Design And Implementation Of Underground Gas Concentration Prediction System Based On Deep Learning

Coal is one of the important energy sources in China,and it occupies a large proportion of consumption.As the demand continues to increase,the mining of coal mines is also expanding.The safety of coal mine production is also a major issue that we must pay attention to.Gas disasters often occur in coal mine accidents,causing great harm to coal mining work.With the rapid development of communication technology,internet technology and big data technology,the underground distributed sensor system has been realized.The large amount of gas data monitored in the coal mining site presents high nonlinearity and high complexity.For the time being,the gas concentration prediction for the collected massive data has serious shortcomings in terms of reliability,accuracy and real-time.Therefore,this paper based on the Deep Belief Network to process and analyze the collected gas data,can make good use of its powerful function expression and feature extraction functions and the advantages of processing nonlinear data;and design a real-time data processing system based on Spark accurate and real-time prediction of gas concentration,which is applied to massive data analysis,is feasible from the aspects of maintenance and scalability in the later period.The main research work is as follows:Firstly,the traditional Back Propagation neural network is analyzed.Aiming at the problem that Back Propagation neural network is prone to gradient diffusion and falling into local minimum value.A Deep Belief Network based on the restricted Boltzmann Machine is proposed to construct the gas concentration prediction model.The model is developed from the artificial neural network,which not only makes full use of the advantages of the artificial neural network,but also makes up for its shortcomings.Compared with Back Propagation neural network prediction model,the feasibility and accuracy of its application in gas concentration prediction are further determined.Then,based on the characteristics of gas concentration data collected under the mine and the rapid growth of data volume,the Spark real-time data processing system is established,and the Deep Belief Network algorithm is combined to accurately and real-time predict the gas concentration.At the same time,the whole system has the characteristics of high availability,easy expansion,versatility etc,which alleviates the burden of system maintenance in the later stage and solves the problem of massive data processing.In the design,the excellent technology frameworks such as Flume,Kafka and Spark in the big data ecosystem are adopted,and a series of optimization strategies are proposed.Finally,the collected large data sets are verified by the offline data processing ystem.The final result shows that the three-month data is predicted and analyzed.and the Deep Belief Network prediction model is compared with the Back Propagation network prediction model.The root error is reduced by about 0.01378,and the processing time of each batch task is about 4.5s in real time,the prediction accuracy is higher,and the real-time performance is better,thus achieving a good gas concentration prediction and reducing the occurrence of gas disasters.