A methodology for scheduling within‐day roadway work zones using deep neural networks and active learning

Abstract

City infrastructure agencies routinely implement road projects that address various elements of urban infrastructure. The majority of these projects are shortterm in nature (e.g., utility repair), as they are completed in a few hours within 8:00 a.m. to 5:00 p.m. of a workday. The implementation of these projects during working hours, in spite of the inconvenience imposed on road users, helps the agency avoid extra labor costs associated with nonregular working hours. Careful scheduling of these projects can prevent unduly increased travel delays (road users' interest) while keeping project costs low (the agency's interest). This study presents a bi-level framework for scheduling short-term urban road projects that analyzes the implicit tradeoffs between the two stakeholders' interests. The upper-level model establishes the optimal schedule considering the project characteristics, such as cost and duration. The lower-level model captures the dynamic user equilibrium conditions that yield the road users' path and departure time choices. The bi-level model is a mixed-integer program with nonlinear constraints. Recognizing the relatively low efficiency of traditional solution methods, this paper proposes a deep-neural-network-ensemble-assisted active learning (DN2EA2L) algorithm and adopts a fixed-point algorithm for solving the bi-level model. The numerical experiment uses the Sioux Falls network to demonstrate the efficiency of the DN2EA2L, compared to conventional metaheuristic methods. It is shown that travel time increases due to the project implementation during the peak hours can outweigh the agency's saving in wage costs. Further, it is shown that a significant reduction in the road users' value of time enables the agency to schedule more projects during regular working hours. INTRODUCTIONRoad project implementation continues to pose a major cause of nonrecurring congestion in urban road networks (FHWA, 2022). These projects, regardless of their shortterm or long-term nature, cause an increase in road user travel times. Since the pioneering research of Adeli

Exploration on Intelligent Interaction Design and User Experience of VR Platform Based on Intelligent Perception and Deep Learning

“…If the sample has poor linearity, it can be directly subjected to a pull-down slope detection experiment. On the contrary, it is necessary to observe whether there is a correlation and a directional change relationship between them, in order to determine whether they have the same or similar characteristics, and provide evaluation conclusions [19][20]. It is based on a large amount of distributed memory and describes typical knowledge points in the network by classifying and merging various types of deep objects.…”

365‐day sectional work zone schedule optimization for road networks considering economies of scale and user cost

“…Miralinaghi et al (2022) considered the benefits of contract bundling for multiple links in a road network and employed the elitist nondominated sorting genetic algorithm to solve the problem. Saneii et al (2023) used the dynamic user equilibrium of the traffic assignment on the road network, wherein the traffic demands change according to the time, and utilized deep neural network ensemble-assisted active learning to solve the problem.…”

A weakly supervised active learning framework for non-intrusive load monitoring

“…Although weak labels reduce the labelling effort, the number of time periods that need to be labelled for fine-tuning could be still large. Active Learning (AL) approaches [42,43] are used in literature to optimise data selection for artificial intelligence algorithms by choosing the most informative data, and that way reduce the number of data segments needed to be labelled and added to the training dataset, but without compromising the algorithm performance [44]. AL approaches have been widely used for deep learning algorithms recently [45].…”