随着开采深度的增加，井下环境变得愈加复杂，为明确各因素对煤层瓦斯含量变化的影响规律。首先对煤层瓦斯含量影响因素采用特征衍生和贝叶斯优化算法（tree-structured parzen estimator, TPE）进行数据处理，然后采用多项式特征衍生、交叉组合特征衍生、分组统计特征衍生出11个新特征，通过决策树与随机森林算法对瓦斯含量的影响因素进行研究。结果表明：在决策树模型下排名前四的影响因素分别是新特征9（泥岩厚度×煤矿构造）、新特征5（砂泥岩比×泥岩厚度）、新特征1（砂泥岩比×砂泥岩比）和新特征3（挥发分×挥发分），得到均方误差（mean square error, MSE）为0.069 2、决定系数（R²）值为0.942 3。在随机森林模型下排名前四的影响因素分别是新特征9（泥岩厚度×煤矿构造）、新特征1（砂泥岩比×砂泥岩比）、砂泥岩比和泥岩厚度，得到的MSE为0.045 83，R²为0.961 8。在经过TPE超参数优化的随机森林模型下排名前四的影响因素分别是新特征9（泥岩厚度×煤矿构造）、新特征6（砂泥岩比×挥发分、砂泥岩比）和新特征7（砂泥岩比×煤矿构造），得到的MSE为0.015 66，R²为0.990 9。在决策树、随机森林和TPE调参随机森林三种算法预测得到的均方误差逐渐减小，R²值逐渐增大。三种模型算法预测中新特征9都是权重最大，是最重要的影响因素；在权重排名2~4的影响因素中都包含了泥岩厚度，可见泥岩厚度对煤层瓦斯含量也是重要影响因素。

With the increase of mining depth, the underground environment becomes increasingly complex. In order to clarify the impact of various factors on the changes in coal seam gas content. Firstly, the factors affecting coal seam gas content were processed using feature derivation and tree-structured parzen estimator (TPE). Then, 11 new features were derived using polynomial feature derivation, cross combination feature derivation, and grouping statistical feature derivation. The factors affecting gas content were studied using decision tree and random forest algorithm. The results show that the top four influencing factors in the decision tree model were new feature 9 (mudstone thickness × coal mine structure), new feature 5 (sandstone to mudstone ratio × mudstone thickness), new feature 1 (sandstone to mudstone ratio × sandstone to mudstone ratio), and new feature 3 (volatile matter × volatile matter), with a mean square error (MSE) of 0.069 2 and a coefficient of determination (R²) value of 0.942 3. The top four influencing factors in the random forest model are new feature 9 (mudstone thickness × coal mine structure), new feature 1 (sandstone to mudstone ratio × sandstone to mudstone ratio), sandstone to mudstone ratio, and mudstone thickness. The MSE is 0.0458 3 and R² is 0.961 8. The top four influencing factors in the TPE hyperparameter optimized random forest model are new feature 9 (mudstone thickness × coal mine structure), new feature 6 (sandstone shale ratio × volatile matter), sandstone shale ratio, and new feature 7 (sandstone shale ratio × coal mine structure). The MSE is 0.015 66 and R² is 0.990 9. The mean square error predicted by the decision tree, random forest, and TPE tuning random forest algorithms gradually decreases, while the R² value gradually increases. Among the three model algorithms, the new feature 9 has the highest weight and is the most important influencing factor; Mudstone thickness is included in the weight ranking of factors 2-4, indicating that shale thickness is also an important influencing factor on coal seam gas content.