ABSTRACT
Machine learning plays an important role in the field of geology, especially in the process of oil and gas exploration. TOC content in kerogen is an important index to determine the quality of source rock, which can affect the optimization of favorable reservoir areas. At present, the prediction accuracy of TOC content in highly heterogeneous source rocks by various seismic interpretation and logging parameter regression methods is relatively low, which seriously restricts the progress of oil and gas exploration. Based on this, this paper takes Wulanhua sag in Erlian Basin as an example, the TOC content of source rocks is predicted by ΔLogR and artificial neural network (ANN) simulation, and the differences are compared. The results show that: (1) the measured TOC values of source rocks in Wulanhua sag are distributed in 0.2%~5.3%, with an average of 2.0%, and the characteristics of source rocks are good. (2) The accuracy rates of TOC content predicted by the two methods are 79% and 90% respectively. The ANN simulation method is feasible to simulate TOC content. (3) ANN simulation method is not limited by geological conditions, and the model will take into account the effects of various parameters. In general, the TOC prediction method of lacustrine source rocks based on machine learning has a broad application prospect for source rock evaluation.
