Shale oil content evaluation and sweet spot prediction based on convolutional neural network

Shale oil content prediction is currently the focus of exploration. Traditional prediction methods are less accurate, and a large number of analytical tests can increase the cost of exploration. Therefore, new methods need to be introduced to accurately predict sweet spot intervals. In this study, a quick and novel convolutional neural network (CNN) method to predict oil content parameters (TOC, S1 and S2) is provided based on well logs (SP, GR, CAL, RT, and AC) in the Dongpu Depression, Bohai Bay basin, China. The results showed that the CNN model showed high accuracy in predicting TOC (R2 = 0.73) than the Delta logR method (R2 = 0.132). The validation sets for S1 (R2 = 0.72) and S2 (R2 = 0.67) also show high prediction accuracy. In addition, Well Wen 318 (located in the same formation) was selected as a blind well to test the performance of the model. The micromigration evaluation method is based on the principle of mass balance, which reflects the changes of the original hydrocarbon generation potential (HGPO) and the real hydrocarbon generation potential (HGPR). Combined with the established CNN model and the micromigration evaluation method, we predicted and evaluated the oil content in lower Mbr 3 of the Shahejie Fm and divided four sweet spots. The differential enrichment of shale oil is controlled by micromigration. Samples with high saturated/aromatic ratio and small main peak carbon of saturated hydrocarbon chromatography are more likely to undergo micromigration, which in turn affects the oil content of shale. Shales with high organic matter abundance and good pore connectivity are more susceptible to micromigration, leading to the differential enrichment of shale oil. This study provides a new idea for evaluating hydrocarbon micromigration of shale oil and predicting favorable zones and the grading evaluation of oil content in petroliferous basins around the world.