Ventilation resistance coefficient prediction of tunnels based on GA-projection pursuit regression
The Genetic Algorithms(GA)-projection pursuit regression method is employed to predict the ventilation resistance coefficient in mines.The method consists of two steps:building a projection regression model and using a genetic algorithm to find the best projection vector.GA is used to search for the optimal projection direction in the projection pursuit,enabling the transformation of the influencing factors of the high-dimensional ventilation resistance coefficient into a lower-dimensional space.This algorithm combines the data dimension reduction characteristics of projection pursuit with the global search capability of genetic algorithms.This study considers different airflow densities and two types of roadways:supported and unsupported.Factors such as the diameter of the wooden column,longitudinal diameter,roadway cross-section area,roadway perimeter,shotcrete part of the perimeter,metal beam thickness,column thickness,shed spacing,roadway height,and roadway width are used for training.The model is established using 142 learning samples collected from the field,and its performance is validated using 42 actual testing samples.The predicted results are compared with those obtained from Principal Component Analysis(PCA)prediction and Backpropagation Neural Network(BPNN)prediction.The outcomes reveal that the average error of this method is 1.76％,with a maximum error of 1.07％.To further validate the model's reliability,an in-depth analysis and validation are conducted focusing on a specific category of roadway,demonstrating the method's consistency and accuracy.Remarkably,the average error rate is found to be only 1.73％,with a negligible maximum error of 0.79％.Additionally,this study researches the ventilation resistance coefficient prediction of unsupported roadways,resulting in an average error of 1.73％and a maximum error of 0.79％.The model's predictive accuracy is minimally affected by different airflow densities.Regardless of various types of supports or the selection of a specific support type,this method exhibits superior predictive accuracy compared to Principal Component Analysis and Backpropagation Neural Network methods.
NETL
NSTL
万方数据
