Study On Fully Mechanized Mining Face Roof Weightings Intelligent Warning Model Based On Multi-Source Information

The intelligent early warning of working face roof pressure is an important part of the intelligent development of coal mine,which is of great significance to the safe and efficient production of coal mine and the study of the migration law of overburden.The existing intelligent early warning model of loading is mainly based on two ideas.One is based on the roof loading theory to complete the static prediction of loading step distance;Second,from the mine pressure data,complete the analysis and early warning of the working face pressure.However,the cycle pressure of working face is a process of dynamic change,and the static prediction cannot realize the dynamic early warning.In addition,the use of undug ore pressure data cannot effectively complete the pressure early warning of working face.Aiming at these problems,the use of fully mechanized working face support and roof state intellisense system(SSRI)to dig deeper into the pressure data,uses data mining after working face to pressure trend prediction model is established,and according to the working pressure influence factors working face to pressure interval prediction model is established,on the basis of the future pressure prediction model and to interval prediction model,built based on multi-source information for pressure intelligent dynamic early warning model.The main research results are as follows:(1)The machine learning method combining genetic algorithm(GA)and random forest(RF)is used to predict the pressure step of working face based on coal seam dip Angle,roof condition,basic roof thickness,direct roof thickness,working face length and working face advancing speed.Sixty sets of data collected from related papers were used to generate data sets for training and validation of the model.The determination coefficient(R2)and root mean square error(RMSE)were selected as evaluation indexes to compare the prediction performance of the RF model and the mixed GA-RF model.Sensitivity studies were also conducted to assess the importance of input parameters.From the importance score,the coal seam dip Angle(0.2543)has the greatest influence,and the importance of the other influencing factors fluctuates within the range of 0.15.The characteristics obtained by partial dependence graph are consistent with the results of theoretical analysis.The results show that the proposed GA-RF model can predict the static pressure step distance of the working face.(2)Machine learning method combining Bayesian optimization(BO)and longshort term memory network(LSTM)was used to predict the pressure trend based on the last cycle time weighted working resistance,the number of safety valve opening,the initial support force,the final resistance,and the resistance increasing rate of each bearing stage of the support.The prediction performance of BP model,LSTM model and BO-LSTM model were verified and compared by means of mean square error(MSE)and mean absolute error(MAE).Sensitivity studies were also conducted to assess the importance of input parameters.The final resistance and the upper working cycle time are weighted to pressure the most important parameters in trend prediction.At the same time,the other features extracted by SSRI are also essential for the prediction of pressure trend.The results show that the proposed BO-LSTM model has high accuracy and good robustness in predicting pressure trend.(3)the future pressure interval prediction model and to the trend prediction model is built based on the multi-source information fusion of the working face for pressure intelligent early warning model,with a face as an engineering background,based on the actual data is verified,increases the relative error(RE)as evaluation indexes,combining image contrast method to the model prediction effect is assessed.The example analysis shows that this model can be effectively implemented to suppress early warning.The paper has 37 pictures,3 tables,and 79 references.