SMOGN过采样下导水裂隙带高度的MPSO-BP预测模型  

A SMOGN-based MPSO-BP model to predict the height of a hydraulically conductive fracture zone

作　　者：刘奇 梁智昊[1] 訾建潇 LIU Qi;LIANG Zhihao;ZI Jianxiao(College of Energy and Mining Engineering,Shandong University of Science and Technology,Qingdao 266590,China;State Key Laboratory of Mining Disaster Prevention and Control Co-founded by Shandong Province and the Ministry of Science and Technology,Shandong University of Science and Technology,Qingdao 266590,China;Anhui Provincial Key Laboratory of Building Structure and Underground Engineering,Anhui Jianzhu University,Hefei 230601,China;Feicheng Mining Group Liangbaosi Energy Co.,Ltd.,Jining 272400,China)

机构地区：[1]山东科技大学能源与矿业工程学院,山东青岛266590 [2]山东科技大学矿山灾害预防控制省部共建国家重点实验室培育基地,山东青岛266590 [3]安徽建筑大学建筑结构与地下工程安徽省重点实验室,安徽合肥230601 [4]肥城矿业集团梁宝寺能源有限责任公司,山东济宁272400

出　　处：《煤田地质与勘探》2024年第11期72-85,共14页Coal Geology & Exploration

基　　金：国家自然科学基金项目(51904168);山东省自然科学基金项目(ZR2023ME021);青岛市博士后基金项目(QDBSH20230202050)。

摘　　要：【目的】导水裂隙带高度是顶板(涌)突水、地下水资源流失的重要影响因素之一,是矿井防治水研究的重点。【方法】为了准确地预测煤层顶板导水裂隙带高度,选取开采深度、采高、煤层倾角、工作面斜长、硬岩岩性比例系数和开采方法作为导水裂隙带高度的主要影响因素,搜集200例导水裂隙带高度实测样本作为模型数据集。首先,采用自适应高斯噪声过采样方法(synthetic minority over-sampling technique for regression with Gaussian noise,SMOGN)对原始数据集进行过采样,结合8折交叉验证,将平均绝对误差(EMA)、均方根误差(ERMS)和决定系数(R2)作为回归模型评价指标,确定最优的BP神经网络结构,然后采用变异粒子群优化算法(mutation particle swarm optimization,MPSO),对神经网络的初始权值和阈值进行优化,最后将优化后的预测模型进行工程现场应用。【结果和结论】结果表明:该数据集下,BP神经网络采用Huber loss和Adam一阶优化算法,训练速度和稳定性均得到提升,最优激活函数为Tanh,最优隐藏层节点数为12。当MPSO种群数量为50时,模型性能最好,经过SMOGN过采样和MPSO超参数优化,最终训练集的EMA为0.163,ERMS为0.216,R2为0.948,验证集的EMA为0.260,ERMS为0.341,R2为0.901。在现场应用中模型预测的相对误差均在9%以下。结果表明结合SMOGN技术和MPSO超参数优化技术,显著提高了模型的稳定性和泛化性能,改善了样本分布特征,提高了样本利用效率和模型预测效果,对导水裂隙带高度模型的训练和预测具有重要的借鉴意义。[Objective]The height of a hydraulically conductive fracture zone,a significant factor influencing roof water inrushes and groundwater resource loss,is identified as a research focus of the prevention and control of mine water disasters. [Methods] To accurately predict the heights of hydraulically conductive fracture zones in coal seam roofs, fiveparameters were selected as the primary factors influencing hydraulically conductive fracture zones the mining depth:mining height, coal seam inclination, the length of the mining face along its dip direction, proportional coefficient ofhard rocks (i.e., the ratio of the cumulative thickness of hard rocks within the statistical height above the coal seam roofto the statistical height), and mining method. A total of 200 measured samples concerning the heights of hydraulicallyconductive fracture zones were collected as the model dataset. First, over-sampling of the original dataset was conductedusing the synthetic minority over-sampling technique for regression (SmoteR) combined with the introduction ofGaussian Noise (SMOGN). In conjunction with 8-fold cross-validation, the optimal back propagation (BP) neural networkstructure was determined by using the mean absolute error (denoted by EMA), root mean square error (denoted byERMS), and coefficient of determination (denoted by R2) as the assessment indices of the regression model. Then, the initialweights and thresholds of the BP neural network were optimized using the mutation particle swarm optimization(MPSO) algorithm. Finally, the optimized prediction model, i.e., the MPSO-BP model, was applied to the engineeringfield. [Results and Conclusions] The results indicate that based on the original dataset, the BP neural network, using theHuber loss and Adam first-order optimization algorithm, enhanced the training speed and stability. Consequently, the optimalactivation function was determined at Tanh and the optimal hidden layer node number at 12. The MPSO-BP modelyielded the optimal performance where the MPSO population num

关 键 词：煤矿防治水 回归过采样 导水裂隙带 高度预测 变异粒子群算法 模型优化 

分 类 号：TD745.2[矿业工程—矿井通风与安全]