To accurately assess the gas safety situation in coal mines, in view of the fact that the prediction of gas concentration in the working face is greatly influenced by multiple interrelated factors and has a lot of data noise, this paper proposes a gas concentration prediction method based on the correlation of environmental parameters in the coal mine working face. Firstly, in response to the issue that the gas concentration is influenced by multiple factors, this paper adopts the Bayesian Optimization XGBoost (BOXGB) algorithm to analyze the correlation of environmental parameters in the working face. By intelligently searching for the optimal combination of hyperparameters, the prediction performance of the model is enhanced. Secondly, considering the problem that both the gas concentration data and the associated factor data contain a large amount of noise, a gas concentration prediction algorithm based on SKNet Multi-scale Residual Network Informer (SMSRN-Informer) is proposed. By adding a multi-scale residual network structure with dynamic kernel selection to the Informer model, different scale data features can be extracted, effectively suppressing the influence of data noise. This paper selected GRU, Linear, iTransformer, etc., as comparison models. The R \(^2\) value was increased by approximately 5.09 \(\%\) , indicating that the model proposed in this paper significantly outperforms the comparison models in terms of gas concentration prediction accuracy and can more accurately analyze the impact of different factors on the model.