Abstract:The prediction of gas emission from working face is very important to the mine safety production.Different-source prediction was used to divide gas emission, according to the gas emission is influenced by different factors,partial least square method (PLS) was used that of cross validity analysis,to determine the principal component number,the principal component was regarded as neural network input layer and correlation model was set up.The result showed that this method not only avoids the interference between the various related factors,solves the problem of multiple correlation among various factors, reduces the variable dimension,but also by nonlinear mapping capability of BP neural network and adaptive learning ability,improve the prediction accuracy and stability.