利用计算图上的多日交通数据估计概率动态起讫点需求
Estimating probabilistic dynamic origin-destination demands using multi-day traffic data on computational graphs
摘要：
系统级交通决策需要了解网络流量的日常变化，这就要求对网络上的概率动态出行需求进行准确的建模和估计。现有的研究大多估计确定性的动态起讫点（OD）的需求，而需求和流量的日常变化被忽视。动态OD需求的概率分布估计是具有挑战性的，由于时空网络的复杂性和高维问题的计算强度。随着海量交通数据的可用性和先进计算方法的出现，本文开发了一个数据驱动的框架，解决了概率动态起讫点需求估计（PDODE）问题，使用多日数据。不同的统计距离（例如，lp-norm、Wasserstein距离、KL散度、Bhattacharyya距离）来衡量交通状况的估计值与观测值之间的差距，发现2-Wasserstein距离在估计均值和标准差时达到了平衡的精度。该框架被铸造成计算图和重新参数化的技巧，同时估计的平均值和标准差的概率动态OD需求。我们证明了所提出的PDODE框架的有效性和效率在小型和真实世界的网络。特别是，它表明，建议的PDODE框架可以通过考虑需求变化来缓解过拟合问题。总体而言，开发的PDODE框架提供了一个实用的工具，公共机构了解需求随机性的来源，评估日常的网络流量的变化，并作出可靠的决策，智能交通系统。
英文摘要：
System-level decision making in transportation needs to understand day-to-day variation of network flows, which calls for accurate modeling and estimation of probabilistic dynamic travel demand on networks. Most existing studies estimate deterministic dynamic origin-destination (OD) demand, while the day-to-day variation of demand and flow is overlooked. Estimating probabilistic distributions of dynamic OD demand is challenging due to the complexity of the spatio-temporal networks and the computational intensity of the high-dimensional problems. With the availability of massive traffic data and the emergence of advanced computational methods, this paper develops a data-driven framework that solves the probabilistic dynamic origin-destination demand estimation (PDODE) problem using multi-day data. Different statistical distances (e.g., lp-norm, Wasserstein distance, KL divergence, Bhattacharyya distance) are used and compared to measure the gap between the estimated and the observed traffic conditions, and it is found that 2-Wasserstein distance achieves a balanced accuracy in estimating both mean and standard deviation. The proposed framework is cast into the computational graph and a reparametrization trick is developed to estimate the mean and standard deviation of the probabilistic dynamic OD demand simultaneously. We demonstrate the effectiveness and efficiency of the proposed PDODE framework on both small and real-world networks. In particular, it is demonstrated that the proposed PDODE framework can mitigate the overfitting issues by considering the demand variation. Overall, the developed PDODE framework provides a practical tool for public agencies to understand the sources of demand stochasticity, evaluate day-to-day variation of network flow, and make reliable decisions for intelligent transportation systems.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
