摘要
基于深度学习,提出了一种分析采动地裂缝成因及预测地裂缝发育程度的方法。通过分析官地煤矿的井田地质测量资料和实地调查,确定了11类影响地裂缝发育的因素;依据地裂缝面积与采空区面积的比值,将地裂缝发育程度分为4类。利用深度学习的方法,构建了全连接深度神经网络模型(DNN)对裂缝发育程度进行预测;以预测准确率为指标,通过6次特征选择对影响因素的重要性进行了分析。特征选择的结果表明:开采层数、开采总厚度、开采宽度、开采深度、砂泥岩比、开采长度是影响地裂缝发育的主要特征,地质构造和地表出露是次要特征,煤层倾角、地形坡度、相对位置是冗余特征。与卷积神经网络(CNN)及循环神经网络(RNN/LSTM)模型的训练结果相比,DNN模型预测准确率较高。
In order to analyze ground fissures caused by coal mining and predict the development degree of ground fissures,on the basis of summarizing the previous researches,11 kinds of influencing factor were determined by analyzing the geological survey data and field investigation of Guandi coal mine field.According to the ratio of the acreage of the ground fissures to the gob area,the development degree of ground fissures was divided into 4 categories.By using the method of deep learning,a fully connected deep neural network model(DNN)was constructed to predict the degree of crack development;with the prediction accuracy as an indicator,the importance of influencing factors was analyzed through 6 feature selections.The feature selection reveals that,the mining layers,thickness,width,depth,length and sandstone-mudstone ratio are the main features affecting the development of ground fissures;geological structures and surface excavation are secondary features;dip angle of coal seam,terrain slope and relative position are redundant features.Comparison with convolutional neural network and recurrent neural network shows that the DNN model has a higher prediction accuracy.
作者
贾杨
吕义清
JIA Yang;LYU Yiqing(College of Mining Engineering, Taiyuan University of Technology, Taiyuan 030024, China)
出处
《太原理工大学学报》
CAS
北大核心
2020年第3期411-417,共7页
Journal of Taiyuan University of Technology
基金
山西省自然科学基金资助项目(201701D121015)。
关键词
地裂缝
裂缝发育
预测
神经网络
深度学习
特征选择
官地煤矿
ground fissure
fissure development
prediction
neural network
deep learning
feature selection
Guandi coal mine
作者简介
通讯作者:吕义清(1970-),男,博士,副教授,主要从事环境与地质灾害方面的研究,(E-mail)lv-yiqing@sina.com。
