摘要
液压支架支护过程是一个时间、空间上的动态变化过程,目前液压支架支护质量评价大多关注支架静态特征,对支架立柱压力动态变化研究较少。针对上述问题,采用深度学习方法构建了一种基于改进型LeNet-5网络的液压支架时空区域支护质量动态评价模型。首先,将工作面液压支架立柱压力数据进行预处理(缺失值填充、异常值处理、筛选、排序等),得到较为完整的液压支架压力数据。其次,将预处理后的液压支架立柱压力数据按照时间和空间排列,并提取反映智采工作面液压支架支护情况的初撑力、循环末阻力、时间加权阻力、阻力空间分布情况等重要特征量,将压力时间序列和空间序列组合为时间-空间二维总时空压力矩阵。再次,根据工作面支护要求,将时空区域支护质量划分为支护质量初步恶化、支护质量持续恶化、支护质量深度恶化、支护质量保持一般、支护质量初步优化、支护质量持续优化、支护质量保持良好7类,在总时空压力矩阵上使用滑动窗口按照一定间隔截取给定大小的子矩阵,将子矩阵与7类时空区域支护质量一一对应,形成样本和标签。最后,将样本和标签输入改进型LeNet-5网络进行训练,构建液压支架时空区域支护质量评价模型,实时评价该区域支架支护情况。实验结果表明:基于改进型LeNet-5网络的液压支架时空区域支护质量评价模型可用于工作面区域内支护质量动态效果辨识,为现场操作人员有针对性地调整液压支架支护状态提供依据,分类准确率为85.25%,比基于LeNet-5网络的模型提高了12%。同时,改进型LeNet-5网络在训练过程中能较快地收敛到最优解,加快了网络训练速度,验证了改进型LeNet-5网络用于智采工作面液压支架时空区域支护质量评价的优势。
The support process of hydraulic support is a dynamic change process in time and space regions.At present, the evaluation of support quality of hydraulic support mostly focuses on the static characteristics of support. There is little research on the dynamic change of support column pressure. In order to solve the above problems, a dynamic evaluation model of support quality of hydraulic support in the space-time region based on improved LeNet-5 network is established by using the deep learning method. Firstly, the column pressure data of hydraulic support in working face is preprocessed(missing value filling, abnormal value processing, screening,sorting, etc.) to obtain more complete pressure data hydraulic support. Secondly, the preprocessed column pressure data of the hydraulic support is arranged according to time and space. The important characteristics are extracted such as initial setting force, circulating end-resistance, time-weighted resistance, and spatial distribution condition of resistance, which reflect the support condition of the hydraulic support in intelligent mining working face. The pressure time-sequence and the pressure space-sequence are combined into the total space-time pressure matrix in 2D space-time region. Thirdly, according to the support requirements of the working face, the support quality of the space-time region into seven types, namely, support quality preliminary deterioration, support quality continuous deterioration, support quality deep deterioration, support quality general maintenance, support quality preliminary optimization, support quality continuous optimization and support quality good maintenance.On the total space-time pressure matrix, the sliding window is used to intercept the sub-matrix with the given size at a certain interval. The sub-matrix is one-to-one corresponding to the seven support quality types in space-time regions to form samples and labels. Finally, the samples and the labels are input into the improved LeNet-5network for training. The dynamic eval
作者
贾思锋
付翔
王然风
王宏伟
王朋飞
JIA Sifeng;FU Xiang;WANG Ranfeng;WANG Hongwei;WANG Pengfei(College of Mining Engineering,Taiyuan University of Technology,Taiyuan 030024,China;Center of Shanxi Engineering Research for Coal Mine Intelligent Equipment,Taiyuan University of Technology,Taiyuan 030024,China;Postdoctoral Workstation,Shanxi Coking Coal Group Co.,Ltd.,Taiyuan 030024,China;College of Mechanical and Vehicle Engineering,Taiyuan University of Technology,Taiyuan 030024,China)
出处
《工矿自动化》
北大核心
2022年第10期26-33,81,共9页
Journal Of Mine Automation
基金
国家自然科学基金项目(52274157,52274092)
国家重点研发计划项目(2020YFB1314004)
内蒙古自治区重点专项项目(2022EEDSKJXM010)
山西省揭榜招标项目(20201101005)。
关键词
智采工作面
液压支架
立柱压力
支护质量评价
时空区域
深度学习
卷积神经网络
intelligent working face
hydraulic support
column pressure
support quality evaluation
spacetime region
deep learning
convolutional neural network
