Abstract: Due to the complex and changing environmental characteristics of mines, the accuracy of underground wireless channel modeling by traditional ray tracing is low.In this paper, by analyzing and evaluating the characteristics of machine learning algorithm and its combination, the optimal channel modeling method is selected.Machine learning algorithm is introduced to learn the scene features, and then more accurate modeling is realized.The application of BP neural network, genetic algorithm and support vector machine in the direction of coal mine channel modeling is studied.Therefore, a field intensity prediction model combining ray tracing method and GA_BP is established in this paper.After that, the least square support vector machine method is proposed to build the prediction model.The field strength is predicted by taking the actual data of the roadway as the training sample of the algorithm.The characteristics of each algorithm and the influence of the parameters in the algorithm on the prediction result are analyzed in detail.Compared with the simulation results, the error between the field strength prediction result and the measured data is -1.206 dbm, the hybrid model is higher prediction accuracy.