Design And Implementation Of Multi-factor Early Warning System Of Mine Gas Based On Deep Learning

The gas disaster poses a serious threat to the lives of mine workers and also restricts the development of the coal mine industry.With the advancement of information technology,Shanxi Zhongxing Coal Mine has actively introduced monitoring technology and applied it to actual production.However,traditional monitoring equipment usually alarms when the gas concentration exceeds the limit.Therefore,the accident prediction ability is insufficient.At the same time,the underground sub-systems are scattered,which is not conducive to integration management.In order to solve the above problems,this thesis establishes the gas concentration prediction models based on deep learning theory,and it designs and implements a gas multi-factor early warning system.This thesis takes the monitoring data of the north wing roadway and 1209 working face of Zhongxing Coal Mine as the research background,and integrates the KJ95 monitoring subsystem to collect real-time data with the help of multi-factor sensors.After storing them in Mongo DB and My SQL,complete the data cleaning,repair and fusion tasks,to reasonably remove outliers,fill in the default values,and obtain the time series of multi-factor historical data.This thesis explores the influence of carbon monoxide,carbon dioxide,oxygen,wind speed and temperature on the trend of gas concentration by constructing gas univariate and multivariable prediction models.It designs the fusion structure of Convolutional Neural Network(CNN)and Long Short-Term Memory network(LSTM)to predict the change trend of gas concentration in the future.The time series prediction is transformed into supervised learning,and the prediction accuracy is improved by adjusting the network structure,optimizing algorithms and learning behavior.At the same time,the Gated Recurrent Unit(GRU)network was established for experimental training,and the Autoregressive Integrated Moving Average model(ARIMA)was established by ADF test and Akaike information criterion(AIC).Finally,the mean absolute error and mean square error results of the three models on the test set are compared.After selecting the preferred ones,they are encapsulated by the Django framework and deployed to the model server to provide interfaces for external access and calling.Based on the prediction model,this thesis determines the warning limits and levels,and then designs and implements the gas multi-factor warning system.This system is divided into five modules,which are trend warning module,data monitoring and visualization module,linkage warning module,formula processor module and permission management module.They jointly complete the early warning function of gas concentration trend,display multi-factor monitoring data in real time,and realize the linkage early warning and root cause analysis between various types of sensors.This system is built using Sping,Spring MVC and Hibernate integration framework,the front end uses Bootstrap framework,and realizes visual interaction with tools such as ECharts,Web GL and GIS.At present,the development and verification of the system have been completed,and it has been deployed and tested in the real environment.The operation results show that the system not only enhances the early warning capability of gas monitoring,but also improves the information fusion management level of coal mine enterprises,so it has a strong application value.