Research On Prediction Of Coal And Gas Outburst Based On SKPCA And NEAT Algorithm

Coal mining continues to develop to the deep,compared with the shallow underground environment is worse,put forward higher requirements for coal and gas outburst warning.In this paper,the gas outburst prediction index is selected from many aspects,a new outburst prediction index system is put forward,and a coal and gas outburst prediction model based on SKPCA and NEat algorithm is established.Through the analysis of outburst mechanism and influencing factors,15 secondary prediction indexes which can reflect outburst information are selected from gas factor,coal factor,geological structure factor,in-situ stress factor and other factors.In view of the information overlapping and non-linear problems among the indicators,this paper uses kernel principal component analysis to process the outstanding data.Because the principal component data is obtained by linear combination of all the original data,the interpretability of principal component is poor.In this paper,the sparse method is introduced to sparse the kernel principal components.The load coefficient of the unimportant elements in each principal component is reduced to zero,so that each principal component only retains the most important element in the corresponding original index.Reduce noise interference in gas outburst data,and the principal component information is more representative.After calculating the contribution rate of principal components,the first four principal components are selected as the input of the prediction model,which reduces the complexity of the model and improves the generalization ability.Overfitting and local optimal solution often appear when neural network is used to predict coal and gas outburst.This paper introduces a new neural network optimization algorithm.NEAT algorithm adopts the principle of genetic evolutionary algorithm,which can not only optimize the weight parameters of the neural network,but also optimize the connection topology of the network.By encoding the neural network connections,each connection is labeled with an innovation number,which is then arranged as an individual’s genome.Mutation,crossover and other operations of individual genes are inherited to the next generation,and the optimal prediction network structure is found after multi-generation evolution.The optimized network structure is not the traditional full connection layer,and the internal parameters are greatly reduced to avoid the phenomenon of over-fitting.In order to avoid the large difference of individual fitness in the early stage of optimization,which makes the network unable to search for optimization globally,dynamic individual fitness parameters are introduced in this paper to reduce the gap of individual fitness in the early stage.Divide coal and gas outburst prediction data into training set and test set,and build SKPCA-NEAT prediction model.Input data for training and prediction,and compare the results with the prediction results of SKPCA-BP,PCA-NEAT,SKPCA-GA and SVM models.The experimental results show that SKPCA is more effective in processing coal and gas outburst data than PCA.The SKPCA-NEAT model has better prediction accuracy and stronger model generalization ability.The paper has 23 figures,12 tables and 61 references.