【作者】 李翔；

【导师】 徐奴文； 李治国；

【作者基本信息】 四川大学 ， 水利工程（专业学位）， 2021， 硕士

【摘要】 岩爆是高地应力条件下深埋地下工程中主要的地质灾害类型之一,不仅对人身安全构成威胁,而且会造成施工设备的损坏,延误施工进度。如果能够在施工过程中有效地对岩爆灾害进行预警,还可以降低岩爆造成的危害。微震监测技术是一种先进的监测手段,它突破了传统的点、线分布监测的局面,可以对岩爆发生前的某些特征信息进行有效的监测。尽管微震监测技术的应用已日趋成熟,但如何提高其应用效率还有待于进一步研究。当前学科交叉渗透十分盛行,人工智能得到了广泛应用,运用智能算法提高岩爆预警效率可为地下工程安全建设提供宝贵的参考。作为智能算法的一种,贝叶斯网络是贝叶斯统计的应用分支,它对不确定性问题具有强大的处理能力,能够有效地表达和融合多源信息。因此本文以贝叶斯理论为背景,以微震震源参数为基础,探讨了一种基于贝叶斯网络的微震多参数岩爆预警方法。通过收集实际工程中的微震监测资料,分析了岩爆发生前微震事件的震源参数特征。选择有代表性的参数作为贝叶斯网络的输入,初步研究构造的岩爆贝叶斯网络模型。本论文的主要研究工作有:(1)结合实际工程实例,对岩爆发生前多起微震事件的震源参数特征进行分析。研究结果表明,矩震级和地震能量两个震源参数都具有一定的规律性,在岩爆发生之前,多数事件的矩震级和地震能量数值偏大可以作为岩爆预警的指标。震源尺度和视体积可以表示岩体破裂的损伤程度,当多数微震事件的震源尺度和视体积偏大时,表明岩体内部损伤加重,岩体稳定性降低。视应力和动态应力降表征岩体内部的发生位移前后的应力释放,当出现数值偏大,说明岩体内部活动加剧。(2)实际工程中所用的微震监测系统是加拿大的ESG系统,该系统获得的微震源参数主要有矩震级、地震能量、震源尺度、视应力、动应力降等,包括岩体破裂时的非弹性变形、应力释放与调整、损伤张量等。选择这五个参数作为岩爆预警模型的输入参数。建立合理的隧洞开挖安全监测系统,得到有效的监测数据,通过实际岩爆分析发现,各参数虽可作为岩爆预警的指标,但单独进行分析效果较差,多个参数相结合可显示出较好的效果。(3)基于静态贝叶斯网络,提出了岩爆综合预警模型。岩爆前,微震事件在一定时期内时空分布较为密集,这些微震事件的参数均可融合为一个特征值。选择岩爆发生前连续发生的50次微震的参数数据,分别求出5个参数的平均值,得出岩爆发生前的5个前兆值。以114例岩爆为研究对象,整理出114×5组前兆值,通过贝叶斯网络学习,建立岩爆预警模型。该模型经自我验证的正确率为95.61%,6倍交叉验证的正确率为92.98%,ROC曲线下面积接近1,均表明该模型具有良好的可靠性。并对五个参数的影响强度和灵敏度进行了分析,发现矩震级和地震能两个参数所占比重较大,可作为岩爆预警的重要参数。通过将某水电站施工排水隧洞数据导入到模型中,预警成功率为83.33%,说明该预警模型有一定的参考价值。(4)基于动态贝叶斯网络,提出了微震多参数岩爆预警模型。针对微震震源参数的时序问题,将岩爆前50次微震资料作为时序片段输入动态贝叶斯网络。通过对静态贝叶斯网络模型的扩展,将其演化为动态贝叶斯网络结构,并将整理好的114×5组特征序列导入动态贝叶斯网络进行参数学习,建立岩爆预警模型。该模型通过了自我验证,6次交叉验证和ROC曲线分析。同时对各参数的影响强度进行了分析,发现矩震级和震源尺度具有很强的代表性。在MFC平台上编制了岩爆预警工具程序,并将某水电站的进厂交通洞和施工排水洞的实际数据导入到模型中,预警成功率分别为81.82%和83.33%,说明预警模型具有一定的应用价值。更多还原

【Abstract】 Rockburst is one of the main types of geological disasters in deep buried underground engineering under high geostress conditions.It not only poses a threat to personal safety,but also causes damage to construction equipment and delays the construction progress.If the early warning of rockburst disaster can be effectively carried out in the construction process,the damage caused by rockburst can also be reduced.Microseismic(MS)monitoring technology is an advanced monitoring method,which breaks through the traditional situation of point and line distribution monitoring,and can effectively monitor some characteristic information before rockburst.Although the application of MS monitoring technology has become more and more mature,how to improve its application efficiency remains to be further studied.At present,interdisciplinary penetration is very popular,and artificial intelligence has been widely used.Using intelligent algorithm to improve the efficiency of rockburst early warning can provide valuable reference for underground engineering safety construction.As a kind of intelligent algorithm,Bayesian network is an application branch of Bayesian statistics.It has strong ability to deal with uncertain problems,and can effectively express and fuse multi-source information.Therefore,based on Bayesian theory and MS source parameters,this paper discusses a multi parameter MS rockburst early warning method based on Bayesian network.Based on the MS monitoring of a practical project,the characteristics of source parameters of MS events before the rockburst are analyzed.The representative parameters are selected as the input of Bayesian network,and the rockburst Bayesian network model is preliminarily studied.The main research work of this paper is as follows(1)Combined with practical engineering examples,the characteristics of source parameters of several MS events before rockburst are analyzed.The results show that the moment magnitude and seismic energy have certain regularity.Before the occurrence of rockburst,the moment magnitude and seismic energy of most events are too large,which can be used as indicators of rockburst warning.When the focal scale and apparent volume of most MS events are too large,it indicates that the damage inside the rock mass is aggravated and the stability of the rock mass is reduced.The apparent stress and dynamic stress drop represent the stress release before and after displacement in the rock mass.When the value is too large,it indicates that the internal activity of the rock mass is intensified.(2)The MS monitoring system used in practical engineering is ESG of Canada.The MS parameters obtained by the system mainly include moment magnitude,seismic energy,focal size,apparent stress,dynamic stress drop,etc.,including inelastic deformation,stress release and adjustment,damage tensor,etc.These five parameters are selected as the input parameters of rockburst warning model.A reasonable tunnel excavation safety monitoring system is established to obtain effective monitoring data.Through the analysis of actual rockburst,it is found that although each parameter can be used as the index of rockburst warning,the effect of single analysis is poor,and the combination of multiple parameters can show better effect.(3)Based on static Bayesian networks,an integrated rockburst warning model is proposed.Before rockburst,the spatial and temporal distribution of MS events is relatively intensive in a certain period,and the parameters of these MS events can be integrated into eigenvalues.The parameter data of 50 microseisms before the rockburst are selected,and the average values of five parameters are calculated respectively,and the five precursory values before the rockburst are obtained.Taking 114 cases of rockburst as the research object,114 × 5 groups of precursor values are sorted out,and the early warning model of rockburst is established through Bayesian network learning.The accuracy of the model is 95.61% by self-verification,92.98% by six times cross validation,and the area under ROC is close to 1,which indicates that the model has good reliability.The influence intensity and sensitivity of the five parameters are analyzed.It is found that the moment magnitude and seismic energy account for a large proportion and can be used as important parameters for rockburst early warning.By importing the drainage tunnel data of a hydropower station into the model,the success rate of early warning is 83.33%,which shows that the early warning model has a certain reference value.(4)Based on dynamic state Bayesian networks,an MS multi-parameter rockburst warning model is proposed.Aiming at the timing problem of MS source parameters,50 MS data before rockburst are input into dynamic Bayesian network as time series.By extending the static Bayesian network model,it is evolved into a dynamic Bayesian network structure,and the 114 × 5 feature sequences are imported into the dynamic Bayesian network for parameter learning to establish a rockburst warning model.The model has passed self-validation,six cross validation and ROC curve analysis.At the same time,the influence intensity of each parameter is analyzed,and it is found that moment magnitude and focal scale are very representative.The program of rockburst early warning tool is compiled on MFC platform,and the actual data of access tunnel and drainage tunnel of hydropower station are imported into the model.The success rate of early warning is 81.82% and 83.33% respectively,which indicates that the early warning model has certain application value.更多还原