关于共享电动车停放优化与运维策略的研究

Research on Parking Optimization and Operation and Maintenance Strategies for Shared Electric Bikes

摘要: 共享电动车作为解决城市短途通勤和“最后一公里”交通需求的关键工具，推广中面临高峰时段车辆短缺、非高峰时段闲置、区域供需失衡等核心痛点。传统人工调度响应迟缓，难以适配动态需求。本研究基于美国华盛顿特区及周边CapitalBikeshare系统2024年120万条骑行数据，构建“精准识别–动态调度–智能决策”全链条优化体系，运用多维度供需平衡评估框架、K-means聚类算法、遗传算法、LSTM需求预测等方法，系统研究共享电动车配置与调度问题。文章针对供需失衡站点识别，融合空间属性、用户行为与资源状态构建多维度评估框架，通过K-means聚类(K = 3，轮廓系数0.6719，Calinski-Harabasz指数684.5430)与多指标体系，识别出15个严重供不应求站点(集中于商圈、地铁口 + 商圈等)与12个供过于求闲置区域(集中于政府/行政中心等)。针对用户等待时间与车辆闲置矛盾，以小时骑行量80%分位数确定15:00~19:00为高峰时段，构建最小化“调度运输成本 + 用户等待时间成本”的整数线性规划模型，结合遗传算法优化路径，方案覆盖85%失衡站点，实现调度距离缩短23%、车辆闲置率降低37%、用户平均等待时间减少42%。针对动态决策需求，整合前两阶段成果，构建LSTM需求预测模型量化天气、节假日等外部因素影响，提出“弹性投放–闭环回收”策略，暴雨场景下用户等待时间减少52%，货车日均里程减少38%。研究为共享交通智能化升级提供可推广方案。

Abstract: Shared electric bikes, as a key tool to address the short-distance commuting and “last mile” transportation demands in cities, face core pain points such as vehicle shortages during peak hours, idle vehicles during off-peak hours, and regional supply and demand imbalances during their promotion. Traditional manual scheduling responds slowly and is difficult to adapt to dynamic demands. This study is based on 1.2 million cycling data from the CapitalBikeshare system in Washington, D. C. and its surrounding areas in 2024, and constructs a full-chain optimization system of “precise identification-dynamic scheduling-intelligent decision-making”. It employs methods such as a multi-dimensional supply and demand balance assessment framework, K-means clustering algorithm, genetic algorithm, and LSTM demand forecasting systematically study the configuration and dispatching issues of shared electric vehicles. For the identification of sites with supply and demand imbalance, this article integrates spatial attributes, user behaviors and resource states to construct a multi-dimensional evaluation framework. Through K-means clustering (K = 3, contour coefficient 0.6719, Calinski-Harabasz index 684.5430) and a multi-index system, fifteen severely oversupplied stations (concentrated in business districts, subway entrances + business districts, etc.) and twelve oversupplied idle areas (concentrated in government/administrative centers, etc.) were identified. In response to the contradiction between user waiting time and vehicle idleness, the peak period from 15:00 to 19:00 was determined based on the 80% percentile of hourly riding volume. An integer linear programming model that minimizes “dispatching transportation cost + user waiting time cost” was constructed. Combined with the genetic algorithm to optimize the path, the scheme covers 85% of the imbalanced stations. Achieve a 23% reduction in dispatching distance, a 37% decrease in vehicle idleness rate, and a 42% reduction in the average waiting time for users. In response to the dynamic decision-making requirements, the achievements of the first two stages were integrated to construct an LSTM demand forecasting model to quantify the influence of external factors such as weather and holidays. A “flexible deployment-closed-loop recovery” strategy was proposed, which reduced the waiting time for users by 52% and the average daily mileage of trucks by 38% in the rainstorm scenario. Research provides scalable solutions for the intelligent upgrade of shared transportation.