Accelerate transit network design problem-solving based on large-scale smart card data and graph-clustering decomposition

Abstract

Transit network optimization problems, such as the Transit Network Design (TND) problem, are generally difficult to be solved within reasonable computational time, especially for large real-world instances potentially containing thousands of bus stops and millions of stop-level OD pairs. Modifications made in existing TND studies for improving the computational performance could vary across different solution methods, assumptions and problem modeling, raising concerns about their transferability and implementation difficulty. This study presents a general framework to accelerate the TND problem-solving for large instances by decomposing the transit network into smaller subnetworks. We first proposed a modified Louvain algorithm to generate an optimal network partitioning based on transit passenger flows extracted from large-scale smart card data. A solution method (i.e., a Genetic Algorithm, GA) is then implemented to address the TND problem in each subnetwork separately and parallelly, thereby saving much computational time. Transit routes generated and optimized in each subnetwork are then combined into one route set as the solution for the entire transit network. The experiment on a well-known benchmark dataset shows that the proposed method, even incorporated with only a basic version of GA, can yield solutions better than most of its existing competitors with an outstanding computational performance. Moreover, the results indicate that our method could speed up the basic GA nearly 1,000 times faster on the iteration process. Another experiment on a real-world network instance (i.e., the Nanjing bus) shows that our method can optimize the currently operating routes, reducing the average travel time by 18.6% and the number of transfers by 17.7%. Code and data for the implementation of the GCGA framework and metrics calculations are provided at https://github.com/Radar-Lei/TBS-D-2023-00544R1.