Identification of thin-layer coal texture using geophysical logging data: Investigation by Wavelet Transform and Linear Discrimination Analysis

Abstract

The well log inversion has been widely applied to forecast coal texture distribution due to its effectiveness and low investment. However, most research missed the low resolution of well logging data caused by the “boundary effect”, which make it difficult to identify thin layers. Here a method to improve the vertical resolution of logging curves by Wavelet Transform was applied. The original logging curve was decomposed by two-level wavelet decomposition using sym 8 mother wavelet, and then the high-frequency wavelet coefficient (d2) was weighted to reconstruct a new log curve to enhance the resolution of the thin-layer. Based on 160 reconstructed training logging datasets, the Linear Discrimination Analysis was used to establish a quantitative coal texture identification model by reducing the dimensionality of the data points and projecting them into a two-dimensional space. The validation results of coal texture are in good agreement with the coal core. Compared with the non-reconstructed curve, the recognition accuracy is greatly improved, especially for layers with a thickness < 20 cm. The combination of Wavelet Transform and Linear Discrimination Analysis provides a new way to predict coal texture using geophysical logging data. This method has been successfully applied to multiple (>20 layers) and thin (generally <2 m in thickness) seams in the western Guizhou province of China, with recognition accuracy >85%.