(1.西安邮电大学 通信与信息工程学院,陕西 西安 710121; 2.西安邮电大学 经济与管理学院,陕西 西安 710121)

瓦斯预测预警系统; 神经网络; 长短时记忆网络

Application and design of LSTM in coal mine gas prediction and warning system
LI Wei-shan1,WANG Lin1,WEI Chen2

(1.School of Communication and Information Engineering,Xi'an University of Posts and Telecommunications,Xi'an 710121,China; 2.School of Economics and Management,Xi'an University of Posts and Telecommunications,Xi'an 710121,China)

gas prediction and warning system; neural network; long short time memory

DOI: 10.13800/j.cnki.xakjdxxb.2018.0621


针对煤矿瓦斯浓度的预测的问题,以亭南煤矿正常生产期间302工作面的监测数据为研究背景,采用深度学习技术LSTM(Long Short Time Memory,长短时记忆网络)建立瓦斯预测模型,研究与设计了基于LSTM的煤矿瓦斯预测预警系统。LSTM网络针对时间序列数据具有较强的建模能力,能够实现信息的长期依赖,自动挖掘数据之间潜在的关联关系。采集煤矿正常生产期间的瓦斯监测数据作为训练数据,利用深度学习框架TensorFlow进行算法的仿真,并研究了不同时间步长、网络深度下的LSTM以及多信息融合对瓦斯预测模型性能的影响。实验结果在1 000条测试数据集上获得了3.61%平均相对偏差,LSTM瓦斯预测模型具有较高的准确度,泛化能力强。在系统研究与设计中,使用Spring,SpringMVC和Hibernate框架按照适应性、易用性、可扩展性等原则对系统进行了设计。系统部署阶段,将训练好的LSTM瓦斯预测模型部署在TensorFlow Serving服务器中,对外提供服务,实现了煤矿瓦斯预警系统,增强了煤矿瓦斯监控系统的预警能力,提高了煤炭企业安全生产管理水平,具有一定的实用价值。

In order to solve the problem of prediction of coal mine gas concentration,taking the monitoring data of the 302face during the normal production period of Tingnan Coal Mine as research background,using deep learning technology LSTM to establish the model of gas prediction,a coal mine gas prediction and early warning system based on LSTM(Long Short Time Memory)is designed.LSTM has strong modeling ability regarding time-series data,can realize the long-term dependence on information and automatically excavate the potential relationship between data.Collecting the gas monitoring data during the normal production of the coal mine as training data,the TensorFlow deep learning framework is used to simulate the coal mine gas prediction model.The effects of different time steps,network depth and multi-information fusion on the performance of the model are studied.The experimental results show an average relative deviation of 3.61% on 1000test datasets.The LSTM gas prediction model has high accuracy with strong generalization ability.In system research and design phase,the system is designed using Spring,SpringMVC and hibernate frameworks according to the principles of adaptability,ease of use,and scalability.In system deployment phase,the trained LSTM gas forecasting model is deployed in the server to provide services to the outside world,to realize the coal mine gas warning system,enhance the early warning capability of the coal mine gas monitoring system and improve the safety management of the coal mining enterprises.