| Coal mining is bound to have an impact on the land resources and ecological environment around the mining area.And the destruction of the ecological environment will also restrict the mining and utilization of coal.In order to reasonably and scientifically mine and reduce the damage of subsidence,accurate prediction of surface movement and deformation becomes the key to solving the above problems.Therefore,how to select accurate prediction parameters is very important and also becomes the subject of this paper.For the study of the method of calculating the predicted parameters,a large number of scholars not only dissect the essential relationship between parameters and geological mining factors but also generally establish the corresponding parameter prediction model.However,these models are often more applicable to the mining areas selected by scholars and can not be effectively extended.In view of the limitations of existing predictive parameter methods,this paper uses machine learning algorithm to build a probability integral method parameter prediction model of "fuzzy clustering first,then parameter prediction".The specific research ideas are as follows:(1)After a brief introduction of the geological and mining factors that affect the probability integral parameters,principal component analysis(PCA)is used to reduce the dimension of the collected geological and mining factors to filter the redundant information in the original data and reduce the impact on the fuzzy clustering and parameter prediction results.(2)To explore the law of change between similar geological and mining conditions and parameters of the probability integral method,the rock movement observation stations are fuzzy clustered.The simulated annealing(SA)algorithm and the genetic algorithm(GA)are combined to improve the fuzzy C-means clustering(FCM)to form a SAGA-FCM clustering algorithm.The dimension-reduced data is input into the SAGA-FCM algorithm to obtain the cluster center and membership matrix.(3)The observation stations are classified according to the clustered membership matrix,and three types of rock movement observation stations are obtained.To explore the impact of data before and after classification on parameter prediction accuracy,the two groups of data are selected from the same test set,and back propagation neural network,support vector machines(SVM),and generalized regression neural network(GRNN)are used for accuracy verification to obtain the prediction results of probability integral parameters.The result analysis mainly includes two aspects:the comparative analysis of the predicted results of classified and unclassified data,and the comparative analysis of the prediction results of three machine learning models.The experimental results show that the classification data can greatly reduce the fluctuation range and improve the prediction accuracy.Under the unclassified data,the prediction accuracy of the three machine learning models is compared,and the experimental results show that GRNN has the highest prediction accuracy;Under the classified data,the prediction accuracy of three machine learning models is compared.The experimental results show that GRNN has the highest prediction accuracy.To sum up,the GRNN model classified by SAGA-FCM is feasible,stable,and accurate in predicting the probability integral parameters,and the average relative error of the subsidence coefficient q,the horizontal moving coefficient b,the tangent of main influence angle tan?,the inflection offset s/H prediction are less than 5%.This also provides theoretical support and basis for accurate prediction,disaster warning,and rational mining of surface movement and deformation and prediction in mining area.Figure [21] table[20] reference [82]... |
PDF Full Text Request