【作者】 彭桂力；

【导师】 庹先国；

【作者基本信息】 西南科技大学 ， 控制科学与工程， 2021， 博士

【摘要】 深埋隧道工程中的岩爆事件具有突发性、猛烈性等特点,会给高速公路、铁路中的隧道施工建设带来严重的安全隐患。岩爆事件的实时监测、预报已成为深埋地下工程建设和岩爆机理研究的共同课题。微地震监测是岩爆监测预警的方法之一,通过监测微地震信号反演岩石破裂位置,但其定位精度受到多种因素影响,可以通过加入主动震源技术去提高岩石破裂定位精度,利用已知的主动震源作为标准,探测未知岩石破裂位置定位方法中是必不可少的,然而加入主动震源信号会带来微地震监测信号更加复杂的问题。本论文围绕如何从复杂微地震混合信号中有效分解、识别微地震信号这一问题,在国内外研究学者的基础上,加入主动震源探测技术,观测岩石微破裂的现象,重点针对在主动震源与噪声干扰下的微地震信号的分解和识别算法展开研究,并利用识别的微地震与主动震源信号进行精准定位。论文主要工作和取得的成果如下:1.针对基于主动震源技术探测岩石微破裂的研究中,增加了复杂微地震混合信号分解困难的问题,提出了基于奇异值经验模态分解(Singular Value Empirical Mode Decomposition,SVEMD)算法,对于含有主动震源信号、微地震信号和噪声信号的复杂混合信号进行有效分解。该算法具有分解精度高,残差功率低,可以得到较高信噪比的特点,对混合信号在频域范围内进行有效分解,依次得到噪声、主动震源与微地震信号。2.针对复杂微地震混合信号中含有大量微地震、主动震源和噪声等信号,存在人工识别效率低下,精度不高的问题,提出将信号识别转化为图像识别,将卷积神经网络(Convolutional Neural Network,CNN)应用于微地震信号识别中,改进现有CNN算法,加入迁移(Inception)结构,形成深度迁移卷积神经网络(Deep Convolutional Neural Network-Inception,DCNN-Inception),利用该网络实现微地震信号的自动识别。利用测试集对该算法进行测试,获得信号的识别准确率达到92.4%,同现有的CNN网络相比,识别精度提高7.1%,同时得到信号的损失率为17.4%,比现有CNN网络降低了13%。DCNN-Inception网络微地震数据特征拟合能力明显强于CNN网络,特征提取能力强,准确率高。3.针对岩石微破裂中微地震事件的定位和精细描绘问题,提出引入主动震源技术,优化速度模型,利用主动震源技术反演得到初至时间和位置信息。同时提出了基于主动震源技术的定位方法,改进了传统到时差定位算法的求解方法,建立线性数学方程组,改善了岩石微破裂的刻画精度,避免了的传统Geiger求解方法中受初值影响较大的问题,优化Geiger算法,为进一步实现微地震事件精确定位,分析岩爆提供可靠的数据支持。借助计算机中的MATLAB数值仿真工具构造微地震信号数据,西南科技大学土木实验室进行主动震源实验采集的主动震源数据和白鹤滩水电站岩爆监测工作获得的有效微地震数据,加载本论文提出的SVEMD算法,通过算法验证和仿真结果分析对比,验证SVEMD算法的可行性和有效性。将白鹤滩监测的大量微地震数据构建数据集,用于训练及验证本文的DCNN-Inception算法,通过算法可以有效识别出微地震信号。将计算机仿真软件构造数据与主动震源实验采集数据加载本文中提出的基于主动震源的岩石破裂定位算法,通过算法验证和定位结果对比,证明该算法可以提高现有Geiger算法的计算效率,提高定位精度。更多还原

【Abstract】 The rock burst in deep tunnel engineering has the characteristics of sudden and violent,which will bring serious security risks to the tunnel construction of expressway and high-speed railway.The real-time monitoring and prediction of rockburst has become a common topic in the deep underground engineering construction and rockburst mechanism research.Microseismic monitoring is one of the methods of rockburst monitoring and early warning.It inverses the location of rock fracture by monitoring microseismic signals.The positioning accuracy is affected by many factors.It can be improved by adding active source technology.This method uses the known active source as the standard to detect the unknown rock fracture position,which is essential in the positioning method.However,adding active source signal will bring more complex problems to the microseismic monitoring signal.This dissertation focuses on how to effectively decompose and identify microseismic signals from complex mixed microseismic signals.On the basis of domestic and foreign researchers,active source detection technology is added to observe the rock micro-fracture phenomenon.This dissertation focuses on the decomposition and identification algorithm of microseismic signals under the active source and noise,and uses the identified microseismic and active source signals for accurate positioning.The main work and achievements are as follows:1.Aiming at the problems of difficulty in decomposition of complex mixed microseismic signals in the detection of rock microfracture based on active source technology.This dissertation proposes an algorithm based on Singular Value Empirical Mode Decomposition(SVEMD).SVEMD algorithm can effectively decompose complex mixed signal including active source signal,microseismic signal and noise signal.This algorithm has the characteristics of high resolution,low residual power and high SNR.It can decompose the mixed signal effectively in the frequency domain to obtain the noise,active source and microseismic signal in turn2.Aiming at the problems of low efficiency and low precision by manual identification microseismic signal in in complex microseismic mixed signals which has a lot of microseismic,active source and noise signals.This dissertation proposes a method of transforming signal recognition into image recognition and Convolutional Neural Network(CNN)is applied to microseismic signal recognition.It improves the existing CNN algorithm by adding the Inception structure to form the Deep Convolutional Neural Network-Inception(DCNN-Inception).The network realizes the automatic identification of microseismic signals.Using the test set to test the algorithm,the recognition accuracy of the signal is 92.4%,compared with the existing CNN network,the recognition efficiency is improved by 7.1%,and the loss rate of the signal is 17.4%,which is 13% lower than the existing CNN network.The feature fitting ability of DCNN-Inception network is obviously better than that of CNN network,the feature extraction ability is strong,and the accuracy is obviously improved.3.Aiming at the problems of the location and fine description of microseismic events in rock microfracture.This dissertation proposes the active source technology which can optimize the velocity model,and get the first break time and position information.Meanwhile,the active source technology improves the solution method of traditional TDOA location algorithm,establishes linear mathematical equations,improves the description accuracy of rock micro fracture,avoids the problem that the initial value has great influence in the traditional Geiger method,and optimizes the Geiger method.It provides reliable data support for further accurate positioning of microseismic events and analysis of rockburst.With the help of numerical simulation,the microseismic signal data are constructed,and the active source data collected by the Civil Engineering Laboratory of Southwest University of science and technology and the effective microseismic data obtained from the Baihetan Hydropower Station rockburst monitoring.Load the SVEMD algorithm and verify the feasibility and effectiveness.A large number of microseismic data monitored in Baihetan are used to build data sets to train and verify the DCNN-Inception algorithm,which can effectively identify the microseismic signals.At last,loading the active source location algorithm verify the positioning results.It is proved that the algorithm can improve the calculation efficiency and positioning accuracy of the existing Geiger algorithm更多还原