煤岩冲击失稳过程中的微震敏感特征演化规律及冲击灾变智能感知预测方法研究

The low accuracy of early warning caused by the low utilization rate of microseismic monitoring data and the relatively backward data analysis technology has become the bottleneck of the development of coal mine information and intelligent early warning technology. In this project, in order to explore the sensitive characteristics of rock burst and realize the intelligent perception and prediction of impact disaster, a combination of theoretical research, numerical analysis and model validation is utilized to carry out research of evolution law of microseismic sensitivity characteristic and intelligent perception prediction method for coal rock shock instability. By means of multi-scale analysis and other methods, the abundant information of time, frequency, spectrum and energy contained in microseismic monitoring data are deeply excavated, and the corresponding relationship between different stages of coal-rock impact instability and spatio-temporal evolution law of characteristics is revealed. In view of the high-dimensional feature space, screening sensitive features that have good stability, sensitivity and regularity before and after coal and rock impact disasters occurrence, and can significantly reflect the process of impact disaster. Using manifold learning method, a low-dimensional fusion sensitive feature set with better classification is constructed. On the basis of the abovementioned study, we intend to perform the analysis of feature data attributes and deep learning technology, construct an intelligent perception and prediction model of impact disaster, which can realize the intelligent perception and improve the performance of the early warning of impact disaster.

微震监测数据利用率低、数据分析技术手段相对落后造成的预警准确率低已成为煤矿信息化、智能化预警技术发展的瓶颈。本项目以探求煤岩冲击失稳敏感特征，实现冲击灾变智能感知和预测为目标，采用理论研究、数值分析和模型验证相结合的方法开展煤岩冲击失稳过程中的微震敏感特征演化规律及冲击灾变智能感知预测方法研究。借助多尺度分析等方法，深度挖掘微震监测数据中蕴含的丰富时、频、谱及能量特征信息，揭示煤岩冲击失稳不同阶段与特征发展演化规律间的对应关系；针对高维特征空间，筛选能够显著反映冲击灾变过程、煤岩冲击灾害发生前后稳定性好、敏感性和规律性强的敏感特征，利用流形学习的方法，构造分类性更好的低维融合敏感特征集；在此基础上，研究特征数据属性，运用深度学习技术，构建冲击灾变智能感知预测模型，实现冲击灾变智能感知和预警性能提升。

以煤矿微震监测技术为研究背景，研究冲击地压全过程的微震敏感特征演化规律及冲击灾变智能感知预测方法。系统地对煤矿微震监测数据进行分析，抽取有效微震事件的监测数据，研究物理意义明确的地震学参数、统计学参数、微震信号时域特征、频域特征、能量特征等特征参数在煤岩冲击失稳过程中的阶段性演化规律、突变特征及显著性，深入探究煤岩状态变化、失稳和破裂全过程与特征演化间的对应关系，并结合各特征参数的物理意义，揭示冲击灾害孕育、失稳破裂的微震前兆机理和能量前兆机理。.系统研究了基于深度学习的冲击灾变智能感知预测方法，验证了不同深度学习网络模型对时序数据预测的效果和作用，选择了一种最新的预测效果最佳的深度网络模型。利用VMD分解方法和Transformer网络建立监测区域煤岩体状态智能感知和预测模型。采用VMD对冲击地压敏感特征的非平稳时序数据进行分解，将时序数据分解为在频域上具有中心频率和有限带宽的模型分量，再使用Transformer网络对敏感特征的未来数据进行预测，同时利用新的冲击灾害数据不断修正智能感知和预测模型。构建冲击灾变智能感知和预测模型，实现监测区域煤岩体失稳状态的智能感知和预测，为冲击地压灾害信息化、智能化预测预警的更广泛应用提供理论支持。