Methods for predicting the thickness of coal seams based on seismic attribute optimization and machine learning

Accurate location of coal seams is the key technology of unattended mining and predicting the thickness of coal seams is an important content in the seismic interpretation of coalfields. This study constructed a forward model involving wedge-shaped coal seams by referencing the actual thickness and physical properties of strata. Moreover, this study compared and analyzed the effects of signal-to-noise ratio (SNR) and regression methods on the prediction of the coal seams’ thickness through the forward modeling of seismic profiles and the extraction and optimization of seismic attributes. The results of this study are as follows: (1) Some seismic attributes were strongly correlated with, and can be used to predict, the thickness of coal seams; (2) The information redundancy among seismic attributes cannot be ignored. However, there was no essential difference between the seismic attributes optimized using principal component analysis (PCA) and multi-dimensional scaling (MDS); (3) When the SNR was low (10 dB), the root-mean-square (RMS) error of the prediction results of different algorithms was in the order of random forest regression (RFR, 1.07)< support vector machine regression (SVR, 1.15)< multivariate linear regression (MLR, 1.84); (4) When the SNR was high (25 dB), the RMS error of the prediction results of these algorithms was in the order of SVR (0.05)<RFR (0.11)<MLR (0.20); (5) The SNR of the input data had significant impacts on the prediction of the coal seams’ thickness, and a higher SNR corresponded to better prediction performance. Coal thickness prediction method based on seismic attribute optimization and SVR is an effective way to realize high-precision interpretations of the coal seams’ thickness.