Research On Prediction System Of Coal Mine Gas Emission Based On Deep Learning

The gas data collected from the coal mining site are non-linear,high-dimensional and fuzzy.For the time being,the large amount of information and laws hidden behind massive data have not been developed.Prediction is still insufficient in terms of reliability,accuracy and real-time.Deep learning can complete the feature extraction and transformation of massive multi-source heterogeneous data information in the mine through the processing phase of multiple internal nonlinear layers,and can realize the purpose of prediction through autonomous learning,providing technical support for the prediction and decision of gas accident.Therefore,the multi-hidden layer model constructed by deep neural network(DNN)is used to train the sample data and complete the autonomous learning process.The error analysis results show that the model has good performance.The encapsulation of Tensor Flow framework is realized to complete the design of gas data prediction system,which can make accurate and real-time prediction and provide technical support for intelligent mining.This paper mainly completed the following aspects of the research content:(1)By comparing the basic structure and training methods of several typical deep learning models,it is concluded that the DNN model is more suitable for the prediction of gas emission.(2)The influencing factors of gas emission are analyzed,and SPSS is used to perform Bartlett spherical test and KMO test on the relevant parameters.The test results show that the partial correlation between the selected parameters is weak,and the significance is 0,which meets the analysis conditions,with reliable reliability and validity.Thus,the basic input parameters of the prediction model are determined.(3)A gas emission prediction model based on deep neural network is proposed and applied to the prediction of gas emission in Gaohe Mine.The pre-processed data is divided into 80% as training set and 20% as test set,which are used for model training and model prediction performance evaluation respectively.The prediction results show that the overall prediction curve basically conforms to the changing trend of gas emission and can accurately reflect the change amplitude of the emission.(4)The effect of the model is checked by error analysis.The standarddeviation of the prediction error of the mining face is 1.776921,and the standard deviation of the prediction error of the tunneling face is 0.310803.The prediction error values are all distributed within the range of about 0,with a decreasing trend to both sides,indicating that the accuracy of the model is better.The applicability and superiority of this model for gas emission prediction is verified.(5)On the basis of completing the construction of the gas emission data prediction model of the Gaohe Mine,through the establishment of a distributed deep learning framework,we completed the design of an accurate and efficient gas emission prediction system with independent learning capabilities.