Study On Surface Deformation Data Analysis And Prediction Model Of A Coal Mine In Shandong Province

As one of the energy minerals,coal resources play a significant role in the modern industry,whether heavy industry,light industry,energy industry,transport industry,etc.Due to the constant drilling of underground coal,it is inevitable that the ground level of the ground floor is going to sink,so it is very important to observe and predict the landscape of the ground and the deformation of the ground.Based on the research background of a coal mine in shandong province,the basic situation of the mining area and the layout and observation of the observation station are introduced.The sedimentary time series of the observation station is obtained by periodic repeated observation.Analysis of surface rock shift parameters by using SODP software.The noise-containing data is processed by means of the wavelet threshold denoising,and the optimal denoising combination method is selected by means of the control variable method,It is found that db3 wavelet base function,soft threshold,rigrsure threshold principle,scal=sln,1 layer decompose are best use by comparing RMS and SNR.To establish a time series prediction model for denoising data,and to use AIC criterion to set the order,at that time,AIC obtained the minimum value 196.062,namely,the time series model was ARMA(1,6),and the prediction results were compared with the measured values,confirming the prediction accuracy of time series model is higher.It establish a traditional BP neural network and analyzes that selection method of the number of hidden layer nodes.Using the variable gradient back propagation algorithm based on L-M optimize the traditional BP network,Results show that the BP network based on L-M can not only speed up the convergence speed,and the prediction precision is higher than that of traditional BP network.For BP network is difficult to select a right number of hidden layer nodes and the parameters of the model more faults,GRNN just a smooth parameters,thus at the same time using GRNN to forecast,Determine the smoothing parameter is 0.1,comparing the predicted results with the measured values,The results show that GRNN is fast and reliable for settlement prediction.Using the curve fitting toolbox in MATLAB to fit the denoising data,To compare the value of the root mean square error and mean absolute error,the selected exponential function for the optimal fitting curve.Using the exponential function to predict the settlement value,The results show that curve fitting can accurately predict the trend of surface subsidence.