Data-Driven Feeder-Bus Network Design with Multiobjective Optimization

Here you can find detailed results about the article "Data-Driven Feeder-Bus Network Design with Multiobjective Optimization".Abstract: Transportation systems increasingly rely on data-driven decision support tools to optimize multimodal transit operations. This paper presents an integrated framework for feeder-bus network design that combines machine learning-based demand synthesis with multiobjective evolutionary optimization. The methodology addresses a critical challenge in metropolitan transit planning: designing feeder routes that efficiently connect peripheral areas to high-capacity rail systems while balancing competing objectives of service coverage and operational efficiency. Using publicly available data (census indicators, socioeconomic indices, origin--destination surveys, and transit infrastructure) we develop two neural network models to predict trip destinations and public transport usage. These models power a Monte Carlo demand generator that produces temporally resolved origin--destination matrices at metropolitan scale. A tailored NSGA-II algorithm with station-preserving encoding and capacity-aware operators then optimizes feeder routes under vehicle capacity and duty-time constraints. Applied to the Monterrey Metropolitan Area (Mexico), the framework generates feeder networks that increase passengers served by approximately 28\% and reduce total travel time by 17\% compared to the current configuration, using the same fleet size. The approach demonstrates how accessible data sources and intelligent optimization techniques can substantially improve transit connectivity and accessibility in growing metropolitan areas with limited real-time data infrastructure.<br>Git Hub with NSGA-II code: https://github.com/Sam126RV/Optimization-of-Metro-Feeder-Lines-NSGAii.git

你可在此获取论文"Data-Driven Feeder-Bus Network Design with Multiobjective Optimization"的详细研究成果。
摘要：交通运输系统愈发依赖数据驱动的决策支持工具，以优化多模式公共交通运营。本文提出一种用于接驳公交(feeder-bus)线网设计的集成框架，该框架将基于机器学习的需求合成与多目标进化优化相结合。该方法解决了大都市交通规划中的一项关键挑战：设计可高效连接外围区域与大运量轨道交通系统的接驳线路，同时平衡服务覆盖范围与运营效率这两大相互竞争的目标。
我们利用公开可用的数据源（人口普查指标、社会经济指数、起讫点(Origin-Destination, OD)调查数据以及公共交通基础设施数据）开发了两款神经网络模型，用于预测出行目的地与公共交通使用情况。这些模型驱动蒙特卡洛(Monte Carlo)需求生成器，可生成大都市尺度下具有时间分辨率的起讫点矩阵。随后，一种经过定制化改造的NSGA-II（Non-dominated Sorting Genetic Algorithm II，非支配排序遗传算法II）算法，采用保留站点的编码方式与考虑容量的操作算子，在车辆容量与执勤时长约束下对接驳线路进行优化。
该框架应用于墨西哥蒙特雷大都市区，在保持车队规模不变的前提下，生成的接驳线网较当前配置可使服务乘客量提升约28%，总出行时间减少17%。该研究方法证明，在缺乏实时数据基础设施的发展中大都市地区，可通过易得的数据源与智能优化技术，显著改善交通连通性与可达性。
含NSGA-II代码的GitHub仓库：https://github.com/Sam126RV/Optimization-of-Metro-Feeder-Lines-NSGAii.git