Application Research Of The Prediction Model For The Coal Working Face Roof Pressure Based On GA-BP Neural Networks

This paper in view of this problem about major source of hidden danger caused by the coal roof weighting from fully mechanized working face,building theprediction model of the pressure from working face roof on GA-BP neural networks to forecast the regularity of coal seam roof weighting in DaTong mining area by the way of theo-retical analysis,data collection and pretreatment,model evaluation and optimization and so on.The main research results are as follows:(1)The training samples of prediction model are 60 sets of datas for geological structure of coal working face and process parameters as well as coal roof weighting collected in DaTong mining area.Sorting the results used grey correlation degree to analyze the correlation degree between the influencing factors and the coal roof weighting.The main influencing factors of the coal roof weighting are working face inclination length,advance speed of work line,roof condition of coal seam,mining height,coal thickness,change rate of coal seam dip angle,thickness of the coal direct roof and main roof,buried depth,change rate of buried depth and coal thickness,coal seam dip angle.Finally,normalizing the sample datas.(2)According to the problem to be solved,building the prediction model based on BP neural network to forecast the coal roof pressure and determining the structure of BP neural network and the value of each parameter with empirical formula and repeated experiment.The target outputs of the BP neural network model are respectively the first weighting intensity,the first weighting interval,the periodic weighting intensity and the periodic weighting length of the working face roof.Each group does 30 times of trainings,and the results show that the average determination coefficient of the four target outputs are respectively 0.751,0.74,0.826,and 0.837,and the average comprehensive determination coefficient of the four groups was 0.788;The relative error between the predicted value and the real value is compared and analyzed.And the result shows that BP can predict the roof weighting condition of the coal working face roof to some extent,but the predicted result is not stable.(3)According to the existing problems of BP neural network,adopting the improved genetic algorithm to optimize it,and building the prediction model based on GA-BP neural network to forecast coal roof pressure.Through the comparative analysis of the predicted values of the model,the average determination coefficient of the BP neural network model is 0.798,and the average determination coefficient of the optimized GA-BP neural network model is 0.896;The average probability that the relative error of BP and GA-BP model prediction results is respectively less than 3% accounted for 54.93% and 69.33%,and the average probability that the relative error is less than 5% respectively accounted for 75.18% and 89.38%.The training results show that the prediction effect of the model based on optimizing GA-BP neural network is obviously higher than that of BP neural network,and it can well predict the first weighting intensity,the first weighting interval,the periodic weighting intensity and the periodic weighting length of the working face roof.(4)The GA-BP neural network model was verified by 5 groups of measured data of fully mechanized mining working face(without participating in model training)in DaTong mining area.Results of prediction model based on the optimized GA-BP neural network show that the relative error of four target values about coal working face roof pressure is respectively less than 2%,4%,3% and 4%,and the total relative error is less than 4% within the error range of industrial production permit.so the prediction model based on GA-BP neural networks to forecast the pressure from mine working face roof can be applied to predict the coal roof weighting.