基于PCA-BP神经网络的TOC测井评价方法研究OACSTPCD

The organic carbon content is the main parameter to evaluate the potential of hydrocarbon source rocks.The commonly used TOC logging inversion model is difficult to deeply analyze the complex collinearity relationship between logging curves,which restricts the comprehensive evaluation effect of multi-dimensional logging information.An intelligent prediction method of organic carbon content based on principal component analysis and back propagation(PCA-BP)neural network is established,based on the pyrolysis experimental results and conventional logging curves of Triassic Baijiantan formation mudstone in Mahu sag.The method is based on the weighted average of sensitive logging curves and TOC test results as the original data set.Firstly,the variance inflation factor is used to detect the collinearity between the logging curves.Then,the principal component analysis(PCA)technology is used to decollinearity and reduce the dimension of the original data set,and two principal components are determined.Finally,combined with neutron,gamma ray,density and acoustic logging curve values,a three-layer back propagation(BP)neural network prediction model with six input nodes is established to evaluate the source rock potential of the Triassic Baijiantan formation in the study area.The prediction results of the cumulative 410 m section of three coring wells show that the determination coefficient of the model is as high as 0.879,the average absolute error and mean square error of the prediction results are 0.220 and 0.107,respectively,and the mean relative deviation is 16.1%.The research results provide a reliable reference for the optimization of exploration domain in Junggar basin.