【作者】 郭立；

【导师】 吴爱祥；

【作者基本信息】 中南大学 ， 采矿工程， 2004， 博士

【摘要】 岩爆倾向性预测是防治岩爆灾害的基础。根据预测结果及时进行岩石工程反馈设计并采取安全防治措施,对于深部硬岩开采和高应力地下工程的建设具有重要的理论意义与应用价值。论文针对岩爆倾向性预测方法的发展趋势及现存的局限性,结合国家“十五”重点攻关项目“复杂难采深部铜矿床安全高效开采关键技术研究”之子项目—冬瓜山高应力开采岩爆预测预报研究,应用岩石工程系统理论和相关的数据挖掘技术,对深部硬岩岩爆倾向性动态预测模型及其应用价值展开了以下主要研究: (1) 建立了岩爆倾向性动态预测的概念,从系统分析与系统控制的研究思路出发,提出了基于先验信息和数据挖掘技术的岩爆倾向性动态预测研究方法并给出了其实现流程。 (2) 引入岩石工程系统理论方法构建工程岩爆倾向性的动态预测模型。以客观存在或易于量测的工程地质因素、工程环境因素和开挖因素作为系统输入参数,通过构造交互作用矩阵列举岩爆倾向性的影响因素并研究其交互作用的机理,既可以实现深部硬岩岩爆倾向性的动态预测,又能够分析各因素的相对作用强度大小,进而确定岩爆启动的主控因素。 (3) 研究了基于改进BP神经网络的岩爆动态预测模型实现方法。在深入研究BP神经网络算法原理及神经网络参数分析原理的基础上,对整体相对作用矩阵的计算式进行改进,提出了(整体)相对作用强度的统一计算式(G)RSE;采用异变换函数法、动量法和自适应调整学习率等综合方法对典型BP网络进行改进,并应用于预测模型的交互作用矩阵编码,丰富了岩石工程系统理论的编码方式;推导了基于改进BP神经网络的(整体)相对作用强度(G)RSE的算法流程。 (4) 以岩爆倾向性的主要影响因素作为输入层,岩爆倾向性作为输出层,构建三层改进算法的BP神经网络模型,对深部VCR法采场和冬瓜山深部巷道的岩爆倾向性进行预测和检验,同时,在选定样本范围内,分别对这两种情况下各因素的交互作用机理进行研究,确定了其相对作用大小、时空变异特性以及岩爆启动的主控影响因素。 (5) 应用模糊数学和模糊信息优化的相关理论和方法,构造了一种新的岩爆倾向性动态预测模型,即模糊神经网络预测模型,变传统神经网络的点-点映射关系为集合-集合的映射关系,变典型的黑箱学习方式为被人容易接受的“IF-THEN”表达方法,同时充分利用模糊理论与神经网络融合于一体的强大优势,较好地解决了复杂条件下岩爆倾向性动态预测可能存在的矛盾样本点带来的网络不收敛、小样本情况下的信息不足以及知识表达困难的问题,拓展了岩爆倾向性动态预测模型的应用范围。 (6) 结合实例,研究了深部采场岩爆倾向性模糊神经网络预测模型的关键环节实现方法。在缺少或无法得到更详细资料的情况下,采用模糊C均值聚类和减法聚类相结合的分析方法对先验信息进行数据处理,根据函数隶属程度的大小和聚类样本的特性对更多还原

【Abstract】 The prediction of rockbursts proneness is the basis of preventing and controlling rockbursts disaster. Based on predicting results, the back-devising to rock engineering and the security measurements can be carried out in time, which have very important theoretical and practical meaning to hard rock mining in depth and underground rock engineering constructing in high stress. Aimed at the developing trends and existing shortage of predicting approach of rockbursts proneness, meanwhile combined with the project of Dongguashan rockbursts prediction in high-stress mining, the dynamic predicting models and pre-controlling approaches of rock engineering rockbursts proneness are studied in this Ph.D dissertation by the use of rock engineering system and data-mining approach. the main researches have been studied as the following:(1) The concept to dynamically predict rockbursts proneness is built. Then the approaches to predict hard rock rockbursts at depth based on previous information and data-mining method are brought forward and its flow chart from the jumping-off place applied with system analyzing and system controlling methods to study problems.(2) The dynamic predicting model utilizing RES approaches to predict engineering rockbursts proneness is constructed, in which engineering geologic factors, engineering environmental factors that should be objective or easily measured are inputted into system as parameters. Then, interactive matrix is built to enumerate influencing factors on rockbursts proneness to study their interactive mechanism, which can not only realize the dynamic predicting to engineering rockbursts proneness, but also analyze each factor’s relative strength of effect, besides this the main influencing factors induced rockburst can be found out.(3) The approaches to dynamically predict rockbursts proneness based on improved BP neural network are studied, which mainly includes the researches as establishing formula of GRSE and unite formula (G) RSE of (Global) relative strength of effect based on deeply studying the principle of BP algorithms and BP parameters’ analyzing , improving BP algorithms to decode the model’s interactive matrix by synthetically applying different transfer function, momentum method and self-recognizing learning rate so that the decoding approach of RES is enriched, deducing the algorithmic process of (G)RSE based on improved BP network.(4) The predicting model with three layers by improved BP algorithms in which influencing factors act as input layer and rockbursts proneness acts as output layer is built, which is used to predict and checkout rockbursts proneness taking place in deep stope by VCR mining and in Dongguashan deep tunnel. Meanwhile, within the special samples, the interactive mechanism of influencing factors is studied individually so that their relativestrength of effect, time-space differentiate characteristics and the main influencing factors induced rockbursts are made clearly.(5) With the application of fuzzy mathematic and fuzzy information optimizing theory and approaches , the improved intelligent model to dynamically predict rockbursts proneness is constructed, in which traditional spot to spot mapping is replaced by muster to muster mapping, typical black box learning manner is replaced by IF-THEN expressing manner which is easily accepted by people. Meanwhile the powerful advantage of fuzzy theory integrated with neural is fully utilized. Therefore, the probably problems such as not constringency caused by inconsistency with samples each other, information absence under the complex condition of few samples and difficult in knowledge expressing are settled preferably. As the result, the application range of the model to dynamically predict rockbursts proneness is widened.(6) Based on examples’ characteristic, the key approaches to predict rockbursts proneness by fuzzy neural model in deep stope is discussed, which involves the researches as that applying c-clustering integrated with subtractive clustering approach to deal with data kept in previous information under the condition of lacking much more datum. Then in terms of subjective degree of function and the characteristics of clustering samples, the violent degree of rockbursts is finely marked off. Also the fuzzy relationship between influencing factors and rockbursts proneness is studied by applying fuzzy information optimizing theory. More over, a new approach to evaluate each influencing factor’s weight is brought forward which fully considers multi-factors’ interaction.(7) The fuzzy neural model is adopted to predicting and checkouting rockburst proneness taking place in deep stope by VCR mining, which indicates that the model has active meaning both in mining previous information and directing engineering practicing.(8) The main influencing factors , stress distributing state probably inducing deep stope rockbursts in Dongguashan copper mine is further studied by on-spot experiments, room experiments combined with numerical simulation, from which sti-stress distribution discipline, especially stress newly distributing at wall-rock which is induced by static pressure, semi-static pressure and dynamic pressure at different mining techniques is discovered. So, by back devising, mining schemes are optimized under rockbursts proneness. The study in this Ph.D dissertation offer a important decision-making reference to efficiently pre-control rockbursts probably induced by stope mining in Dongguashan copper mine.更多还原