Evaluating and improving the performance of Iowa truck parking information and management system

Thumbnail Image
Date
2023-08
Authors
Yang, Yilun
Major Professor
Advisor
Dong-O'Brien, Jing
Chu, Lynna
Seeger, Christopher
Sharma, Anuj
Wood, Jonathan
Committee Member
Journal Title
Journal ISSN
Volume Title
Publisher
Authors
Research Projects
Organizational Units
Journal Issue
Is Version Of
Versions
Series
Department
Civil, Construction, and Environmental Engineering
Abstract
The surge in roadway truck freight has intensified the difficulties faced by truck drivers in locating safe and legal parking spaces. Parking shortage on the highway system not only creates logistical challenges, but also has substantial economic implications. To help drivers make safer, more efficient parking decisions and better utilize existing truck parking facilities, Truck Parking Information Management Systems (TPIMS) have been deployed in several states to provide real-time parking information to truck drivers through various communication channels. This dissertation first presents an anomaly detection method to identify sensor failures by tracking structural changes in time series parking data. In addition, a data dashboard is developed to evaluate and visualize the performance of the TPIMS through multidimensional aggregation of parking flow data. It can be observed from the dashboard that after the implementation of TPIMS, the utilization among parking facilities along I80 is more evenly distributed. Such intuitive tools can help agencies evaluate the system performance and make informed decisions. Despite the availability of real-time utilization information provided by truck parking information and management systems, there remains a significant requirement for truck drivers to know the anticipated availability of parking spaces at their intended arrival time. The second part of this dissertation develops predictive models that aid truck drivers in making informed decisions regarding trip planning, specifically focusing on two scenarios: pre-trip planning and en-route decision-making. Several tree-based ensembled machine learning models and a recurrent neural network based architecture is developed to forecast the utilization of parking sites and the results show that the prediction can be achieved with small errors. Models with and without other input attributes (e.g., weather, truck volume data) are discussed and compared at different prediction horizon, which enables various levels of forecasting. The integration of long-term and short-term prediction models aligns more closely with drivers' travel planning tendencies and this will help build a more efficient planning and decision-making process, thus optimizing the transportation industry's overall performance. Finally, to address the issue that the reliability and accuracy will hinder the future adoption of the system, a novel approach to enhance TPIMS is discussed by utilizing image processing techniques. In particular, the state-of-the-art You Only Look Once (YOLO) v5 algorithm is used to detect the occupancy of truck parking sites. By comparing the image processing results with the sensor reported data, sensor fault can be detected. Accordingly, a real time reporting system is proposed to communicate sensor fault result to the system operators in a timely fashion. The experimental results demonstrate the effectiveness and efficiency of the proposed system, and it also provides insights for the development and improvement of truck parking information and management systems.
Comments
Description
Keywords
Citation
DOI
Source
Subject Categories
Copyright