Abstract: Changes in coal thickness have an important impact on safe and efficient coal mining. In order to solve the problem of large errors in coal thickness prediction results when the 3D seismic data contains noise, a method in which variable modal decomposition(VMD) and support vector machine(SVM) methods are combined for coal thickness prediction is proposed. Firstly, a coal-thickness wedge model is constructed and seismic forward modeling is performed on it. Based on the condition of thin coal seam thickness, the amplitude attribute and bandwidth attribute have a good positive correlation with the coal thickness, while the instantaneous frequency attribute has a good negative correlation with the coal thickness. With noise applied to forward seismic records, the experimental results show that noise has a greater impact on coal thickness prediction by using seismic attributes. After VMD denoising, based on SVM, the prediction results of actual seismic data are basically consistent with the coal seam information revealed through existing boreholes. The minimum absolute error is only 0.02 m and the maximum absolute error is 0.52 m, showing the feasibility and effectiveness of the coal thickness prediction method. It provides reference for coal thickness inversion in the low SNR area.