Chunho Yeom, Ph.D., Ali Hajbabaie, Ph.D., Nagui M. Rouphail, Ph.D., William Rasdorf, Ph.D., P.E., and Bastian J. Schroeder, Ph.D., P.E. 2017-10-26 03:44:17
Freeways with uninterrupted flow are expected to provide seamless movement of traffic resulting in reliable and safe travel conditions with a high level of service. In a time of aging transportation infrastructure, repair and rehabilitation work on freeways is becoming increasingly frequent, causing congestion on the affected facilities. While some construction activities can be limited to night-time or off-peak hours, others require lane closures and intense work activity even during peak periods. Transportation agencies need tools and predictive models to estimate the impacts of various work zone configurations on freeway traffic operations to better evaluate and weigh the impacts of different construction staging schemes. Although it is not uncommon to encounter studies about freeway work zone capacity and speed-flow relationships, predicting free-flow speed (FFS) has not been as commonly studied. Research Objectives The objective of this paper is to present a predictive freeway work zone free-flow speed model based on key variables that influence traffic operations: speed (under non-work zone and work zone conditions), lane closure severity index, barrier type, day or night activity, and total number of ramps in the vicinity. Literature Review Benekohal et al. developed a work intensity speed reduction model using data from a survey of freeway drivers in rest areas.1 They reported a 7.25 mph free-flow speed reduction in short-term work zones (i.e., with temporary barriers such as cones) and a 3.27 mph free-flow speed reduction in long term work zones (i.e., with concrete barriers). Chitturi and Benekohal compared speed reduction on freeways due to decreased lane width and lateral distance at eleven rural freeway work zones (all 2 to 1 lane configurations) in Illinois.2 The authors determined that the narrower lane width and reduced lateral distance had a high impact on free-flow speed. However, they did not provide any specific model explaining freeway work zone free-flow speed, nor did they indicate the extent of the impact. Porter and Mason observed the 85th percentile speed of passenger vehicles and trucks in freeway work zones.3 They collected data from 17 work zones in Pennsylvania and Texas. After extensive statistical analysis they determined relationships between speed and several variables, including work zone type (lane closure or median crossover), downstream distance, posted speed limit, vertical alignment, width of pavement, and barrier type concluding that work zone posted speed limit and work zone type had the greatest effect on free-flow speed. Hajbabaie et al. studied the effects of various speed reduction treatments on both free-flow and average speed at freeway work zones.4 They collected data from three sites in Illinois with two lanes open upstream of and in the work zone, a 65 mph speed limit upstream of the work zone dropping to 55 mph in the work zone, and work activity on the left hand side of the travel lanes. They found that police presence reduced the free-flow speed of passenger cars and heavy vehicles by averages of 7.3 mph and 3.2 mph, respectively. Methodology The freeway work zone free-flow speed model was developed using regression with speed and flow data from three sensor databases, from a project website, from satellite images, and by directly contacting road agencies to obtain the data. Using the data collected, a statistical analysis was performed to find key variables affecting freeway work zone free-flow speed. It is noteworthy to mention that the research team considered both additive and multiplicative free-flow speed models. Additive models are those which assume that the different components affect the time series additively. On the other hand, multiplicative models focus on the percentage of change between the values of the variables. The additive approach was ultimately selected since it was more intuitive in assessing the impact of variables on freeway work zone free-flow speed and it produced results almost identical to the multiplicative approach. This study aimed at developing predictive models for free flow speed on freeway work zones under different geometric and prevailing conditions. As a result, free flow speed data under different geometric and prevailing conditions was needed and collected. The data was analyzed for missing values and unreasonable observations before fitting the predictive models. In addition, independent variables were tested for potential correlations using either a correlation coefficient test or a Pearson correlation matrix.5 Different independent variables were analyzed independently to identify the potential mathematical form that is the most suitable to represent the relationship between independent and dependent variables. A ranking and sequential variable selection approach was used to select independent variables that are more suitable to be included in the final model. Finally, the final model was validated in three sites (2 in Richmond and 1 in Los Angeles). In the remainder of this section details will be provided on: Data collection; Variable correlation test; Variable selection on model fitting process; and Model validation. Data Collection The free-flow speed model development was based on data collected from: the Regional Integrated Transportation Information System (RITIS); Caltrans Performance Measurement System (PeMS); and the NCDOT Traffic.com database.6,7,8 Supplemental field data collected in Richmond, VA (I-95) and Los Angeles, CA (I-5) were used for validation purposes.9 Collecting freeway speed and flow data from a sensor database was very efficient, provided that each sensor was located within the work zone and remained well calibrated. Once a work zone is identified and its configuration is known, sensors can be strategically located to provide quality data. Figure 1 shows an example of sensor data collected on I-895 in Maryland. The figure plots flow volume vs time (two days) for a left (brown) and right (purple) lane. Additional data were gathered using satellite and street view images, direct inquiries to work zone site agencies, and the work zone project website, a small amount of which is presented in Table 1. Note that here we are showing a few lines of data to illustrate the overall scope of the variables but in no way was this the entire data set. Information was collected from 14 freeway sites providing a total of 24,633 rows of speed and flow rate data (either 5-or 15-minute interval speed and flow rate data). Free-flow speed was obtained based on the following five steps: Obtain 15-minute speed and volume data from sensors' database; Select 15-minute traffic data with flow rates below 500 vphpl; Calculate the average and standard deviation (SD) of the step 2 speed data; Exclude traffic data falling outside of the ±2 SD threshold; and Calculate free-flow speed by averaging the remaining speed observations from step 4. Specification of Independent Variables The following independent variables were used to test the impact of different work zone configurations on free-flow speed: ● Speed Ratio (Sr): The ratio of non-work zone speed limit to work zone speed limit; ● Posted Speed Limit (S in mph) ● Lane Closure Severity Index (LCSI): The inverse of the open lane ratio (total/open) multiplied by the inverse of the number of open lanes; ● Area Type (A): 0 for urban, 1 for rural; ● Barrier Type (Br): 0 for hard, 1 for soft; ● Daytime or Nighttime (DN): 0 for day, 1 for night; ● Lane Closure location (LC): 0 for right hand side lane closure, 1 for left hand side lane closure, ● Work zone Length (l in mile); and ● Number of ramps (freeway interchange connections) within 3 miles upstream or downstream of the work zone (Nr). Note that the speed limit in the work zone may be similar to that upstream of the work zone or up to 10 mph lower. Therefore, the speed ratio is expected to account for the differences in free-flow speed for various speed limits under non-work zone and work zone conditions. Finally, also note that the variables shown were consistently available across our data. It is worth noting that the other variables were not considered in our model because we did not have data sets that adequately included them. Weather and truck volume are examples of such variables. Clearly, weather affects free-flow speed and capacity. Truck volume may as well. In future studies to enhance this work such variables should be further considered. Free-Flow Speed Model Correlation test results indicated that area type (A), lane closure side (LC), and day or night condition (DN) were not independent. As a result, they were not included in the variable selection process simultaneously. A forward variable selection approach was used to determine independent variables to appear in the free flow speed prediction model. The following eight different variable combinations were tested. ● Combination 1: Sr, S, LCSI, Br, Nr ● Combination 2: Sr, S, LCSI, Br, Nr, A ● Combination 3: Sr, S, LCSI, Br, Nr, A, LC ● Combination 4: Sr, S, LCSI, Br, Nr, A, DN ● Combination 5: Sr, S, LCSI, Br, Nr, A, LC, DN ● Combination 6: Sr, S, LCSI, Br, Nr, LC ● Combination 7: Sr, S, LCSI, Br, Nr, LC, DN ● Combination 8: Sr, S, LCSI, Br, Nr, DN Combinations 1, 2, 3, 7, and 8 have p-values less than 0.05 for all their coefficients. Accordingly, these combinations were retained for further analysis. Combinations 4, 5, and 6 had at least one variable with a p-value exceeding 0.05 and were thus eliminated. Combinations 2 and 3 were also dropped from further consideration as they used area type, which was highly correlated with work zone speed limit, a key variable in free-flow speed model development. After examining the remaining combinations (1, 7, and 8) the research team selected combination 8 as the proposed model as presented in Table 2 and by the following equation. Freeway WZ FFS (mph) = 9.95 + 33.49×fSr + 0.53×fS – 5.60×fLCSI – 3.84×fBr – 1.71×fDN – 1.45×fNr To illustrate, consider a 2-to-1 lane closure configuration with posted non-work zone and work zone speed limits of 65 mph and 55 mph, respectively. The work zone uses a cone (soft) barrier during the day time and has a total of 6-nearby ramps. The free-flow speed for this scenario would be: Freeway WZ FFS = 9.95+33.49×(65/55)+0.53×55–5.60×2– 3.84×1–1.71×0–1.45×6 = 54.8 (mph). Model Validation The freeway work zone free-flow speed model presented above was independently validated with field data collected at three additional work zones (2 in Richmond, VA and 1 in Los Angeles, CA). These sites showed very different operational conditions, with two having very low free-flow speeds (two Richmond locations) and the other having moderate to high free-flow speeds (LA location). The validation results are shown in Table 3. The free-flow speed at each site was collected using radar speed detectors. Both Richmond sites had a 3-to-1 lane closure while the Los Angeles site had a 3-to-2 lane closure. For these three sites, the model predicted successfully free-flow speed. An evaluation of the 95th percentile confidence interval around the field-observed mean showed that the model prediction fell within that confidence bound in the Los Angeles site showing a 1.1 percent difference, while the Richmond sites showed 8.6 percent and 14.6 percent differences, respectively. A very large data set from multiple sources was used to develop this freeway work zone free-flow speed model. The sensors' database was the only available way to obtain the necessary volume of data in a practical way. These data were used to develop the model. To validate the model we used a smaller data set made up of data collected from three field sites. In doing so we were able to set up an approach to apply the free-flow speed model to actual field data. The model can now be calibrated and improved. Results This section presents both the freeway work zone free-flow speed regression model results and the model validation results. As noted earlier, free-flow speed for non-work zone and work zone freeways was calculated using field data from the sensor database similar to that shown in Figure 2. The first two columns of Figure 2 refer to the data collection location and route identification. The minimum, 1st quartile, median, 3rd quartile, and maximum free-flow speed values with outliers (white dot inside circle) are shown in the form of a box plot, separated by an indicator of work zone presence. Sample size, speed difference between work zone and non-work zone conditions, and p-values are presented for each site as well. The presence of a work zone significantly changed free-flow speed (p-values <0.0001) at all sites except for one in Maryland (R-03-MD). In addition, the R-04-MD site shows a counterintuitive result, as the free-flow speed with a work zone is higher than without one. It should be noted that the data for both R-03-MD and R-04-MD came from the same site, having a work zone with a lane shift (two lanes open) and no lane closure. All other sites had at least one lane closed and showed free-flow speed differences ranging from 0.1 mph to 8.9 mph. Findings and Conclusions This paper presents a freeway work zone free-flow speed model based on an abundance of multi-state sensor data and extensive data analysis. The model predicts free-flow speed using the speed limit ratio between non-work zone and work zone conditions, the posted work zone speed limit, lane closure severity index, barrier type, day or night condition, and the total number of ramps in the vicinity. The model was validated using field data that was not used in the model calibration process and we found that it successfully predicted both relatively high and low free-flow speed. As hoped, we were able to effectively quantify the degree to which the work zone presence reduced free-flow speed on freeways compared to normal conditions. Thus we present the model here in the hope that others may benefit from its use. The authors hope that transportation agencies and traffic engineers find the model to be useful, practical, and simple to use. We recommend its use as a starting point for free-flow speed prediction and we recommend that users calibrate the model for local conditions, thus further enhancing its future utility. We additionally recommend assessing the effect of nearby on-and off-ramp traffic. It may be the case that there is an effect on free-flow speed due to the presence of significant merging and diverging maneuvers in the vicinity of the work zone. We also recommend the inclusion of weather as a factor that will affect model results and a study of its impact on the model. Intuitively we know that serious weather patterns reduce both speed and capacity but quantifying the extent to which this occurs would be valuable. We also recommend serious consideration of including a similar study of truck volume and its impact on speed. While we have conducted a thorough study and created and validated a solid and useful model there are additional considerations to further explore. Finally, earlier in this paper we suggested that our speed model should be combined with one or more of the existing capacity models referred to in the introduction. What we have not done in this paper is describe how such a "comprehensive" speed-flow model would be developed. However, doing so is critical to taking this work to the next level, as is making use of the improved model we provided herein. References Benekohal, R. F., et al. "Methodology for Estimating Operating Speed and Capacity in Work Zones." Transportation Research Record: Journal of the Transportation Research Board, Vol. 1883, No. 1 (2004): 103-111. Chitturi, M. and Benekohal, R. "Effect of Lane Width on Speeds of Cars and Heavy Vehicles in Work Zones." Transportation Research Record: Journal of the Transportation Research Board, Vol. 1920, No. 1 (2005): 41-48. Porter, R. J. and Mason, J. M. "Modeling Speed Behavior of Passenger Cars and Trucks in Freeway Construction Work Zones: Implications on Work Zone Design and Traffic Control Decision Processes." Journal of Transportation Engineering, Vol. 134, No. 11 (2008): 450-458. Hajbabaie, A., et al. "Sustained and Halo Effects of Various Speed Reduction Treatments in Highway Work Zones." Transportation Research Record: Journal of the Transportation Research Board, Vol. 2265, No. 1 (2011): 118-128. Ott, R. and M. Longnecker. "An Introduction to Statistical Methods and Data Analysis." Cengage Learning, 2010. RITIS. http://www.cattlab.umd.edu/?portfolio=ritis, Accessed 11/11, 2015. PeMS. http://pems.dot.ca.gov, Accessed 11/11, 2015. Traffic.Com. https://www.here.com/traffic, Accessed 11/11, 2015. Yeom, C., et al. "Innovative Work Zone Capacity Models from Nationwide Field and Archival Sources." Transportation Research Record: Journal of the Transportation Research Board, Vol. 2485, No. 1 (2015): 51-60. Chunho Yeom, Ph.D. is a senior researcher for the Korea Expressway Corporation Research Institute. He works in the transportation research division researching for traffic safety and operation. He has around 20 years of experience in the construction and transportation sectors in not only Korea but the United States. He worked for AASHTO as the first international engineering management fellow in 2011 and for the Institute for Transportation Research and Education located in North Carolina State University. He holds a masters and doctorate in civil engineering from North Carolina State University. Ali Hajbabaie, Ph.D. is an assistant professor in the Civil and Environmental Engineering Department at Washington State University. He holds a doctorate of philosophy, a master of science, and a bachelor of science degree all in civil engineering and another master of science degree in industrial engineering. His research focuses on cooperative traffic control utilizing connected and autonomous vehicle and the Internet of Things technologies. He is the secretary of the Work Zone Traffic Control Committee and a member of the Traffic Signal Systems Committee of the Transportation Research Board. He also chairs the Asset Management Subcommittee of the Traffic Signal Systems Committee. Nagui M. Rouphail, Ph.D. is professor of Civil Engineering at North Carolina State University. Previously he served as director of the Institute for Transportation Research and Education (ITRE) for a period of 15 years. Nagui is an internationally known researcher in the field of traffic operations, transportation modeling and the intersection of operations and environmental impacts. He was a principal contributor to several chapters in the U.S. Highway Capacity Manual releases from 2000 until 2015. He also held leadership positions in Transportation Research Board, American Society of Civil Engineers, and ITE technical committees. William Rasdorf, Ph.D., P.E. is a professor of Civil, Construction, and Environmental Engineering at North Carolina State University. He received bachelor's and master of science degrees in architectural Engineering from Penn State University and master of science and doctorate degrees in civil engineering from Carnegie Mellon University. Dr. Rasdorf's research focuses on the use of database and information modeling in the broad arena of infrastructure asset management, primarily in transportation. Bastian J. Schroeder, Ph.D., P.E. is a principal engineer with Kittelson and Associates, Inc. in Wilmington, NC, USA. He specializes in multimodal transportation system planning and analysis, simulation analysis and modeling, data analytics and visualization, and guidebook development. Bastian co-chairs the national ITE Simulation and Capacity Model Users Group (SimCap), and chairs SimCap-NC. He is a member of the Transportation Research Board (TRB) TTSM Task Force, the TRB Committee on Highway Capacity and Quality of Service, and the TRB committee on roundabouts.
Published by ITE Journal. View All Articles.