Research On Dynamic Prediction Method Of Coal Seam Geological Model Based On LSTM

The realization of intelligent unmanned mining is a frontier technology in the field of international coal mining,and also a technological leap forward to solve the problems of safe,efficient and green mining in China's coal mines? The plan for manufacturing 2025 in China? emphasizes that the manufacturing industry should conform to the trend of "Internet plus" and take the integration of informationization and industrialization as the main line.This paper combines the initial coal seam geological model data with the historical cutting data of the shearer,puts forward the dynamic prediction method of the coal seam geological model,draws the three-dimensional prediction map of the coal seam,and designs the dynamic prediction software of the coal seam geological model.The main research results are as follows(1)The data acquisition of initial coal seam geological model and shearer historical cutting data are introduced;A series of data preprocessing are carried out on the historical cutting data and initial coal seam geological model data,including data feature extraction,data combination,and supervised sequence data conversion.There are two kinds of data combination methods.Based on the principle that the two kinds of data are independent and identically distributed,the first data combination method is proposed;On the basis of sequence prediction,regression elements are added,and the second data combination mode is proposed.Combined with the specific problems of coal seam thickness prediction and coal seam floor surface prediction,the corresponding data sets are divided and converted.(2)Based on LSTM,bilstm,convlstm and encoder decoder LSTM,the prediction models of coal seam thickness are established respectively.It is found that the prediction results of the four models have large errors when the super parameters are not optimized.By optimizing the super parameters of the four models,the root mean square errors of coal thickness prediction of the four models with data combination mode 1 are 0.05 m,0.053 m,0.05 m and 0.044 m respectively;The root mean square errors of the four models of data combination mode 2 are 0.051 m,0.051 m,0.051 m and 0.049 m respectively.Among them,encoder decoder LSTM model has the best prediction accuracy.(3)Based on LSTM,bilstm,convlstm and encoder decoder LSTM,the prediction models of coal floor surface are established respectively,and the super parameters of the prediction model are optimized by using the grid search method.The average absolute errors of the four models are 0.112 m,1.168 m,0.116 m and2.02 m respectively;The average absolute errors of the four models of data combination mode 2 are 0.594 m,0.983 m,0.074 m and 0.323 m respectively.Among them,the prediction accuracy of convlstm model is the best.(4)The prediction results of coal seam thickness and coal seam floor surface are superimposed to obtain the prediction results of coal seam roof surface,and the influence of coal seam thickness prediction on the prediction results of coal seam roof surface is discussed.The conclusion of data combination one is that within the error range of(0,0.1m),the prediction error of coal seam thickness has the effect of error mutual growth on the prediction error of coal seam roof surface,and within the error range of(0.3m,0.4m),the prediction error of coal seam thickness has the effect of error mutual elimination on the prediction error of coal seam roof surface;The conclusion of the second data combination method is that the error is within the range of(0,0.05m),and the prediction error of coal seam thickness plays an important role in the prediction error of coal seam roof surface;In addition,an interactive software system for predicting coal seam geological model is designed and implemented in detail.