Research On Dynamic Prediction Method Of Coal Seam Surface Morphology Based On Spatio-Temporal Hybrid Model

High-precision coal seam geological information on the fully mechanized face is the key to implementing intelligent unmanned mining.However,the current coal seam surface models constructed have low vertical accuracy,which cannot meet the actual requirements of intelligent mining.To solve this problem,this study aims to propose a dynamic prediction method for coal seam surface morphology based on a spatio-temporal hybrid model by using initial coal seam three-dimensional model data combined with cutting state data generated by the shearer during dynamic mining.This model is capable of dynamically predicting the coal seam morphology of the unmined area based on the mined coal seam data.The coal seam morphology is predicted both in the direction of working face advance and along the working face direction.Our research focuses on providing support for achieving high-precision coal seam state information on the fully mechanized face.The main research work and achievements are presented as follows:（1）In this study,raw cutting state data from the onboard sensors of a shearer were collected and obtained on the 18201 fully mechanized face.Through data conversion,function fitting and difference method,the state data with exact correspondence in time is obtained;it is combined with the mechanical structure parameters of coal mining machine to calculate and solve the cut-off trajectory according to different working conditions;and it is processed into the same data type as the top and bottom plates of coal seam established at the initial stage.The abnormal cut-off data is processed by ordinary kriging difference method,and the cut-off trajectory surface of 764 m～923.2 m in the direction of working face advance is finally obtained.Using the truncated surface as the real coal seam surface.The distribution characteristics of the coal seam surface were analyzed,and the vertical coordinates of the coal seam multi-source data,the slope(z-value,k_{YO Z},k_XOZ)along the working face advance direction and the working face direction were extracted as the characteristics of the response coal seam surface.The coal seam data along the working face advance direction and the working face direction are divided into data sets and corresponding data preprocessing respectively,and three data combinations are given in order to fully combine the two types of coal seam data.Lastly,the final sliding window method was used to construct the coal seam data as supervised sequence data.（2）Based on the spatio-temporal hybrid model,the dynamic prediction models of coal seam surface morphology with convolutional neural network-long short-term memory network-attention（CLA）and convolutional neural network-gated recurrent unit-attention（CGA）were developed along the working face advancement direction.The model is tested with real coal seam data.The z-value andk_YOZ are used as model feature inputs to demonstrate the capability of the proposed model to predict coal seam surface morphology.Using the established CLA and CGA models,extensive experiments are conducted based on three different data combinations,and the results show that the CLA model has the best effect in predicting the upper coal seam surface by using the upper coal miner cut-off trajectory;the CGA model has the best effect in predicting the lower coal seam surface by using the lower coal miner cut-off trajectory combined with the initial coal seam floor data.Finally,the effectiveness of the Attention module is verified by using the ablation experiment.The results confirmed the effectiveness of the established models,and demonstrated the advantage of the models in dynamic prediction of coal seam surface morphology.（3）In this study,a dynamic prediction model based on CLA and CGA was established along the working face,and the model was tested with real coal seam data.Using z-value,k_YOZ andk_XOZ as model feature inputs to verify the prediction effect of the model.The results show that the CGA model has the best result in predicting the upper surface of coal seam by using the upper truncation trajectory;the CGA model has the best result in predicting the lower surface of coal seam by combining the lower truncation trajectory with the initial base plate data.The ablation experiments of CLA and CGA models respectively confirm that the added Attention module has significantly improved the prediction of coal seam surface.After comparing and analyzing the coal seam surface morphology prediction in the direction of working face advance and in the direction of working face,the optimal dynamic prediction model for coal seam surface morphology was established,and a more accurate dynamic prediction of coal seam surface morphology in the unmined area is achieved.This thesis contains 88 figures,26 tables,and 109 references.