Abstract: The migration of blasting fume in single-end roadway is analyzed，and the mathematical model of blasting fume diffusion in the casting area is established from the viewpoint of mass conversation.Blasting fume monitoring in the -118 m tunnel in Guilaizhuang gold mine showed that the density of blasting fume at monitoring point increases firstly and then decreases.The process of blasting fume diffusion basically coincides with the exponential decay.The factors for fume-drain time are analyzed from four aspects of blasting，explosive，ventilation condition and tunnel.With BP neural network model，a stable network structure is obtained，regarding 10 groups data of explosive quantity，hole number，distance between fan drum and heading face，distance between monitoring point and heading face，air output，air temperature，roadway temperature，relative humidity of roadway as input，and fume-drain time as output.5-group of experimental input data was introduced into the BP network，obtaining that relative errors between measured results and network-training results are lower than 7%.Better prediction effect is achieved with BP neural network model.Accurate prediction of fume-drain time can not only arrange tunneling in a reasonable way，but also avoid occurrence of blasting fume poisoning incident，which creates significant meanings to safe and high-efficient production of mining.

Key words: Single-end roadway, Blasting fume, Fume-drain time, BP network model, Prediction