| Coal mine gas accidents are an important content of safety production management,and the occurrence of accidents poses a great threat to the safety of workers.Due to the complex mechanism of gas accidents,prediction and abnormal identification of gas concentration are priority coping strategies for gas accident prevention and control.However,existing technologies have problems such as high complexity of prediction models,low prediction accuracy,simple abnormal identification methods,and poor scalability.Based on the multi source online monitoring data of coal mines,this paper conducts in-depth research on gas concentration prediction models and anomaly recognition methods,and designs a set of gas anomaly recognition processes and technical implementations.The main research conclusions are as follows:Theoretical analysis of the causes of abnormal gas emission in mining areas,comprehensive analysis of the structure of the underground ventilation system and the internal relationship between online monitoring indicators,collection of online monitoring data sets related to gas concentration,gray correlation analysis of the indicators and gas concentration,and extraction of indicators with a correlation degree above 0.9 to build a gas concentration impact factor set.Using entropy weighted data fusion technology to fuse and validate the gas concentration data of the upper corner,working face,and return air roadway at the data level.The results indicate that the fused gas concentration more comprehensively reflects the gas concentration in the mining area;Aiming at the non-linear characteristics of gas concentration influencing factors,the kernel principal component analysis(KPCA)method was used to fuse the characteristics of gas concentration influencing factors,and a comparative analysis was conducted with principal component analysis(PCA).The results showed that the kernel principal component analysis method fused the original features into two principal components,while reducing redundant information while retaining 91.87% of important information,resulting in better fusion results.The time series prediction model based on KPCA-GRU was constructed to predict the processed gas concentration data.Based on the training results and evaluation indicators,the number of neurons in the hidden layer,training batch and training times parameters in the model were determined.The comparison and analysis with the single GRU model and LSTM model showed that KPCA-GRU’s root mean square error,average absolute error and determination coefficient index results were better than the single model,It is proved that information fusion technology can improve prediction accuracy while simplifying model complexity.By introducing Bayesian inference theory,a gas concentration anomaly recognition method integrating KPCA-GRU-Bayes is proposed.Based on prior information,Bayesian estimation is performed on the mean and variance of the prior residual set,and the probability of abnormal gas concentration is quantified and identified through the calculation of Bayesian factors and inference of abnormal conditions.Based on the above technical theory,a set of abnormal identification process for coal mine online monitoring system is designed and applied to a coal mine engineering example for verification.The results showed that compared to the traditional threshold method,the early warning time was successfully achieved about 30 minutes earlier,providing time for advanced treatment of gas risk hazards.Figure [28] Table [12] Reference [87]... |
PDF Full Text Request