Maximum Gradient Decision-Making for Railways Based on Convolutional Neural Network

AbstractMaximum gradient (MG) decision-making is among the most important in railway alignment design because it greatly affects railway transport capacity, construction costs, and operation costs. However, existing studies mainly focus on optimizing railway alignment for cases with predetermined MG values. Studies on MG decision-making are rare. In this study, a data-driven method is proposed for MG decision-making based on a convolutional neural network (CNN). A total of 246 existing established railway cases are compiled whose total length is nearly 30,000 km. Factors that influence MG decision-making are characterized as a multichannel image. The 246 railway cases are characterized as 246 multichannel images and cropped into 20,170 images. Using the cropped images as the input data, a CNN model is designed to explore the relations among the factors and the MG value in order to make MG decisions. The method’s performance is tested on 36 existing railway cases. The test accuracy is 94.44%, which demonstrates that the proposed method can match experienced human experts in determining MG values for railway cases.