Car-sharing mobility is an emerging sustainable transportation mode, but it poses great challenges to operators and urban traffic management due to the imbalance between supply and demand across time and space. To address the problem, this research proposes a spatio-temporal rebalancing optimization framework for the urban car-sharing system (CSS) based on geospatial big data and machine learning. In the spatial dimension, we construct the urban car-sharing network data set using geospatial big data. The graph deep learning is used to mine the car-sharing space demand patterns for location planning. This data-driven graph neural network approach breaks through the limitations of complex mathematical models in the previous location planning and can cope with large-scale CSS in real time when data is available. In the temporal dimension, we construct a combined optimization model of dynamic relocation and pricing based on the optimized car-sharing station layout. A multi-threaded reinforcement learning algorithm is proposed to solve the optimal relocation and pricing scheme. Dynamic relocation and pricing strategies are obtained by reinforcement learning algorithms based on accumulated historical operational data and real-time market demand, aiming at maximizing profits and optimizing resource utilization. The simulation results show that the combined optimization model of dynamic relocation and pricing provides a more effective solution than the non-combined model. The proposed optimization framework provides systematic decision support for solving urban CSS supply–demand imbalance and yields extensive theoretical and practical implications, especially in urban traffic management.

中文翻译：

---

基于地理空间大数据和机器学习优化城市汽车共享系统：时空再平衡视角


汽车共享出行是一种新兴的可持续交通方式，但由于跨时空的供需不平衡，给运营商和城市交通管理带来了巨大的挑战。针对这一问题，本研究提出了一种基于地理空间大数据和机器学习的城市汽车共享系统（CSS）时空再平衡优化框架。在空间维度上，我们利用地理空间大数据构建城市汽车共享网络数据集。利用图深度学习挖掘汽车共享空间需求模式以进行选址规划。这种数据驱动的图神经网络方法突破了以往选址规划中复杂数学模型的限制，可以在数据可用的情况下实时应对大规模CSS。在时间维度上，我们基于优化的汽车共享站点布局构建了动态搬迁和定价的组合优化模型。提出了一种多线程强化学习算法来求解最优搬迁和定价方案。基于累积的历史运营数据和实时市场需求，通过强化学习算法得出动态搬迁和定价策略，旨在实现利润最大化和资源利用最优化。仿真结果表明，动态搬迁与定价组合优化模型比非组合模型提供了更有效的解决方案。所提出的优化框架为解决城市CSS供需失衡问题提供了系统的决策支持，并产生了广泛的理论和实践意义，特别是在城市交通管理方面。