Fine detection of water-bearing collapse column based on information fusion of seismic exploration and TEM

Abstract

When exploring water-bearing collapse columns (WBCCs) in coal mines, a high-resolution seismic method can roughly locate the collapse column manually but cannot identify the collapse column water abundance. Comparatively, the transient electromagnetic method (TEM) is sensitive to the low resistivity of WBCCs, but it is difficult to delineate the boundary due to the volume effect. Therefore, the fusion processing combined with seismic wave field and transient electromagnetic field is necessary for automatic fine detection. A multifield and multiattribute information fusion model based on a neural network was constructed. The geometry, kinematics, dynamics, and statistical features contained in seismic data are often described by seismic attributes, and the resistivity of the water in TEM is relatively low. Therefore, we first extracted the TEM inversion resistivity and optimized eight seismic attributes sensitive to the target geologic structure. Then, a backpropagation neural network (BPNN) optimized using particle swarm optimization (PSO-BPNN) was selected as the fusion method for model test and engineering applications. By the test model, we verified the feasibility and effectiveness of using PSO-BPNN to predict different water-bearing collapse columns in coal mine. In engineering applications, the attributes in borehole were used for PSO-BPNN training, and a good fusion result was achieved. Results show that the nine attributes can be used as the main feature set for the information fusion of the WBCC based on seismic method and TEM. Combined with the features, the PSO-BPNN can detect the boundary and water abundance of the collapse column in coal mine, which can greatly improve detection accuracy.