Stratigraphic automatic correlation using SegNet semantic segmentation model
Yue Hong Dai, Xuri Huang, Haojie Liu, Hongwei Yang, Guohua Wei, Ning Lu, Zhiying Han, Haibo Song · 2021
A multi-task encoder-decoder based on SegNet architecture is proposed for automatic stratigraphic correlation in this work. In order to have higher resolution correlation, logs and their wavelet transformed results are combined to form the training datasets. In addition, two types of loss functions for the SegNet are used to achieve high-resolution results. By applying this method to a field in Shengli Oilfield, the result demonstrates that this network can obtain accurate stratigraphic correlation and is significantly efficient compared to the conventional manual method. Using the correlated results and combined with dip attribute from seismic data, an isochronal stratigraphic framework is built for geological modeling and study. This work demonstrates that SegNet can be a reliable automatic well log correlation technique with high efficiency and accuracy.