Abstract: According to serious problems such as gas accumulation and gas concentration exceeding limits on the working face underground coal mines, Introducing wavelet packet and neural network model in prediction of mine gas emission. First, The collected data decomposition, reconstruction and extracted feature vectors by wavelet packet transform, Then input to generation algorithm based on dynamic node RBF neural network model training and learning, While using the simplified model delete policy. Finally, By joint time-frequency simulation, The results show that WP-IRBF model is far superior to QPSO-RBF model in prediction accuracy and training error and is a very suitable for effective method for prediction of coal mine gas quantity.