Coal mine gas calamity is the most dangerous accidents in production, which brought serious danger for safety production. The forecast of gas calamity is particularly important. Based on the previous gas forecast method cannot solve the problem of multi dimensional disasters. A new adaptive prediction method was proposed for the prediction of multi dimensional by combining the multidimensional gas forecast model and regression neural network. Experimental results showed that the autoregressive neural network prediction had highly precision and strongly adaptability, which could make a good effect on gas disaster forecast.