Abstract: For the purpose to achieve more accurate prediction of gas concentration around working face through effective analysis of gas measuring data in mines，based on Bayesian network method，chaotic phase space reconstructive technology，and Gaussian process regression model，we studied the prediction method for gas concentration by time series analysis in this paper．By applying Bayesian network method，the sample data sets which have strong relative features with gas concentration around face in time series were extracted to construct the prediction model of multivariate time series．Additionally，the sample space of gas concentration in multivariate time series was reconstructed by using chaotic phase space reconstructive technology．Furthermore，the gas concentration prediction around the face was carried out by using Gaussian process regression model，namely，the prediction values and the corresponding confidence intervals were used to describe the dynamic variation of gas concentration around the face．The case study shows that the prediction results of this method have much higher accuracy than that of univariate method，and the prediction interval is optimal in the same confidence level，which means it can better reflect the dynamic variation of gas concentration around the working face．

Key words: Bayesian network, phase space reconstruction, Gaussian process, multivariate time series, interval prediction, gas concentration