Research On Prediction And Early Warning Of Mine Gas Disaster Based On Recurrent Neural Network

Coal is an important basic energy for China's economic development.While promoting the steady development of the national economy,it is easy to cause gas disasters in the process of energy mining due to rough economic development and backward energy mining methods,which poses a serious threat to coal mine production and miners' life safety.Therefore,in order to avoid the occurrence of gas disasters and prevent them in advance,coal mine gas disaster prediction technology has become the focus of research in the field of coal mine safety.However,due to the problems of short prediction time and low prediction accuracy,the existing gas disaster prediction technology can not meet the needs of actual coal mine disaster prediction.Aiming at the problem of how to improve the prediction accuracy of gas disaster,this paper carries out in-depth research on the prediction method of gas disaster in coal mine.The main research contents are as follows:(1)Based on information fusion perception technology,an entropy weighted data fusion algorithm is proposed to solve the problems of coal mine data sources and complex formats.The entropy weighted data fusion algorithm is used to fuse the gas concentration data on the working surface,the gas concentration data at the inlet and the gas concentration data at the upper corner.The traditional recurren t neural network model is used to predict the data before and after fusion,and the mean square error of input data and output data is calculated.Through experimental comparison and analysis;the prediction accuracy of the data after entropy weighted fusion is 39.59%higher than that of the data without fusion in the training set and 36.16%higher in the test set.(2)Aiming at the problem that the accuracy of traditional prediction methods in gas disaster prediction needs to be improved,a gated cycle element neural network(Adam-GRU)gas concentration prediction model based on adaptive moment estimation was proposed.The model first obtains the gas concentration data after fusion,then carries on the sp atial reconstruction processing to the data,constructs the training set of the network prediction model;Then,the adaptive moment estimation optimization algorithm was used to optimize the GRU model parameters,the mean square error was selected as the loss function of the prediction model,and dropout technology was introduced to reduce the overfitting phenomenon in the prediction model.Finally,the model training is carried out to establish the gas concentration prediction model based on Adam-GRU network.Compared with the support vector regression(SVR)model,the back propagation neural network(BP),recurrent neural network(RNN)and long-term and short-term memory network(LSTM),the Adam-GRU model achieves the minimum error on the test set.The errors of SVR,BP,RNN and LSTM models are reduced by 58.23%,37.67%,27.88%and 25.57%respectively,and the operating efficiency of Adam-GRU model is improved by 18.41%compared with LSTM model.(3)To solve the problem that gas concentration will be affected by various factors in the mines,a PSO-CNN-aGRU gas concentration prediction model was proposed based on integrated learning.Firstly,particle swarm optimization(PSO)algorithm is used to optimize the model parameters of CNN structure,and the local variation trend and correlation characteristics of gas data are extracted automatically.Then the GRU neural network model based on adaptive moment estimation is used to predict the temporal data based on the correlation characteristics.In the final experiment,sequence data features extracted by the PSO-CNN model were fed into the SVR,BP,and LSTM models after being processed by the full connection layer,respectively.Compared with the other three models,the mean square error of the PSO-CNN-aGRU model was reduced by 55.47%,43.55%,and 23.34%,respectively,and the operating efficiency of the PSO-CNN-LSTM model was increased by 15.15%.