基于贝叶斯正则化改进BP神经网络的页岩气有机碳含量预测模型

页岩气总有机碳(TOC)含量是评价岩性气藏的关键指标，受复杂地质及岩芯采集等多种因素的影响，常规室内测试分析获得的TOC含量的数据有限且结果有失准确。为合理准确预测页岩气TOC含量，本文首先通过对页岩气储层TOC含量测井资料综合分析选取8条测井曲线，并结合主成分分析法(Principal Component Analysis，PCA)提取四个主成分；其次基于贝叶斯正则化(Bayesian Regularization)改进的BP神经网络方法建立页岩气TOC含量预测的BR-BP模型；最后利用该模型对研究区A区页岩气TOC含量进行预测，并与常规的LM-BP神经网络模型的预测结果进行对比。结果表明：BR-BP模型有较强的非线性拟合能力，能够真实地反映出页岩气TOC含量与各测井参数之间的非线性关系，其模型预测结果与实际值基本吻合，与常规的LM-BP神经网络模型相比，其数据敏感性增强，预测精度有所提高，该研究方法具有一定的理论意义和参考价值，为我国TOC含量预测提供了一种新的技术方法和手段。

A prediction model for shale gas organic carbon content based on improved BP neural network using Bayesian regularization

Total organic carbon (TOC) content in shale gas is a key indicator for evaluating lithologic gas reservoirs. The data of this parameter from conventional laboratory analysis are limited in amount with poor accuracy owing to many factors such as complex geology and core recovery. This work attempted to solve this problem. We selects 8 logging curves by comprehensive analysis of logging data of TOC content in shale gas reservoirs and four principal components were extracted by Principal Component Analysis (PCA) from these curves. Then, a BR-BP model was established to predict TOC content in shale gas based on improved BP neural network with Bayesian regularization. Finally, the model is used to predict the TOC content of shale gas in the area A under the study, and compared with the prediction results by the conventional LM-BP neural network model. The results show that the BR-BP model has strong nonlinear fitting ability which can truly reflect the nonlinear relationship between the TOC content of shale gas and each logging parameter and the model prediction largely accords with the actual values. Compared with the conventional LM-BP neural network, the data sensitivity of this model is enhanced and the prediction accuracy is improved. This research method has certain theoretical significance and reference value, which provides a new technique for the prediction of TOC content in hydrocarbon exploration.