Personalized Freight Route Recommendations With System Optimality Considerations: A Utility Learning Approach
Personalized Freight Route Recommendations With System Optimality Considerations: A Utility Learning Approach
复制标题
DOI：
10.1109/tits.2022.3213773
复制
发表时间：
2023-01
影响因子：
8.5
通讯作者：
Aristotelis-Angelos Papadopoulos;Ioannis Kordonis;M. Dessouky;Petros A. Ioannou
中科院分区：
文献类型：
--
作者：
Aristotelis-Angelos Papadopoulos;Ioannis Kordonis;M. Dessouky;Petros A. Ioannou
交通拥堵会对经济和环境产生负面影响。在卡车流量大的地区，交通状况变得更加糟糕。在本文中，我们为卡车司机提出了一种协调的定价和路线方案，以有效地将卡车安排到网络中并改善整体交通状况。我们方法的一个基本特征是，我们根据驾驶员的个人路线偏好提供个性化的路线指示。与以前提供个性化路线建议的工作相比，我们的方法通过组合定价和路线方案优化了整个系统的总成本，满足平均财产的预算平衡，并通过保证卡车司机在他/她决定参与该机制时的预期总效用（包括付款）大于或等于他/她不参与时的预期效用，确保每个卡车司机都有动力参与所提出的机制。由于估计每个卡车司机的效用函数是计算密集型的，因此我们首先根据卡车司机对少量二元路线选择问题的回答将卡车司机分为不相交的集群，随后我们建议使用基于最大似然估计（MLE）原理的学习方案，该方案允许我们学习描述每个集群的效用函数的参数。然后使用估计的效用来计算具有上述特征的定价和路线方案。苏福尔斯网络的仿真结果证明了所提出的定价和路由方案的效率。
Traffic congestion has a negative economic and environmental impact. Traffic conditions become even worse in areas with high volume of trucks. In this paper, we propose a coordinated pricing-and-routing scheme for truck drivers to efficiently route trucks into the network and improve the overall traffic conditions. A basic characteristic of our approach is the fact that we provide personalized routing instructions based on drivers’ individual routing preferences. In contrast with previous works that provide personalized routing suggestions, our approach optimizes over a total system-wide cost through a combined pricing-and-routing scheme that satisfies the budget balance on average property and ensures that every truck driver has an incentive to participate in the proposed mechanism by guaranteeing that the expected total utility of a truck driver (including payments) in case he/she decides to participate in the mechanism, is greater than or equal to his/her expected utility in case he/she does not participate. Since estimating a utility function for each individual truck driver is computationally intensive, we first divide the truck drivers into disjoint clusters based on their responses to a small number of binary route choice questions and we subsequently propose to use a learning scheme based on the Maximum Likelihood Estimation (MLE) principle that allows us to learn the parameters of the utility function that describes each cluster. The estimated utilities are then used to calculate a pricing-and-routing scheme with the aforementioned characteristics. Simulation results in the Sioux Falls network demonstrate the efficiency of the proposed pricing-and-routing scheme.