| In terms of the international and domestic energy production and consumption,the position of coal is crucial.China's coal production conditions are complex,and all kinds of disasters and accidents happen frequently,among whith gas accidents are most likely to cause casualtie.Therefore,it is very important to predict and early warn gas disasters.The previous research on the early warning of underground dynamic disasters is mainly focused on the direction of gas,stress,electromagnetic and microseismic.Gas is a very important basis for identifying disaster,and it is necessary to study the early warning methods based on abnormal gas emission.At present,a mine mainly conducts disaster prediction and warning through the monitoring system based on whether the gas concentration exceeds its limit,but it cannot accurately judge the degree of risk of the working face,even some overrun is caused by the breakage,bump and other interferences.Therefore,studying the change of gas concentration curve in a certain period of time is more in line with the actual needs of disaster warning.The traditional methods to describe the gas concentration curve are divided into two kinds.One is to fit the gas concentration.The mathematical equation is used to describe the change of gas concentration curve in a period of time,and then predict the gas concentration in the next time,so as to achieve the purpose of disaster forecast.However,this method needs to use different formulas,parameters,and coefficients to set up a large number of equations when describing the gas curves under different types and degrees.Even if the high accuracy equations to describing all kinds of gas curves in underground mine can be found,only very similar gas curves can be predicted.With the advancing of working face,the high precision equation library is easy to fail when the similar degree of gas curves is low.What's more,the underground conditions are complex and there are many interference factors,which will bring many unfavorable factors to the application of mathematical equations.Another is to extract features of gas concentration time series,and use statistical indexes to describe gas time series,such as time and frequency domain indexes(the mean,variance,RMS,peak value,the value,kurtosis,skewness,etc.)and statistics indexes(non-stationary,nonlinear,non-Gauss,etc).With the purpose of identifying the risk degree,different models of characteristic indexes are used to distinguish the gas curves of different risk levels.However,the statistical indexes are limited,and it is not easy to describe the weak difference of gas concentration curves.Besides,experts estimate that the dangerous degree of gas concentration curve from underground monitoring is not determined by calculating statistical indicators,but by identifying observed visual characteristics of the gas curve's overall trends,changes,evolution,etc.In this paper,a relatively mature image recognition technology has been adopted to identify the risk of gas abnormal emission in the heading face.Image recognition technology is effective in fingerprint recognition,face recognition,remote sensing image recognition,historical relic restoration and other fields.It can extract and identify various macro features and weak features.As each one-dimensional time series feature can be mapped into a two-dimensional image feature,but many two-dimensional image features cannot be used to describe the one-dimensional time series.After converting the time series of one-dimensional gas concentration into a two-dimensional gas concentration image,more features of the gas concentration curve can be extracted to fully and accurately identify the danger of abnormal gas emission.In this paper,the image characteristics of the gas curves after the abnormal gas emission in the heading face are firstly analyzed and the image recognition synthesis model of the gas abnormal emission in the heading face based on ReliefF,Dimension Interval and Support Vector Machine(ReliefF-Dimension Interval-SVM)is established,which is used to identify the dangerous degree of gas abnormal emission in the gas image.Secondly,using the self-made tunneling face gas abnormal emission test system,the effects of different gas emission volumes and different wind speeds on gas concentration curves are studied.A whole gas image is segmented into a large number of local images through image segmentation and image resampling.The integrated model of image recognition is used to dynamically identify the risk of abnormal emssion of the experimental gas local images,through which the optimization of this comprehensive model is realized.Thirdly,in order to verify the effect of the optimized model,a geometric model for gas emission and migration is developed.The effect of different gas emission volumes and different roadway wind speeds on the gas emission concentration is simulated using ANSYS_FLUENT software.The model dynamically identified the dangerous degree of the numerical simulation gas local image and the model's effect is verified.Finally,the optimized image recognition model of abnormal gas emission for the tunneling face is used to dynamically identify the dangerous degree of abnormal gas emission in the gas concentration image obtained from the 1011 roadway in Wolong Lake Coal Mine,and the field application of the model is verified.The main results of this paper are as follows:(1)The comprehensive model of gas abnormal emission image recognition model based on ReliefF-Dimension Interval-SVM is proposed.In this model,four normalization options,four kinds of kernel function algorithms and two support vector classifier algorithms are initially set up,so a total of 32 sub-model configurations are provided.After the model optimization,only 1 sub-model with the highest recognition accuracy and the strongest robustness is retained,and the others are eliminated.By the built-in ReliefF algorithm,the model evaluates the weight of each characteristic index to identify the risk of gas emission,and eliminates the characteristics of low weight,and realizes the feature dimension reduction.By the built-in feature dimension interval,the model avoids the risk of low sensitivity or failure in the single dimension model.The ReliefF algorithm and the dimension interval of the model make it unnecessary to configure the parameters of the artificial model when updating the training sample.(2)The optimization and evaluation method of image recognition model for gas abnormal emission in heading face is put forward.Taking into account the actual needs of coal mine disaster early warning,the evaluation indexes and evaluation criteria for evaluating the performance of the image recognition model of gas abnormal emission in heading face have been determined.The evaluation indexes are the recall,accuracy,cross validation accuracy(CV),precision,specificity,F1-Measure and G-mean.The criterion is that all indexes are as high as possible.When the performance of each evaluation index can not be all taken into account,the recall is the most important,especially the recall rate of high risk images,and the following are the accuracy of model recognition,the cross validation of model training,and the other indexes.Based on this,the recognition performance of the 32 sub-models is evaluated to determine the optimal configuration of the model.(3)The image recognition characteristic index library for gas abnormal emission in heading face is set up.Seventy visual characteristics are excavated,and the characteristic index library is used to describe gas concentration curve,and it can distinguish the weak correlation and difference among gas images.The characteristic index library includes 3 kinds of characteristic indexes:(1)area characteristics: the area,axis direction of minimum external moment,minimum convex polygon solid degree,rectangle,centroid coordinate,maximum inner tangent circle,etc.(2)boundary characteristics: the circumference,boundary diameter,sphericity,variation coefficient,standard deviation of boundary and center distance,Fourier descriptors,etc.(3)moments characteristics: central moments,Hu moments,Zernike moments,etc.(4)The typical signal database for experimental abnormal gas emission is established based on the experimental research.A test system for abnormal gas emission of gas concentration change characteristics in heading face is designed and built,as well as the influence of different gas emission volumes and different wind speeds on the curve after the abnormal gas emission in the heading face.The test results show that:(1)the trend of the curve after abnormal gas emission is consistent,first rapidly rising and then slowly descending to the initial level;(2)when the initial gas emission pressure is certain,the amount of gas emission is certain,the gas concentration rising speed is not related to the wind speed.The greater the wind speed,the smaller the peak concentration,the peak time and the back time;(3)when the wind speed of roadway is certain,if the initial pressure is larger,the amount of gas emission is larger,the concentration rise velocity,the peak concentration and the back time are both larger.The gas emission volume represents how fast the gas concentration rises,and represents the dangerous level of gas image,based on which the typical signal library of experimental gas abnormal emission under different dangerous levels is set up.(5)The accuracy and reliability of the image recognition integrated model is optimized based on the experimental gas images.The accuracy and reliability of the comprehensive model are optimized by using the experimental gas images.The integrated model is finally set up to the standardization method of Min-Max,the linear kernel function,the standard C-SVC classifier and the dimension interval of N=11,22,,30.This model is used to identify whether the global image is easy to occur gas outburst with the accuracy rate of 100%,and the model is used to identify which category the global image belongs to when it comes to 4 types of dangerous degrees of gas abnormal emission(high,medium,low and no dangerous)with the accuracy rate of 100%.After that the global image of the experimental gas is divided into local images.A large number of local images are formed by resampling.The model is used to dynamically identify the risk of gas abnormal emission in the experimental local image.The overall accuracy rate is 91.60%.The recall of 4 types of local images with high,middle,low and no danger are 95%,59.09%,98.67% and 88.13%,respectively.The model can identify quickly.The risk degree of experimental gas abnormal emission local image.(6)The geometric model of gas emission and migration and diffusion is set up and the typical signal library for abnormal gas emission of numerical simulation is established.The geometric model of gas abnormal emission and diffusion is set up.Combined with the fluid numerical simulation software of ANSYS_FLUENT,the gas concentration characteristics under different gas emission volumes and different wind speed conditions are simulated.The numerical results show that:(1)after gas emission,the gas curve rapidly rises,and then slowly decreases to the initial level with the continuous flow of fresh air flow;(2)when the wind speed is certain,the higher the gas emission pressure,the greater the emission volume,the greater the gas rising speed,the greater the peak,and the longer the back time;(3)when the gas emission volume is certain,the greater the wind speed,the earlier the concentration began to rise,the smaller the peak concentration,the shorter the peak time,and the shorter the back time.The gas emission volume depends on dangerous degree of gas image,which has a consistent meaning with experimental gas curve image.Therefore,the gas emission is used to characterize the danger degree of gas image,and the typical numerical signal library of abnormal gas emission is established.(7)The accuracy and reliability of the image recognition integrated model is verified based on the simulated gas images.The accuracy and reliability of the comprehensive model are verified by using the simulated gas images.The model is used to identify which category the global image belongs to when it comes to 4 types of dangerous degrees of gas abnormal emission(high,medium,low and no dangerous)with the accuracy rate of 100%.The model is also used to dynamically identify the risk of gas abnormal emission in the simulated local image.The overall accuracy rate is 88.57%.The recall of 4 types of local images with high,middle,low and no danger are 91.01%,84.00%,98.70% and 80.50%,respectively.The model can identify quickly the risk degree of experimental gas abnormal emission local image.(8)The accuracy and reliability of the image recognition integrated model are further verified based on the field gas image.Using the gas image from the 1011 roadway in Wolong Lake Coal Mine,the accuracy and reliability of the optimized image recognition model is further verified.The model is also used to dynamically identify the risk of gas abnormal emission in the field local image.The overall accuracy rate is 98.39% and the recall of 4 types of local images with high,middle,low and no danger are 90.24%,91.57%,63.16% and 98.60%,respectively.The model can identify quickly the risk degree of site gas abnormal emission local image.Combined with the gas monitoring system,the new gas image is dynamically generated,and its risk degree is quickly identified in order to achieve the early warning of risk degree in the working surface.In this paper,specific research has been carried out using theoretical research,experimental research,numerical simulation,and field application.The risk of abnormal gas emission reflected by the gas concentration curve after gas emission has been studied.Through real-time monitoring of gas data,and timely adding of new concentration data and eliminating old concentration data to the gas local image,the gas local image is updated in real time and its degree of danger can be dynamically monitored and identified by the optimized model.The model can realize rapidly recognize the dangerous degree through the features of gas curve image shortly after the beginning of the abnormal emission without having to wait for the concentration to rise to a considerable extent.The research results will help to increase the recognition accuracy of abnormal gas emission in the working face,to identify the degree of danger of abnormal gas emission in the working face early,to enrich the characteristics of gas emission and their recognition theory of early warning,to improve the monitoring and early warning technologies of underground dynamic disasters. |
PDF Full Text Request