Prediction Of Coal And Gas Outburst Based On Improved Artificial Neural Network

Coal and gas outburst prediction based on improved artificial neural network(ANN)is a kind of coal mine production safety accident which is common and will have serious consequences once it happens.The traditional methods for predicting coal and gas outburst always lead too deviation results.This paper analyzes the advantages and disadvantages of the conventional methods for predicting coal and gas outburst and introduces a neural network with excellent ability to solve complex linear or nonlinear problems.In view of disadvantages of the present standard BP and RBF neural network in the prediction of coal and gas outburst existing in the research,for example,the BP neural network weights threshold is determined randomly,number of hidden layers and neurons is difficult to determine,sample characteristics of coal and gas outburst is generally not after pretreatment,based on the idea of enumeration method,the improved algorithm of the BP and RBF neural network and principal component analysis(PCA)are put forward to address the pretreatment of characteristics of coal and gas outburst samples.And the improved BP and RBF neural network,particle swarm optimization(PSO)and genetic algorithm(GA)were used to select the parameters of the two neural networks.Based on the comprehensive consideration of 13 factors of outburst coal seams,46 coal seams with gas were trained and predicted by the above improvement methods.The results show that:(1)When BP neural network was used,PCA had a negative effect on the pretreatment of coal and gas outburst sample characteristics,while when RBF was used,PCA did not significantly improve the prediction accuracy.(2)The improved BP and RBF neural network indicates that the prediction accuracy is the highest when selecting the combination of burial depth,fault number,surrounding rock assemblage,failure type of coal body,coal thickness,coal seam dip angle,initial velocity of gas dispersion and characteristic coefficient of firmness.This combination comprehensively covers the gas characteristic,the coal body characteristic and the geological structure three parts information.(3)In this prediction of coal and gas outburst,the optimization effect of PSO on the initial weight threshold of BP neural network is better than GA algorithm,and the prediction performance of the improved BP neural network adaptive determined topology structure is better than that of the traditional empirical formula.(4)The improved BP and RBF neural network showed good prediction performance after studying the coal and gas outburst samples.Considering that the internal complexity of RBF neural network is much higher than that of BP neural network under the same accuracy requirement.When the number of coal and gas outburst samples is small,the improved RBF neural network can be used;when the number is large,the BP neural network is more appropriate.Figure[21]table[3]reference[64]...