Research On Gas Content Prediction Method Based On Improved Extreme Learning Machine

Since the coal seam gas prediction problem is a complex nonlinear prediction problem,the application of intelligent machine learning algorithms to coal seam gas prediction has high applicability.Combined with the actual situation of the 15 th mining area of Jiulishan Mine,the data of the influencing factors of gas content are summarized through relevant data to establish a prediction sample data set,and the grey relational analysis method is used to screen the data samples as the input of the prediction model.In order to break through the limitation of the performance improvement of a single intelligent algorithm,the simulated annealing algorithm is connected in series with the genetic algorithm and mixed into a genetic simulated annealing algorithm.Since the gas content prediction problem is a nonlinear prediction problem,this paper introduces a kernel function based on the extreme learning machine to improve it.Taking the mean square error of the model predicted coal seam gas content value and the field measured gas content value as the fitness function,the GASA-KELM gas content prediction model was established by using the designed GASA algorithm to improve the KELM initial parameter assignment method.Predictive models for networks and support vector machines.The 15-mining area data set is divided into training set and test set by 10-fold 10-fold cross-validation method,and the prediction model is trained 100 times with samples to complete the modeling.The model is used for on-site gas content prediction in 15 mining areas,and the results show that:(1)In the optimization process of Rastrgin’s function,particle swarm optimization,genetic algorithm,simulated annealing algorithm and genetic simulated annealing algorithm completed the convergence in 146,147,153 and 78 iterations,respectively,and the variances of the results were 3.05,4.65,6.25 and 1.59,but the quality of GA and SA optimization is not as good as PSO and GASA,so the GASA algorithm designed in this paper has stronger function optimization ability and search stability.(2)During the model training process,the GASA-BPNN model failed to meet the target requirement after 800 iterations.The GASA-SVM model and the GASA-KELM model reached the target minimum error at 610 and 378 iterations,respectively.The cross-validation gas content prediction total average The relative errors are 13.11 %,9.68 %,and 8.29 %,respectively,and the overall mean variance of the test set is 3.15,2.73,and 2.52,respectively.Therefore,GASA-KELM has the advantages of fast convergence speed of mean square error,low total average relative error of cross-validation prediction and strong stability.In the process of sampling test,GASA-KELM model has high prediction accuracy and strong generalization ability to new sample data.Advantage.(3)In the process of on-site prediction application,the GASA-KELM model established in this paper has high gas content prediction accuracy,strong prediction stability and generalization ability,which is of great significance for accurately predicting gas content and proposing effective gas control measures.