A data-driven approach to manpower planning at U.S.–Canada border crossings

We investigate the staffing problem at Peace Arch, one of the major U.S.–Canada border crossings, with the goal of reducing time delay without compromising the effectiveness of security screening. Our data analytics show how the arrival rates of vehicles vary by time of day and day of week, and that the service rate per booth varies considerably by the time of day and the number of active booths. We propose a time-varying queueing model to capture these dynamics and use empirical data to estimate the model parameters using a multiple linear regression. We then formulate the staffing task as an integer programming problem and derive a near-optimal workforce schedule. Simulations reveal that our proposed workforce policy improves on the existing schedule by about 18% in terms of average delay without increasing the total work hours of the border staff.     

The allocation problem of electric car-sharing system: A data-driven approach

Car-sharing is an emerging transportation mode with increasing applications of electric vehicles (EVs). One of the important issues for one-way electric car-sharing systems (ECS) is unbalanced vehicle distributions and high relocation costs. To improve its efficiency and overall profit, this research proposes a data-driven optimization model with the consideration of demand uncertainty. Firstly, a large amount of historical order data from an ECS company are analyzed to characterize the dynamics of the vehicles and the behavioral features of the users. An important observation is that the daily demand by users, i.e., pick-ups, follows Poisson distribution; and the arrival rates vary across time exhibiting four major temporal stages. Based on this observation, this research constructs the ECS reallocation problem as a data-driven optimization model which is a combination of a probability expectation model and a linear programming problem with real-time data as input. More importantly, different from existing research, this research formulates the profit as the mathematical expectation of a discrete random variable with uncertain consumer demands. This allows for a comprehensive consideration of all possible future demands. Furthermore, driving range constraint has been considered in the proposed model as EV is the focus of this paper. A linear solution method is proposed to obtain the global optimal. At the end, the model is validated using real data from 30 ECS stations. The results indicate the daily improvement of profit could be as high as 19.05% with an average of 10.16%.     

A Bayesian network approach for population synthesis

Agent-based micro-simulation models require a complete list of agents with detailed demographic/socioeconomic information for the purpose of behavior modeling and simulation. This paper introduces a new alternative for population synthesis based on Bayesian networks. A Bayesian network is a graphical representation of a joint probability distribution, encoding probabilistic relationships among a set of variables in an efficient way. Similar to the previously developed probabilistic approach, in this paper, we consider the population synthesis problem to be the inference of a joint probability distribution. In this sense, the Bayesian network model becomes an efficient tool that allows us to compactly represent/reproduce the structure of the population system and preserve privacy and confidentiality in the meanwhile. We demonstrate and assess the performance of this approach in generating synthetic population for Singapore, by using the Household Interview Travel Survey (HITS) data as the known test population. Our results show that the introduced Bayesian network approach is powerful in characterizing the underlying joint distribution, and meanwhile the overfitting of data can be avoided as much as possible.     

Predicting station-level hourly demand in a large-scale bike-sharing network: A graph convolutional neural network approach

This study proposes a novel Graph Convolutional Neural Network with Data-driven Graph Filter (GCNN-DDGF) model that can learn hidden heterogeneous pairwise correlations between stations to predict station-level hourly demand in a large-scale bike-sharing network. Two architectures of the GCNN-DDGF model are explored; GCNN_reg-DDGF is a regular GCNN-DDGF model which contains the convolution and feedforward blocks, and GCNN_rec-DDGF additionally contains a recurrent block from the Long Short-term Memory neural network architecture to capture temporal dependencies in the bike-sharing demand series. Furthermore, four types of GCNN models are proposed whose adjacency matrices are based on various bike-sharing system data, including Spatial Distance matrix (SD), Demand matrix (DE), Average Trip Duration matrix (ATD), and Demand Correlation matrix (DC). These six types of GCNN models and seven other benchmark models are built and compared on a Citi Bike dataset from New York City which includes 272 stations and over 28 million transactions from 2013 to 2016. Results show that the GCNN_rec-DDGF performs the best in terms of the Root Mean Square Error, the Mean Absolute Error and the coefficient of determination (R²), followed by the GCNN_reg-DDGF. They outperform the other models. Through a more detailed graph network analysis based on the learned DDGF, insights are obtained on the “black box” of the GCNN-DDGF model. It is found to capture some information similar to details embedded in the SD, DE and DC matrices. More importantly, it also uncovers hidden heterogeneous pairwise correlations between stations that are not revealed by any of those matrices.