Vertical alignment optimization of mountain railways with terrain‐driven greedy algorithm improved by Monte Carlo tree search. (22nd September 2022)

• Record Type: Journal Article
• Authors: Zhang, Hong; Song, Taoran; Schonfeld, Paul; Pu, Hao; Li, Wei; Hu, Jianping; Liu, Linya
• Is Part Of: Computer-aided civil and infrastructure engineering. Volume 38:Number 7(2023)
• Page Start: 873
• Page End: 891
• Publication Date: 2022-09-22
• Subjects: Civil engineering -- Data processing -- Periodicals; Computer-aided engineering -- Periodicals; 624.0285
• DOI: 10.1111/mice.12923
• Languages: English
• ISSNs: 1093-9687

Abstract

Abstract: Vertical alignment design is an important process for railway construction which fundamentally affects the infrastructure investment cost. Determining an optimized vertical alignment is a challenging task since the objective function is non‐linear, non‐differentiable, and quite unsmooth. Great efforts have been invested in solving the vertical alignment optimization problem and many methods have been proposed. However, for vertical alignment designs in complex mountainous regions, the terrain conditions impose great difficulties and, hence, many bridges and tunnels are generally required. Thus, reasonably locating bridges and tunnels along the entire alignment (EA) is a major concern that deserves further investigations. To solve this problem, this study develops a terrain‐driven greedy algorithm improved by Monte Carlo tree search (T‐GRA‐MCTS). A terrain‐driven method is proposed to determine the number and longitudinal distribution of vertical points of intersection (VPIs). In order to trade off the local section of an alignment versus the EA when optimizing each VPI along the alignment to locate bridges and tunnels reasonably, an MCTS is employed and integrated with a GRA. The basic MCTS is modified for vertical alignment optimization with a novel equation for computing the upper confidence bounds for trees and a customized termination criterion is provided. A real‐world railway case is used to demonstrate the effectiveness of the proposed method. The results show that the T‐GRA‐MCTS performs better than a greedy search method without MCTS or a widely used nature‐inspired algorithm (i.e., a particle swarm optimization). Moreover, it can find a less expensive solution than the one designed by experienced human engineers.