煤矿瓦斯事故破坏性强、危害范围大，回采工作面瓦斯涌出量及其回风瓦斯体积分数预测可以为制定瓦斯治理措施、预防瓦斯事故提供重要依据。为解决瓦斯涌出量及其回风瓦斯体积分数受多因素影响、数据波动大而难以准确预测的问题，在灰色预测与BP神经网络的基础上引入麻雀搜索算法（SSA），建立了一种ISSA-GM-BP模型用于回采工作面瓦斯涌出量及其回风瓦斯体积分数预测；该模型利用Chebyshev 混沌映射、动态惯性权重、Lévy飞行策略算法对SSA进行改进，在灰色预测中引入动态生成系数$ alpha $建立动态灰色GM(1，1，$ alpha $)模型并与BP神经网络组合使用，再通过改进SSA对组合模型进行优化，利用该模型对某矿山回采工作面瓦斯涌出量及其回风瓦斯体积分数进行预测，并与SSA-BP神经网络、BP神经网络的预测结果作出对比分析。结果表明：在回采工作面瓦斯涌出量、工作面回风瓦斯体积分数2方面，ISSA-GM-BP模型预测结果与实测值之间平均相对误差分别为2.95%、2.65%，SSA-BP 神经网络的平均相对误差分别为9.50%、8.00%，BP 神经网络的平均相对误差分别为12.49%、9.76%，且ISSA-GM-BP模型的决定系数为0.9609、0.9587，预测值完全符合实际回采工作面瓦斯涌出量和工作面回风瓦斯体积分数的变化趋势，在预测精确性与适应性方面具有显著优势。

Coal mine gas accidents are highly destructive and have a wide range of hazards. Predicting the amount of gas emitted from the mining face and the volume fraction of gas in the return air roadway can provide important basis for formulating gas control measures and preventing gas accidents. To solve the problem of mining face gas emission and gas volume fraction in mining face return air being affected by multiple factors and difficult to accurately predict due to large data fluctuations, we introduce the sparrow search algorithm (SSA) based on grey prediction and BP neural network, and establish an ISSA-GM-BP model for predicting the mining face gas emission and gas volume fraction in mining face return air roadway. This model utilizes Chebyshev chaotic mapping, dynamic inertia weight, and Lévy flight strategy algorithm to improve SSA. In grey prediction, a dynamic grey GM (1,1,$ alpha $) model is established by introducing dynamic generation coefficients and combined with BP neural network. The combined model is then optimized by improving SSA. Use this model to predict the mining face gas emission and gas volume fraction in mining face return air roadway, and compare and analyze the prediction results with SSA-BP neural network and BP neural network. The results showed that in terms of mining face gas emission and gas volume fraction in mining face return air roadway, the average relative errors between the prediction results of the ISSA-GM-BP model and the measured values were 2.95% and 2.65%, respectively. The average relative errors of the SSA-BP neural network were 9.50% and 8.00%, respectively. The average relative errors of the BP neural network were 12.49% and 9.76%, respectively. The determination coefficients of the ISSA-GM-BP model were 0.960 9 and 0.958 7, respectively. The predicted values fully conform to the trend of actual mining face gas emission and gas volume fraction in mining face return air roadway, and have significant advantages in prediction accuracy and adaptability.