smallest root means square error(5.2%-6.31%) and the highest correlation coefficient (0.75-0.87). |
| Key words: reservoir parameter neural network interpretability prediction |
|
|