Multi-scale classification and evaluation of shale reservoirs and ‘sweet spot’ prediction of the second and third members of the Qingshankou Formation in the Songliao Basin based on machine learning

• Published in: Journal of petroleum science & engineering 2022-09, Vol.216, p.110678, Article 110678
• Main Authors: Niu, Da-Ming, Li, Yi-Lin, Zhang, Yun-Feng, Sun, Ping-Chang, Wu, Hai-Guang, Fu, Hang, Wang, Ze-Qiang
• Format: Article
• Language: English
• Subjects: Machine learning; Multi-scale; Shale reservoirs; ‘Sweet spot’ area
• ISSN: 0920-4105; 1873-4715
• DOI: 10.1016/j.petrol.2022.110678

Owing to the unique structure of shale reservoirs intercalated with thin siltstone in the second and third members of the Qingshankou Formation (K1qn2+3) in the north of the Central Depression of the Songliao Basin, it is difficult to objectively predict and evaluate multi-scale (i.e., from the micro to macro scale) physical properties of the reservoir and the ‘sweet spot’ area (the area with the best reservoir quality). Using machine learning, we present herein a new machine learning framework (GCA-CE-MGPK) specific to reservoirs for studying the shale reservoirs in K1qn2+3. According to the results of a high-pressure mercury-injection experiment, organic geochemical analysis and scanning electron microscopy. Through grey correlation analysis, clustering ensemble and the Kriging model combined with macro geological parameters, an efficient, accurate and objective multi-scale evaluation of the shale reservoirs and the prediction of the ‘sweet spot’ area were realised. This method can be used to overcome difficulty in parameter selection, time-consuming classification of a large amount of data and difficulty in macro-scale reservoir prediction in the absence of seismic data with an average accuracy of 82.4%. The reservoir prediction results showed that the Class-I reservoir, the ‘sweet spots’ area, is mainly distributed in the north of the study area with an area of 3.15 × 108 m2; the Class-II reservoir is mainly distributed in the north of the study area with an area of 9.88 × 108 m2; the Class-III shale reservoir is the most widely distributed reservoir type with an area of 4.90 × 109 m2. Overall, compared with K1qn1, K1qn2+3 offers more realistic oil and gas exploration potential and advantages. •A new machine learning framework of shale reservoir multi-scale evaluation and ‘sweet-spot’ prediction is established.•The classification evaluation standard of shale reservoir in the K1qn2+3 is established.•The ‘sweet spot’ distribution of shale reservoir in the K1qn2+3 is revealed.•The advantages of shale reservoir in the K1qn2+3 are evaluated compared with the K1qn1.