A Novel Method for Predicting Free Oil Based on Machine Learning

ABSTRACT China possesses abundant continental shale oil resources. Compared with marine shale in North America, continental shale is characterised by a complex geological background, strong heterogeneity, and significant variability in oil content and mobility. The pyrolysis parameter S 1 (free hydrocarbon content) is often used to characterise the amount of free oil in shale oil resource evaluation, but it is prone to errors such as light hydrocarbon volatilisation. Multi‐temperature pyrolysis technology can subdivide hydrocarbon occurrence states into gaseous hydrocarbons ( S 1‐1 ), medium oil ( S 1‐2 ), adsorbed hydrocarbons ( S 2‐1 ) and kerogen cracking hydrocarbons ( S 2‐2 ). This approach provides a new pathway for the accurate quantification of free oil and adsorbed oil. To accurately evaluate shale oil content, the accurate free oil content (defined as S 1‐1 + S 1‐2 ) was obtained based on multi‐temperature pyrolysis, and corresponding machine‐learning training and test datasets were constructed. Combined with conventional pyrolysis parameters ( S 1 , S 2 , T max and TOC), an S 1 prediction model was developed. Random forest, back propagation neural network, and convolutional neural networks were systematically compared. The results show that the CNN algorithm exhibits the best performance in predicting accurate free oil content. The R 2 values of the training and test sets reach 0.971 and 0.912, respectively, and the average absolute error is less than 6%, which is significantly better than that of traditional empirical models. We predict that the resource grade evaluated after free oil is 11% good, 34% medium, 6% poor and 49% useless. The amount of movable oil resources in Dongpu area is calculated by the Monte Carlo method. The model not only effectively overcomes the deviation caused by light hydrocarbon loss and carry‐over effect, but also has strong generalisation ability. This study provides a novel machine‐learning‐based method for predicting shale oil free oil content, significantly improving the accuracy and efficiency of resource evaluation and providing reliable technical support for shale oil exploration decision‐making and dessert prediction.