|
The significant heterogeneity of Xujiahe reservoirs always makes a real challenge for natural gas exploration due to its deep burial depth and intense diagenetic reformation. Traditionally, diagenetic facies is an effective method to evaluate the vertical heterogeneity of tight sandstone. However, qualitative classification caused by incomplete workflow could reduce the fidelity and reliability of diagenetic facies recognition results. In this study, we designed a complete technical workflow combined with quantitative classification and automatic prediction exploiting machine-learning methods. Firstly, by applying the principal component analysis (PCA) algorithm, we reduced the data dimension and extracted four principal components from our dataset collected from Xujiahe reservoir that originally contains fifteen petrological parameters. These four principal components will be fed into hierarchical clustering analysis (HCA). Then we analysis our dataset using the pedigree chart of HCA which integrating the petrographic images and XRD data. Results show that the Xujiahe sandstone can be divided into five diagenetic facies. Secondly, we calibrated the response characteristics of conventional well-logging by coring interval technique, and chose six sensitive logging parameters for diagenetic facies prediction. Based on the diagenetic-electrical data, the prediction model of diagenetic facies is established using fisher discriminant analysis (FDA), which can automatically recognize diagenetic facies. Test results show that the coincidence rates of four validation methods were all over 87%. Finally, we explained the vertical heterogeneity of tight sandstone by results of quantitative identification of diagenetic facies, and proposed the CDF parameter to predict favorable areas of petroleum exploration, which aims to improve exploration efficiency greatly.
|
