A data-driven time-dependent routing and scheduling for activity-based freight transport modeling
van Lint, J.W.C.
Truck flow patterns can best be understood through the study of freight activities on a transportation network. These activities are often observable after the execution of tours resulting from a routing and scheduling optimization process where the traditional Capacitated Vehicle Routing Problem (CVRP) is used to minimize multiple objectives under spatial, temporal, and vehicle constraints. Researchers have admitted in recent years that there should be distinctions between freight and passenger transport modeling due to the complexity associated with logistics (Gonzalez-Calderon and Holguín-Veras, 2019). The most important distinction is to involve the tour behavior of carriers considering all spatial and temporal constraints (You et al., 2016). In current tour-based models, tours are constructed through incremental trip chaining in such a way that the next destination in a tour is estimated based on the conditional probability of the current stop. These types of models are based on choice modeling and provide agencies with descriptive statistics and insights into freight demand. However, discrete choice methods are not subject to constraints and therefore cannot capture spatial-temporal characteristics of tours (Heinitz and Liedtke, 2010). Additionally, trip chaining decisions are made once at the tactical level and hence incrementally reconstruction of the tour at the operational level is not identical to the tour planning process in reality. To deal with this problem, micro-simulations have used a family of vehicle routing problems (VRP) to model the pickup and delivery of carriers within a multi-agent microsimulation framework (Donnelly et al., 2010). Although normative models can perfectly capture space-time constraints, their outcome could deviate from observed tours due to heterogeneity in the tour planning decisions of planners. To deal with this issue, You et al. (2016) proposed an inverse optimization approach to calibrate a family of VRP using the method of successive averages. They estimate the weights of a weighted sum of multiple objectives from a set of observed tours. Parameter estimation of these methods requires fully observed truck movement patterns. However, tour data, if available, are often partially observable to traffic agencies and policymakers due to privacy issues. To the best of our knowledge the development of a method to calibrate the VRP model based on shipment flows with partially observed tour data has not yet been explored and is therefore of interest to the current paper. To address this gap, we propose an efficient surrogate-based optimization method to calibrate a VRP model based on partially observed tour information.
Activity-Based Freight Modeling
To reference this document use:
Data-Driven Vehicle Routing Problem
Mobility & Logistics
Extended abstract submitted for presentation at the 11th Triennial Symposium on Transportation Analysis conference (TRISTAN XI), June 19-25, 2022, Mauritius Island, 1-4