关键词: 总有机碳含量, 烃源岩, 多元回归模型, BP神经网络模型, 曲线叠合模型, 文昌组, 陆丰凹陷

Abstract: There is a certain response relationship between well logging parameters and total organic carbon (TOC)content of source rocks,so TOC content can be predicted by well logging parameters. The multi-variate regression model,BP artificial neural network model and curve overlapping model were established between TOC and conventional well log data,including resistivity log,acoustic log,neutron porosity log,gamma-ray log and density log of Wenchang Formation source rocks in Lufeng Sag. The differences of the three models in TOC prediction effect were discussed. The results show that the multi-variate regression model has better TOC prediction effect for semi-deep lake facies and delta front facies,but worse for shore-shallow lake facies. The prediction effect of BP artificial neural network model is better than that of multi-variate regression model,while the curve overlapping model has worse prediction effect. In practical application,the BP artificial neural network model is suitable for areas where logging parameters and TOC are difficult to express with explicit functions and have a large enough data volume,the multi-variate regression model is suitable for areas where logging parameters are significantly correlated with TOC,while the curve overlapping model is suitable for areas where gamma curve responds significantly to clay and organic matter content,and the target curve can be well superposed in non-hydrocarbon source rock beds. Through the analysis of the above models,it can be applied to other sub-sags in the depression.

Key words: TOC content, source rocks, multi-variate regression model, BP artificial neural network model, curve overlapping model, Wenchang Formation, Lufeng Sag