冲击地压预测预警有助于全面掌握灾害风险程度,提前采取针对性防冲措施,可以有效降低灾害影响｡冲击危险性等级､冲击危险性区域与潜在冲击时间的实时量化预测是冲击地压预测预警的核心｡为此,提出“双驱动”冲击地压智能预警架构,结合物理驱动与数据驱动,动态实时确定工作面冲击地压的等级､时间及区域｡在物理驱动的框架下,通过贝叶斯概率的综合模型,使用工作面实时微震数据､巷道实时应力､地震CT-微震等相关参数,对工作面冲击地压危险性等级进行动静态协同综合实时评价;在数据驱动的框架下,以微震大事件定量预测为切入点,构建一个结合普通卷积模块､循环神经网络模块与AR 自回归模型的深度学习模型MSNet,利用工作面历史微震事件作为模型输入,定量动态预测未来几次微震事件发生的时间､位置与能量,继而确定冲击时间与危险性区域,并基于Unity3D软件开发了相应的冲击地压三维智能预警平台｡创新性提出了基于物理及数据“双驱动”的冲击地压灾害预警技术,实现了冲击地压风险特征信息的动态智能预警预测和危险区域的三维可视化显示｡平台的现场应用表明,具有冲击危险性的等级预测精度可达0.88｡和现场微震记录相比,工作面微震事件位置坐标预测的平均精度达到0.977,时间预测的平均精度达到0.523,基于微震事件定量预测结果的冲击区域与时间确定和现场实际情况表现一致,显示出了良好的工程适用性｡

The prediction and pre warning of the coal burst are helpful to mastering the degree of disaster risks and taking some prevention measures in advance, which is important for reducing the disaster consequences. The real time quantitative predictions for the coal burst hazard level, hazardous zone and occurring time are key issues for the coal burst prediction. This study proposes a “dual driven” intelligent pre warning framework which integrates the physics driven and data driven models, predicting the disaster hazard level, hazardous zone and occurring time. Under the framework of physics driven, this study introduces the Bayes probability model performing a dynamic static, synergistic, real time evaluation for the coal burst hazard level using field parameters including micro seismic data, real time stress data and seismic CT micro seismic detection data. Under the framework of the data driven, based on the quantitative predictions of micro seismic events, this study builds a deep learning model MSNet which combines a vanilla convolutional module, recurrent neural network modules and autoregressive module. The MSNet takes historical micro seismic data as model input and quantitatively predicts the time, zone and energy for imminent micro seismic events whereby determining the potential time and location for the coal burst. This study also develops an intelligent pre warning platform which embeds the “dual driven” model for the coal burst for a deep coal mine using Unity3D software. This study proposes a novel intelligent coal burst technology based on the integration of physics driven and data driven. The developed pre warning platform can predict and display the coal burst hazard level, hazardous zone and potential time in a real time. The application of the platforms demonstrates the success in a long wall workface with the predicting accuracy of hazard level reaching 0.88. The predicting accuracies for micro seismic event coordinates and time are 0.977 and 0.523, respectively. The predictions for the potential coal burst zone and time based on the micro seismic event predications are consistent with the field log, which indicates the feasibility of “dual driven” model in the project field.

1 冲击地压“双驱动”智能预警技术架构

2 冲击地压危险性动静态协同评价模型

   2.1 静态评价

   2.2 基于动态应力监测的冲击危险性评价

   2.3 微震多维信息动态评价

   2.4 基于贝叶斯方法的动静态协同评价模型

3 微震事件定量预测数据驱动模型

   3.1 模型基础数据

   3.2 数据驱动模型设计

   3.3 模型训练

   3.4 模型结果分析

4 基于“双驱动”模型的冲击地压智能预警平台开发

   4.1 平台三维几何模型

   4.2 “双驱动”预警模块实现流程

5 工程应用

6 结论与展望