Research On The On-line Discriminant Theory And Methods Of Mine Inrush Water

With the development of coal mining in depth,the geological conditions of the mine are becoming more and more complex,The problem of water hazard in the safe mining of coal mines are more severe due to the development of water-conducting fissures and multi-level goafing.Different types of water bodies may cause mine water inrush,but its harm degree and prevention methods are different.For example,in all coal mine water inrush accidents,the occurrence rate of water inrush accidents caused by water in goaf and the casualty rate of personnel are all up to 90%.It is of great significance for the prevention and management of water inrush disaster to identify the water inrush source type quickly and accurately.At present,the identification of mine water inrush source mainly depends on the artificial underground sampling,which is sent to the laboratory for concentration detection of six ion.It is impossible to provide real-time warning of the change of Water Inrush Source type,and it is difficult to meet the timeliness requirement of water disaster prevention and control.In this paper,an on-line monitoring and warning system model of mine water inrush source type based on“Internet of things”is proposed.Combined with the hydrology and water quality test data of Xinji No.2 Mine in Huainan Mining area,and the concentration ofCa^2?10??Na^?10??HCO₃^-?Cl^-as well as pH value,conductivity,fluorescence spectrum were used as the discriminant factors of water source type,which provided the theoretical basis for sensor selection.An improved SFLA-BP algorithm is proposed for the comprehensive identification of water source types on the basis of the space-time law of water migration,in which the coordinates of the monitoring points?longitude,dimension and depth?,monitoring time and discriminant factors are taken as parameters,and the need for real-time and accuracy of on-line monitoring and early warning of water inrush sources is also considered.The depth belief net?DBN?and convolutional neural network?CNN?are applied to water source classification for the first time,and the recognition accuracy of CNN water inrush discrimination model is further improved on the premise of guaranteeing the recognition efficiency.The main contributions and innovations of this paper include:1.Limited by the development of sensor technology,currently available on-line monitoring ion sensors include Na^?10??Ca^2?10??Cl^-andHCO₃^-,and lack of on-line monitoring ofSO₄^2-andMg^2?10?,the discriminant factors that can be replaced should be studied.The Xinji No.2 Mine in Huainan mining area is taken as the research object,and according to the hydrological history account data of the mining area and the newly collected water samples,The Ion characteristics,origin and formation mechanism of groundwater in five aquifer groups of this region are analyzed quantitatively,such as pore water of Cenozoic loose stratum,water accumulation in goaf,sandstone fissure water of coal measure formation,karst fissure water of limestone of Taiyuan formation and karst fissure water of limestone of Ordovician system.In order to compensate for the lack of expression of aquifer water features caused by the shortage ofSO₄^2-andMg^2?10?,conductivity is introduced as an on-line discriminant factor.Except for the Cenozoic water,the remaining aquifers have the following relationship:Ca^2?10?andMg^2?10?are highly correlated,and SO₄^2-is highly linearly related to conductivity,Ca^2?10??Na^?10??HCO₃^-?Cl^-.In addition to the Cenozoic water,Ca^2?10??Mg^2?10??Na^?10??HCO₃^-?SO₄^2-and Cl^-can be replaced by conductivity,Ca^2?10??Na^?10??HCO₃^-and Cl^-.Considering that the goaf water is mostly acidic,pH is added as one of the online discriminant factors.In order to measure the water characteristics comprehensively,dissolved organic matter?DOM?was increased as an on-line discriminant feature,and the DOM was detected by fluorescence spectroscopy.2.Aiming at the similarity problem of water body type in Xinji No.2 Mine,and in order to improve the accuracy of water source discrimination,laser induced fluorescence technique was introduced to detect the DOM content in water.By studying the variation law of fluorescence spectrum characteristics of aquifer water body,it is found that the overall variation trend and classification characteristics of fluorescence spectrum curve of water samples are obvious,which provides a new way for water source identification.Further quantitative analysis of spectral characteristics and chemical oxygen demand?COD?,oxidation reduction potential?ORP?and TDS in water showed that the fluorescence intensity of water samples was positively correlated with DOM.Then,the effects of turbidity,pH and temperature on the spectral characteristics of the fluorescence were measured,and the fitted curves were respectively used for spectral curve correction.3.A set of on-line discriminant monitoring system of mine water inrush water source based on Internet of things is designed.The monitoring node is composed of conductivity sensor,pH sensor,Ca2,K Na,HCO3-,Cl-concentration sensor and fluorescence spectrum sensor,which realizes real-time data transmission,visual display and on-line identification of water source type.4.From the observation of inorganic content and DOM,the identification model of mine water inrush source and the construction method of comprehensive discriminant model were studied.Firstly,the identification model of Mine Water Inrush Source Based on support vector machine?SVM?and BP neural network is constructed.On this basis,an improved Shuffled Frog Leading Algorithm?SFLA?-BP neural network model for mine water inrush source identification is proposed to solve the problem of local optimal solution caused by the randomness of initial weights of BP neural network.Compared with BP and SVM,the recognition rate of the optimized SFLA-BP has been improved.The comprehensive discriminant model was used to continuously monitor the water source in goaf No.150801 of a mine,and the online recognition accuracy rate reached91%.In the network structure,a comprehensive discriminant model based on deep belief network?DBN?deep learning is proposed.The results show that the highest recognition of inorganic matter is 59.37%,the average recognition rate of organic matter is 81.07%,and the comprehensive discrimination recognition rate is up to 94.01%.5.An improved recursive average first-order hysteresis smoothing method is proposed to solve the random high-frequency wave interference in the spectrum.Aiming at the problem that the cumulative contribution rate of principal component analysis?PCA?is less than 85%,which is not enough to fully express spectral characteristics,the autocorrelation calculation of the filtered fluorescence spectrum is carried out,and the two-dimensional autocorrelation fluorescence spectrum is obtained.The results show that the fluorescence spectra can filter out the noise interference and show obvious differences for different water bodies.Based on the fluorescence spectrum,a discriminant model of convolutional neural network?CNN?Water inrush source type based on deep learning model framework was constructed.The method directly recognizes the characteristics of the fluorescence spectrum,avoids the lack of PCA dimension reduction,the average recognition rate reaches 94.95%,and the highest recognition rate reaches 98%,which provides a new idea for the on-line mine water source type identification method.