Estimate the impact of work zones using deep neural networks-ScienceDaily

Roadside construction (detours, closed lanes, or slow weaving past workers and equipment) affects traffic flow and travel time at the overall system level. The ability to accurately predict and plan what those effects will be will be of great help to both transportation and road users. The latest Small Starts project, led by the University of Utah’s Abbas Rashidi, is funded by the National Institute for Transportation and Communitys and introduces a robust and deep neural network model for analyzing the impact of construction zones on vehicle traffic.

The top three causes of non-regular traffic delays are crashes, work zones and bad weather, with work zones accounting for 10% of all non-regular delays. Accurate work area impact prediction can significantly reduce fuel consumption and air pollution.

“Machine learning and deep learning are powerful tools for building different types of data and predicting future situations. Using AI to analyze data is generally the future of transportation engineering,” Rashidi said. Mr. says.

The Utah Department of Transportation (UDOT) collects various types of data related to the operation of work areas. Using these data, Rashidi and graduate research assistant Ali Hassandokht Mashhadi sought a way to assess the impact of various variables on traffic and mobility across the road system. This analysis allows UDOT to better understand and plan more efficient work zone operations, select the most effective traffic management system for the work zone, and assess the hidden costs of construction work in the work zone. Useful for.

What are the factors that affect car traffic?

The impact on work zone traffic depends on other existing conditions and how they intersect with work zone factors.

  • Work area factors: Work area layout and location, road closure length, traffic speed in the work area, and daily activity time.
  • Traffic factors: Percentage of heavy vehicles, highway speed limits, capacity, mobility, flow, density, congestion, occupancy.
  • Road factors: Total number of lanes, number of open lanes, pavement grade and condition.
  • Time factor: Year, season, month, day of the week, time, and darkness / light.
  • Spatial factors: Lane width and presence and number of nearby highway ramps.

UDOT collects a large amount of raw data about work zones, including data about the above factors that made this project possible.

The Deep Neural Network (DNN) model developed by researchers can evaluate the effects of multiple factors and the interactions between them. DNN can capture the relationship between input variables and outputs, as opposed to traditional machine learning algorithms.

How does the model work?

DNN was trained and evaluated on approximately 400,000 data points collected from approximately 80 projects on Utah roads. Researchers evaluated the performance of the model using three different measurements, including R2 score, root mean square error (RMSE), and mean absolute error (MAE). The accuracy of the results for all types of work areas, including short-term and long-term, day and night, interstate and arterial work areas, was acceptable and the expected traffic error was less than 2%. This is the first study to investigate the impact of work zone functionality on hourly traffic.

The main advantage of the proposed model is that the user does not have to set various adjustment factors based on actual experience. Previously developed models usually require some adjustment factors in the mathematical model to estimate the capacity of the work zone. However, the model developed by Rashidi and Mashhadi allows you to estimate the amount of traffic per hour without having to manually add factors. It is also worth noting that by using work area features, road features, and temporal features as input variables, the model can estimate work area traffic even in areas without traffic sensors.

Method implementation

This pilot project, funded by the NITC Small Starts grant, has shown promising results.To get into the hands of experts who can use this model, the research team has already published one treatise. Transport research record: This is a review of how to estimate the capacity of a construction work area, and another journal article is being published. The next step involves sharing the findings with UDOT for feedback and seeing how the model can help UDOT. Rashidi also wants to extend the functionality of the model in future research.

“This study focused on Utah data, so it would be great if we could do a similar study and compare the results with other states. See how the behavioral patterns are similar. Please, “said Rashidi.

Story source:

material Provided by Portland State University.. Note: The content can be edited in style and length. Estimate the impact of work zones using deep neural networks-ScienceDaily

Back to top button