Application Of Support Vector Regression In Thickness Prediction Of Tectonic Coal

China's coal industry is in the forefront of the world,but gas outburst accidents always endanger coal mining.According to research,the development of tectonic coal affects gas outburst,and the structure of different types of tectonic coal will make the nature of gas occurrence different,and the thickness distribution of tectonic coal affects regional mining.Therefore,if the thickness of different types of tectonic coal can be accurately predicted,it will play an important role in the safe mining of coal mine.Seismic attribute data can be used to predict the thickness of tectonic coal,but the seismic attribute data will be disturbed by some restrictions,environmental impact and human factors,so the seismic attribute data often has a large noise.In view of this situation,the feature engineering method based on gradient lifting decision tree(GBDT),the feature engineering method based on principal component analysis and the feature engineering method based on denoising automatic encoder are used to process the seismic attribute data.Through comparison,the best performance of feature engineering method based on GBDT combining with new features is selected as the feature engineering method of seismic attribute data.The twin support vector regression(TSVR)is selected as the prediction model of coal thickness.Then,the GBDT-TSVR model is constructed.By comparing the experimental results of genetic algorithm and gray wolf optimization algorithm,gray wolf optimization algorithm is selected as the parameter optimization algorithm of the model to adjust the parameters of the model.After that,the horizontal comparison experiments were carried out with different support vector regression models,and then the longitudinal comparison experiments were carried out with other prediction models.The final experimental results verified the good generalization ability and reliability of GBDT-TSVR model.Finally,the prediction model of tectonic coal thickness based on GBDT-TSVR is applied to 8 coal seam of Luling coal mine,and the data of 19 mines in the mining area are predicted.At the same time,the least squares support vector regression with GBDT combining with new features and the multi-layer perceptron with GBDT combining with new features are used as the comparison model to make the prediction comparison.The comparison results verify that GBDT-TSVR model has better accuracy and generalization ability in the prediction of data beside mine and drill hole data.The seismic attribute feature engineering method based on the GBDT and the prediction model of different kinds of tectonic coal based on TSVR have higher accuracy and generalization ability,which can be extended to the actual thickness prediction of different kinds of tectonic coal.This paper contains 28 pictures,14 tables and 84 references.