针对华北型煤田煤层底板突水监测预警问题，以底板“下三带”理论为基础，开展了微震-电法耦合系统水害监测预警。在分析微震-电法耦合系统监测数据基础上，采用Flume设计数据迁移子系统，以流处理方式对监测源数据进行预处理，对关键目标数据进行采集、聚合和传输，实现了有效监测数据的实时迁移。此外，针对煤矿水害多源监测预警过程中数据规模大、数据实时处理要求高等特点，结合多源异构数据关联分析和时空属性数据分析处理需求，基于Spark和HDFS设计实现了具备TB级数据存储处理能力的煤矿水害多源监测大数据存储平台。该平台采用HDFS设计构建统一的多源时序大数据存储体系，通过MapReduce实现大数据并行处理，利用YARN实现资源的调度与管理，为海量数据存储提供支撑。平台采用Spark Streaming框架搭建了数据实时处理中心，通过流处理方式实现监测数据高速处理，并通过智能预警算法模块和远程服务接口为预警系统现场应用提供支撑。在智能预警技术方面，结合监测数据的时空属性特点，提出了基于深度学习时空序列预测方法——长短时记忆循环网络智能预警模型的底板突水模型预警技术。该预警技术基于LSTM方法，以“下三带”理论为基准对模型进行初始化，形成初始预警判据;将电法、微震监测数据作为输入变量，实际涌(突)水事件作为干预输出量，对智能预警模型进行半监督分类学习训练，形成动态化、参数最优的模型预警准则，将监测数据动态划分为4个预警等级，从而实现了水害智能动态预警和数据可视化表达。在冀中能源葛泉矿东井的实际应用中发现，该平台能够基本达到预期目标。

A microseism-electric coupling system to alert the coal disaster of water inrush has been proposed,aimed at North-China-Type coalfields and based on the theory of Down Three Zones. Firstly,a migration subsystem to collect o- riginal target data by using Flume has been designed,enabling the effectiveness and instantaneity of data preprocess- ing. It also aggregates and transmits these data from raw signals to a database system. This database as a storage center employs Spark and HDFS to store this large-scale,real-time,multi-source,heterogeneous,and spatial-temporal data, and to possess the TB-level capacity. This storage center has a unified multi-source sequential storage architecture to support saving big data by using MapReduce to process the data in parallel and YARN to dispatch and manage the re- source efficiently. It also leverages Spark Streaming to design real-time and high-efficiency data processing with an in- telligent warning module and a remote service interface to serve appli-cations. By performing the migration subsystem and storage center to obtain processed coal-water inrush data,a Long Short-Term Memory ( LSTM) model is used to predict the disaster of water inrush using the theory of Down Three Zones corporately. In the semi-supervised training of this model,the microseism-electric data are inputted,and then the model’s output is the alert level that divides the coal disaster of water inrush into four ranks. The LSTM model can warn the level of water inrush automatically and show the visualization of alert information. In practice,the system can provide timely and effective warning capabilities for east well in Gequan coal mine,Jizhong Energy Group,China.