刘晓，程明果，卫少任，等. 基于FI筛选的DNN瓦斯抽采半径快速确定方法［J］. 河南理工大学学报（自然科学版）， doi:10.16186/j.cnki.1673-9787.2026030019

LIU X, CHENG M G, WEI S R, et al. A rapid determination method for gas drainage radius based on FI screening and DNN［J］. Journal of Henan Polytechnic University（Natural Science），doi: 10.16186/j.cnki.1673-9787.2026030019

基于FI筛选的DNN瓦斯抽采半径快速确定方法

刘晓¹，程明果¹，卫少任¹，杨程涛²，倪小明¹，张宇博¹，王永龙¹，
李立康¹，韩磊¹，李博¹

（1.河南理工大学 能源科学与工程学院，河南 焦作 454003；2. 河南能源化工集团研究总院，河南 郑州 450046）

摘要: [目的] 为解决现行标准下的瓦斯抽采参数确定方法普遍存在井下钻孔工程量大、测试周期长、数据易受人为影响等问题，开展基于FI筛选的DNN瓦斯抽采半径快速确定方法研究。 [方法] 提出采用特征重要性-深度神经网络的瓦斯抽采半径快速确定方法。首先，将瓦斯抽采半径作为目标值，通过基于随机森林算法的特征重要性计算方法（FI）计算各影响因素在瓦斯抽采半径确定过程中的权重大小，筛选确定主控因素，将其作为特征值；其次，采用抽采半径历史数据与COMSOL模拟数据相融合的方法，建立神经网络训练数据集，以特征值为训练样本，以目标值为训练结果。 [结果] 通过神经网络模型的学习、迭代、优化建立瓦斯抽采半径计算模型并开发相应的可视化软件；通过现场实际抽采半径测试数据对比考查，与线性回归和支持向量回归模型相比，DNN模型在测试集上的平均绝对误差分别降低34.7%和48.3%，均方根误差分别降低22.4%和26.8%，决定系数提高至0.988。 [结论] 始瓦斯压力、煤层厚度、渗透率、抽采时间、钻孔半径等5个因素为主控因素，以该模型为核心的瓦斯抽采半径快速确定方法在瓦斯抽采半径确定过程中具有高效性与便捷性，为瓦斯抽采半径的快速、准确确定提供了新的技术思路。

关键词: 瓦斯抽采半径；影响因素；特征重要性；神经网络；数值模拟

中图分类号：TD712

doi: 10.16186/j.cnki.1673-9787.2026030019

基金项目: 国家自然科学基金资助项目（42572222，42502169）；河南省重点研发专项项目（241111321000）

收稿日期：2026-03-16

修回日期：2026-08-27

网络首发日期：2026-09-01

A rapid determination method for gas drainage radius based on FI screening and DNN

Liu Xiao¹, Cheng Mingguo¹, Wei Shaoren¹, Yang Chengtao², Ni Xiaoming¹, Zhang Yubo¹, Wang Yonglong¹, Li Likang¹, Han Lei¹, Li Bo¹

(1.School of Energy Science and Engineering, Henan Polytechnic University, Jiaozuo 454003, Henan, China; 2.Research Institute of Henan Energy and Chemical Industry Group, Zhengzhou 450046, Henan, China.)

Abstract: [Objective] The gas extraction radius was considered a core parameter for mine safety and efficient gas extraction. It was influenced by multiple factors. Under current standards, direct measurement methods, such as pressure method and flow method, were used. These methods involved large underground drilling work, long testing periods, and data prone to human influence. [Methods] A rapid determination method of the gas extraction radius was proposed based on feature importance and deep neural network (DNN). The gas extraction radius was set as the target value. Feature importance (FI) was calculated using a random forest algorithm to evaluate the weights of influencing factors. The main controlling factors were selected as features. Historical extraction radius data were combined with COMSOL simulation data to build a neural network training dataset. Features were used as input samples, and the target value was used as output. [Results] A calculation model of the gas extraction radius was established through DNN learning, iteration, and optimization. Corresponding visualization software was developed. The model was validated using field test data. Compared with linear regression and support vector regression models, the DNN model reduced the mean absolute error by 34.7% and 48.3%, and the root mean square error by 22.4% and 26.8%, respectively. The coefficient of determination reached 0.988. [Conclusion] Five factors, including original gas pressure, coal seam thickness, permeability, extraction time, and borehole radius, were identified as main controlling factors. The proposed rapid determination method, based on this model, demonstrated high efficiency and convenience. A new technical approach was provided for quick and accurate determination of the gas extraction radius.

Key words: gas extraction radius; influencing factors; feature importance; neural network; numerical simulation

CLC: