| Nowadays,how to ensure its efficient and safe mining of coal mines has always been a topic of great concern to scholars at home and abroad with coal as the main energy source.In recent years,with the continuous expansion of the depth and breadth of coal seams in the process of coal mining,the difficulty of dealing with water inrush disasters in mines has also increased,which seriously threatens the safety of coal mine production.Moreover,the rapid and accurate identification of mine water inrush sources is a crucial link in the prevention and control of mine water hazards.Although there are various methods and means for mine water inrush source identification at this stage,the water inrush source identification method based on a single technical means has problems such as poor applicability and large errors.Therefore,it is still urgent to break through the existing water inrush source identification methods and build a more accurate and rapid water source identification technology system.Selected the Huangyuchuan coal mine in Inner Mongolia Autonomous Region,the Donghuantuo mine in the Fengrun District of Tangshan City,Hebei Province,and the Baode coal mine in Baode County,Shanxi Province as the research objects,summarized the current model types for identifying mine water sources by combining the spectrophotometric experimental techniques with the UV-Vis spectrophotometer as the measuring instrument and the characteristics of water chemical field.Proposed the model for identifying water sources under different circumstances.The main research results of the thesis were as follows:(1)Based on the five categories of water temperature and level data,trace ion content data,isotope content data,water chemistry data and spectral data,summarizes the theories related to mine water source identification,analyzed the basis for identifying mine water sources in detail,summarized the characteristics and applicability of the existing water source identification models,and proposed a model for the selection of mine water source identification methods with wide application range.(2)UV-Vis spectrophotometer was used to measure individual water samples from different aquifers to obtain spectral data based on absorbance value.The water samples were analyzed in the experiment.Taking the absorbance of water sample measured by spectrophotometer as the spectral data set,the spectral data set obtained from the experiment was analyzed and processed in detail.The spectral data of water sample measured by spectrophotometer was easy to operate and easy to use.The data can be read within 1 minute,and the spectral data can be quickly obtained.In addition,combined with the established water source identification model,the availability of spectral data and the accuracy of the water inrush source identification model were verified.(3)Propose a single water source identification model based on full-spectrum data.Taking Huangyuchuan mining area as the research object,on the basis of fully considering the influence of geographical factors,with the help of spectrophotometer and other test equipment,collected the water sample data of different aquifers in the mining area,and the spectral data of water samples of different aquifers in the study area were obtained experimentally.At the same time,established the probabilistic neural network optimized by elite genetic algorithm(EGA-PNN),compared with the probabilistic neural network(PNN)and the probabilistic neural network optimized by genetic algorithm(GAPNN).On the one hand,the results verify that the spectral data obtained by spectrophotometry can be used as the data index of water inrush source identification.On the other hand,demonstrate the reliability of the EGA-PNN identification model.Moreover,the EGA-PNN recognition model can quickly achieve global convergence in the process of determining the smoothing factor,and the most stable σ value can be obtained,its recognition accuracy was stable at 95.69%.(4)Propose a single water source identification model based on hydrochemical data.Based on 195 actual monitoring hydrochemical ledger data of Donghuantuo mining area from the mid-1990 s to 2019,selected the sample data parameters such as conventional hydrochemical ion concentration,PH value and total hardness in each aquifer.Used Aquachem water chemical analysis software,SPSS 25 and python 3.10 to draw the Piper triple diagram,Durov diagram,stone diagram,conventional ion box diagram and Pearson correlation analysis diagram,respectively,so as to analyze the hydrochemical characteristics of each major water-filled aquifer and the hydraulic relationship between each aquifer.It was found that the water-chemical data of the roof aquifer of No.5 coal in Donghuantuo mining area have changed significantly since 2015.At the same time,based on the water chemical data,the gradient lifting tree identification model(PSO-LightGBM)optimized by particle swarm optimization algorithm was constructed,and compared with the Gradient Lifting tree algorithm model(LightGBM),random forest model(RF),classification and regression tree model(CART).The results show that the recognition accuracy of PSO-LightGBM was higher than that of LightGBM,CART,RF and other models,which can meet the demand of water source identification in the process of mine construction and was suitable for mining areas with rich hydrochemical data.In addition,since the groundwater environment in study area No.5 coal changed significantly before and after coal mining,it was verified that the change of hydrochemical data had a slight impact on the accuracy of water source identification results.Therefore,it was necessary to monitor and analyze the changes of water samples in this area regularly,so as to improve the accuracy of identification results of water inrush sources.(5)Proposed a mixed water inrush source identification model based on full spectral data.Taking Baode mining area as an example,collecting and integrating four kinds of water samples in this area,including Permian sandstone water,Permian goaf water,Carboniferous sandstone water and Ordovician limestone water.The water sample data were tested in the laboratory for a single water sample,and mixed water test experiments with different proportions were obtained at the same time.In addition,established a fast discrimination model of mixed water inrush source based on Bat algorithm optimized radial basis function neural network(BA-RBF).It was compared with radial basis function neural network(RBF),genetic algorithm optimized radial basis function neural network(GA-RBF)and particle swarm optimization optimized radial basis function neural network(PSO-RBF).Verified the applicability of BA-RBF identification model.The results show that the BA-RBF identification model has the highest recognition accuracy for the spectral data of water samples in different aquifers,and the accuracy can reach 96.67%.Analyzed and compared the results of a single water source identification model based on full spectrum data and hydrochemical data.Analyzed and compared the data of mixed water samples with different proportions,and the model can get good and stable identification results.In addition,the mixed water identification model established at present can judge that the water sample was mixed water quality,and can also quantify the proportion of different sources in the mixed water sample. |
PDF Full Text Request