Study On Coal And Gas Outburst Prediction Method Based On Multi-model Fusion

Coal and gas outburst is a phenomenon in which gas and broken coal suddenly gush out of coal mine in large quantities,which is a major natural disaster in underground production.Improving the accuracy of coal and gas outburst prediction has important practical significance for the safe production of coal mines.In recent years,related researches have made remarkable progress in accuracy of prediction based on machine learning.However,the current researches generally face a problem of poor quality of the original data set used for prediction,which is manifested as insufficient sample size,one-sided sample source,high vacancy rate and redundancy of features.In order to solve the above problems,a coal and gas outburst prediction method based on multiple models fusion is proposed,which obtains comprehensive and sufficient training samples through web mining technology,mainly including Feature Selection based on AHP combined Gini and domain Experience(AGE),Fractional Imputation based on Multiple Model Blend(FIB)and an Enhanced Scorecard Model using Integrated Learning Algorithm as Amendment.Firstly,full coal safety production news are crawled by web and structured data are obtained by regular expression;secondly,AGE is used for feature selection on the converted high-dimensional data set to reduce data dimension and eliminate redundant features;then,FIB is used to impute missing values;finally,an enhanced scorecard model with additional items is established.The experimental results show that AGE feature selection algorithm increases the accuracy of prediction by an average of 5.70%;FIB imputation algorithm improves the accuracy of prediction by 3.61%,AUC by 0.74%,Recall by 0.82%;the enhanced scorecard model using CatBoost predicted results as additional items performs best,with accuracy improved by 0.88%,AUC by 4.08%,Recall by 17.47%;meanwhile,the method of using AGE,FIB and enhanced scorecard model is the best among all comparison algorithms,with accuracy,AUC and Recall reaching99.57%,98.69% and 99.73% respectively.This paper has 22 figures,13 tables and 68 references.