Study On Multi-Source Heterogeneous Prediction Model Fusion Method And Its Application In Mine Subsidence Prediction

Mining subsidence is expected to have important theoretical and practical significance for guiding mining planning,safe production,"under-three" coal mining,and environmental management of mining areas.The prediction method of mining subsidence has always been a research hotspot in the field of rock movement control.The essence of mining subsidence is the complex mechanical deformation process under multi-factor stress such as geological mining conditions.The deformation process is highly nonlinear in time and space.Therefore,it is theoretically feasible to use the nonlinear prediction theory for mining subsidence.Literature research shows that some scholars have established some mining subsidence prediction models based on nonlinear prediction methods,and have achieved some experimental results.However,the following problems still need to be further studied:(1)The prediction models are mostly based on a single nonlinearity.It is constructed by the prediction method and is not suitable for mining prediction of the whole period of the surface movement(starting period,active period and decay period);(2)rarely considers the complementarity of the advantages and disadvantages of the classical nonlinear prediction method,and establishes multiple sources The model of mining subsidence prediction is integrated;(3)The nonlinear mining subsidence prediction model constructed has not taken into account the impact of data freshness,and the model is not expected to be robust.In view of the above problems,this paper carried out systematic research and compiled engineering application programs based on MATLAB platform.The research results enriched and perfected the theoretical theory of mining subsidence prediction.Through research,the main results are as follows:(1)Through the study of domestic and foreign literatures,the research status of mining subsidence prediction based on Strata Movement mechanism and the nonlinear prediction method based on the time characteristics of mining subsidence are analyzed in detail.Studies shows that the current research on mining subsidence prediction is mostly for the numerical simulation and prediction of the full cycle of mining subsidence,and there are few cases to accurately study predictive models by stages.Therefore,the settlement prediction method that accurately reveals the subsidence deformation mechanism of mining area needs further research.(2)The MS23 full-cycle data of the measured maximum subsidence point of the Guqiao North Mine Observatory was preprocessed,and the MS23 settlement data of the active period and the recession period were used to establish a typical nonlinear prediction model.The accuracy and stability of nonlinear prediction models at different stages of mining subsidence were verified by experiments.The nonlinear models were screened by qualitative and quantitative analysis methods.The results show that both the ARMA prediction model and the BP neural network model can predict the settlement of the monitoring points relatively accurately and stably during the active mining subsidence period;In the mining subsidence recession period,BP neural network prediction model,Kalman filter prediction model,ARMA prediction model and cubic exponential smoothing model can better predict the settlement of monitoring points.(3)Based on the multi-source heterogeneous fusion criterion and the nonlinear model weight determination criterion,the filtered multi-source heterogeneous model was established by using the filtered nonlinear prediction model.Taking the measured data of Guqiao North Mine in Huainan as an example,the prediction values of the nonlinear prediction model and the measured values were used to test.The results show that the multi-source heterogeneous model established by the minimum variance reciprocal method in the active period of mining subsidence is superior to the single prediction model in prediction accuracy and stability.The multi-source heterogeneous model established by entropy method in the recession period is superior to the single prediction model in prediction accuracy and stability,the multi-source heterogeneous model can extract the prediction information of the single nonlinear prediction model and fully integrate the advantages of the single nonlinear prediction model,which can effectively improve the prediction accuracy and stability.(4)The data fresh degree function with the mining subsidence characteristics of the mining area was studied to establish a multi-source heterogeneous model with the fresh degree of the data,and predicted the settlement value of the MS23 active period and the recession period of the maximum subsidence point of the Guqiao North Mine.The results show that due to the rapid development of the subsidence during the active period,the settlement changes obviously.Considering the degree of old and new of measured data can effective improve the prediction accuracy and stability of the multi-source heterogeneous model,in the recession period,the surface movement deformation process is stable and the trend tends to be stable.The multi-source heterogeneous model with data freshness(entropy method)has lower prediction accuracy than the multi-source heterogeneous model,but improves the robustness of prediction and prediction.Performance meets engineering application needs.Figure[32]table[36]reference[79]...