Construction work zones are inevitable parts of daily operations at roadway systems. They have a significant impact on traffic conditions and mobility of roadway systems. The traffic impacts of work zones could significantly vary due to several interacting factors such as work zone factors (work zone location and layout, length of the closure, work zone speed, intensity, and daily active hours), traffic factors (percentage of heavy vehicles, highway speed limit, capacity, mobility, flow, density, congestion, occupancy etc.), Road factors (number of total lanes, number of open lanes, and pavement grade and condition), temporal factors (e.g., year, season, month, weekday, daytime, and darkness), weather conditions (rainy, sunny, and snowy), spatial factors (road lane width, proximity and number of ramps), etc.
Utah Department of Transportation (UDOT) is continuously collecting and storing the aforementioned data and is interested in evaluating the impacts of those factors on mobility and traffic conditions of roadway systems within the state of Utah. Such an analysis will help the UDOT personnel to better understand and plan for more efficient work zone operations, select the most effective traffic management systems for work zones, and also assess the hidden costs of construction operations at work zones.
To help UDOT to address this problem, we propose a robust deep Recurrent Neural Network (RNN) model, capable of evaluating the impacts of the above-mentioned factors on mobility conditions of Utah roadway systems. Compared to traditional machine learning algorithms, deep RNN models contain robust features enabling them to predict results more efficiently in complex environments such as this project. A comparison between results obtained from implementing a deep RNN model, as well as traditional machine learning techniques (e.g. ANN and decision trees) will reveal the capabilities of the proposed system in accurately predicting the traffic conditions based on various work zone factors.