Research On Distributed Gas Concentration Prediction Method With Coal Mine Big Data

Gas accident is the main threat to the safe production of coal mines.Once it occurs,it will cause huge economic losses and even casualties.Gas accidents are mainly caused by excessive gas concentrations.Gas is a flammable gas.If its concentration is too high,there is a risk of burning or even explosion.Accurate prediction of gas concentration in coal mines is an effective method to reduce the frequency of gas accidents.However,the performance of the traditional gas concentration prediction system has great limitations.The establishment of a more scientific and effective gas concentration prediction model is of great significance for guiding coal mine safety production.This thesis proposes a single hidden-layer random weight neural network(SRWNN)gas concentration prediction method.The method uses the structure of a single hidden layer feedforward neural network and introduces random input layer weights and hidden layer biases.It combines neural networks and lower and upper bound predictions,so that interval predictions are used instead of traditional point predictions.In addition,it uses the highly robust NSGA-II as the training algorithm,which simplifies the complex training process of traditional neural networks.In order to solve the problem of SRWNN training in the big data environment for too long,and problems in computing and scheduling when facing large-scale prediction tasks,this thesis further studies the distributed gas concentration prediction method of SRWNN in big data environment of coal mine,which proposes distributed model training on the Apache Spark and distributed real-time prediction on the Apache Storm.A large number of experiments prove that the method proposed in this thesis is superior to other comparison algorithms,and the proposed distributed gas concentration prediction method can significantly improve the performance of model training and real-time prediction.The main research contents and innovations are as follows:(1)Based on the actual coal mine scene,SRWNN is proposed.Previous gas concentration prediction model failed to make full use of the coal mine big data in the coal mine monitoring system,and only used the short-term gas concentration series for gas concentration prediction,so that the performance of the prediction model was low.Through multi-dimensional coal mine data correlation analysis,8-tuples for gas concentration prediction are proposed,and then the potential rules between the data are mined with the proposed neural network model,so that gas concentration prediction can be performed more accurately.(2)In order to better evaluate the gas concentration prediction model,this article introduces LUBE,which combines interval prediction with neural networks and applies it to gas concentration prediction.Then,several objectives of interval prediction optimization are proposed,and the multi-objective genetic algorithm NSGA-II is used to train the gas concentration prediction model.Compared with the traditional neural network training method,this method simplifies the complex parameter adjustment process of the neural network,and effectively prevents the model from falling into the local optimal solution during the training process.(3)The distributed application of the SRWNN in the big data environment of coal mine is realized.In order to reduce the time consumed for training the SRWNN in large-scale data samples,this thesis migrates the training process of the SRWNN to Apache Spark.In order to solve the problem of large-scale real-time prediction task execution and load balancing,this thesis transfers the real-time gas concentration prediction task based on SRWNN to Apache Storm.(4)In order to verify the effectiveness of the proposed method,we conducted a large number of experiments on 5 datasets,including real coal mine data from Zhujidong Mine.The experimental results show that the SRWNN proposed in this thesis is superior to other comparison models,and the application of the SRWNN in the big data environment can greatly reduce the model training time and achieve large-scale accurate real-time gas concentration prediction.