Abstract: Total organic carbon (TOC) content is an important parameter reflecting the amount of shale reservoir resources, and nuclear logging data can provide a lot of information about the abundance of organic matter in source rocks. In this paper, the correlations between the response values of various nuclear logging curves and TOC in Wells X1, X2 and X3 in the southeastern Sichuan basin are analyzed. Considering the geological factors that affect the content of TOC, the single nuclear logging curve which is sensitive to TOC content and the combination of the nuclear logging curves that can reflect the genesis of TOC are selected. Then a TOC content prediction model with nuclear logging curves suitable for this area is established through BP neural network. Finally, the model was applied to Well X4 in the southeastern Sichuan basin. Compared with the TOC contents obtained from the core analysis of 113 core samples, the mean relative error of the model prediction results is 0.41, indicating that the prediction accuracy of the new model is high, which can meet the actual production demands in the area. Then the model was applied to the marine shale reservoirs in W and Y blocks of the southern Sichuan basin and good effects have been gained, which can prove the good operability, wide versatility and high evaluation accuracy of the model. The new method provides an effective technological mean for the evaluation of total organic carbon content in marine shale reservoirs.

Key words: nuclear logging, marine facies, shale reservoir, total organic carbon content, combined sensitive parameter, BP neural network