An advanced prediction model of shale oil production profile based on source-reservoir assemblages and artificial neural networks. (1st March 2023)

• Record Type: Journal Article
• Authors: Liu, Yazhou; Zeng, Jianhui; Qiao, Juncheng; Yang, Guangqing; Liu, Shu'ning; Cao, Weifu
• Is Part Of: Applied energy. Volume 333(2023)
• Publication Date: 2023-03-01
• Subjects: Sweet spots -- Production evaluation -- Machine learning -- Quantitative evaluation -- Shale oil; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2022.120604
• Languages: English
• ISSNs: 0306-2619

Abstract

Highlights: Propose a universal multidisciplinary data-driven workflow to predict shale oil production through ANN. Comprehensive quantitative evaluation of geological factors in the source-reservoir assemblages. The geometric and harmonic averaging algorithm strategies are recommended to characterize reservoir heterogeneity. Reservoir storage capacity has the most significant contribution to shale oil production. Abstract: Over the past decade, hydrocarbon production from shale oil reservoirs has become increasingly common, and successful shale oil exploration and development depends significantly on the accurate evaluation of the sweet spots. However, different scholars have established different evaluation standards for sweet spots under different geological settings, and it is difficult for these standards to form a universal evaluation standard. The sweet spots should be synonymous with the overall combination of geological, engineering and economic sweet spots. The shale oil production evaluation would be a valid indicator due to the comprehensive combination of the above three perspectives. This paper demonstrates a multidisciplinary data-driven workflow to predict shale oil production through machine learning and quantitative evaluation of geological variables. 48 test sections from 30 exploratory wells in the Lucaogou Formation of the Jimusaer Sag are taken as an example for application demonstration. The proposed 13 geological variables based on source-reservoir assemblage types, source rock quality, reservoir quality, migration dynamics, and conduit conditions realize a systematic and comprehensive characterization of hydrocarbon generation, storage, dynamics, and flow stimulation. Based on the different averaging algorithms in the ANN model, good agreement has been observed between predicted and simulated data for training (R > 0.95) and validation (R > 0.87). Moreover, the geometric and harmonic averaging algorithms are preferentially recommended to characterize reservoir heterogeneity. In unconventional reservoirs, there is no single attribute that can be used to predict success or failure. The training results of the advanced prediction model are better than the other five single reservoir characterization models. On the well J174 dataset, the sweet spot predicted by the model matches well with the oil test results. The increase in liquid hydrocarbon content, mud gas content, TOC and normal faults percentage has positive effects on shale oil production, while the increase in reverse faults percentage has negative effects on shale oil production. This research provides ideas for intelligent prediction of the distribution of sweet spots in unconventional resources, and is also important for the development of intelligent hydrocarbon exploration technology.