Coal structures are widely regarded as a critical influencing factor for the dynamic behaviors of CH4 migration in coalbed methane (CBM) reservoir. In this paper, geophysical logging data were analyzed to explore the logging response characteristics of coal structures, and their application on identification of coal structures by using the machine learning methods. Meanwhile, the correlations between coal structures and gas-bearing properties were revealed. The results show that with the increase in coal deformation intensities, acoustic transit time, caliper logging, compensated neutron, and natural gamma values positively increase and that for density logging and lateral resistivity show a negative correlation. The multi-logging parameter identification models of coal structures were constructed by using random forest algorithm, radial basis function neural network, and long short-term memory neural network, with their accuracy reaching to 96.67%, 93.33%, and 91.67%, respectively. Based on the identification results of RFA model, the highest distribution percentages of cataclastic coal are 50.2%, which is controlled by tectonic activities and buried depth. The origins of gases are mainly thermogenic gases whose average value of δ13C(CH4) is −37.51‰. The gas content in granulated coal is smaller than 12 cm3/g, but it is higher than 15 cm3/g in cataclastic coal, resulting the higher gas saturation of cataclastic coal. The average extension length of artificial fractures in cataclastic coals is nearly two times as long as in granulated coals. It is suggested that cataclastic coal zone is the favorable area for CBM development.