Evaluation of ramp control algorithms using microscopic traffic simulation
Introduction
Application of advanced information and communication technologies has made possible the implementation of innovative freeway control techniques. One of the most effective freeway control measures is ramp metering. Ramp meters regulate the entering traffic to the freeway to avoid traffic breakdowns and ensure smooth flow. Ramp meters also help break the “platoon” of entering vehicles, giving rise to efficient merging (Elefteriadou, 1997). They have been found to improve freeway capacity utilization, reduce extent and duration of recurrent congestion, reduce the occurrence of non-recurrent congestion, reduce average travel times, and increase throughput (Papageorgiou et al., 1997). This paper presents a detailed simulation evaluation of two algorithms. The two algorithms are: a local ramp metering algorithm called ALINEA (Papageorgiou et al., 1991) and an area wide coordinated algorithm called FLOW (Jacobson et al., 1989).
Several attempts have been made in last four decades toward the development of efficient ramp control strategies. Various methods have been proposed to calculate the metering rate that determines the number of vehicles allowed to enter the freeway from on-ramps. These strategies can be broadly divided into two categories: local metering and area wide metering. The metering plan in local control is based on locally measured traffic conditions; whereas, in area wide control, there is a coordination in metering rate calculations among ramp controllers. The parameters for a set of controllers are estimated jointly to achieve a system-level objective. A similar ramp control strategy that combines both local and area wide control is known as hierarchical control. In this approach, there is a system-wide optimization model at the upper level that calculates the desired network states, and a local controller at the lower level that adjusts the metering rate to minimize the difference between actual and desired network states (Chen et al., 1997).
The performance of ramp metering depends on various factors such as traffic volume, downstream traffic condition, and policy of handling queue spillbacks. These variables have complex interactions with ramp metering. To the best of our knowledge, there has been no study that has identified the effect of downstream bottleneck or queue spillback policy on metering performance till date. This is the first study to systematically investigate the sensitivity of ramp metering with respect to all these variables. In order to identify these effects one has to perform either field tests or simulation experiments. Considering that it may not be possible to control the above mentioned variables (for example downstream condition) in the field, simulation provides an ideal alternative approach to evaluate the performance of ramp control algorithms over a range of values for those variables.
Its uniqueness arises form its detailed and elaborate experimental design of ramp metering evaluation with respect to several key variables which have not been tested before. Furthermore, the majority of the simulation based evaluation studies have been performed with macroscopic traffic simulators. Field data has strongly demonstrated a complex nature of traffic pattern in and around merging areas (Cassidy and Bertini, 1999; Hall and Hall, 1990). Traffic flow in merging areas is emergent from a complex interaction between mainline and ramp traffic and depends on several factors including directional demand, driver behavior, road geometry and such. Macroscopic simulators are based on a coarse representation of traffic flow that fails to represent the said interactions among vehicles. Thus, macroscopic simulators may not be adequate to evaluate the performance of ramp control algorithms. We use the microscopic simulation laboratory, called MITSIM, for evaluating ramp control algorithms. Our motivation to use MITSIM laboratory lies in an explicit and accurate modeling of merging behavior, which is critical for evaluation ramp metering strategies.
MITSIM laboratory consists of a microscopic simulator that is responsible for moving traffic, and a traffic management simulator that is responsible for simulating control operations. It is designed for the evaluation of dynamic traffic management systems. A detailed description of the MITSIM laboratory is beyond the scope of this paper and can be found in Ben-Akiva et al., 1997). In the simulator, vehicles are moved based on car following and lane changing behavior. Behavioral models in MITSIM have been calibrated and validated with a large amount of data from various sites (Ahmed, 1998). MITSIM laboratory is used to evaluate how level of traffic demand, queue handling policy and downstream bottleneck affect the performance of ramp metering and derive insights from the evaluation results.
A realistic modeling of lane changing behavior is critical for modeling ramp-freeway merge area. Lane changing models in MITSIM are based on a rigorous econometric approach and calibrated and validated against the real field data. In this section, we present a brief overview of lane changing models. Interested readers are encouraged to consult Ahmed (1998) for a detailed description of this part of MITSIMLab. There are three kinds of lane changing operations in MITSIMLab, namely discretionary, mandatory and forced lane changing. A discretionary lane changing is desired when a driver wants to improve his driving condition by changing lane (for example, changing lane to gain speed) whereas, a mandatory lane changing is in effect when driver has to change lane in order to maintain his path to the destination (for example, changing lane from a closed lane or changing lane to take an exit).
The execution of discretionary and mandatory lane changing is based on critical gap acceptance behavior i.e. drivers seek a minimum acceptable gap called critical gap. The critical gap is estimated using a random utility model. The traditional gap acceptance models fail to capture drivers' lane changing behavior in a very congested network where the acceptable gaps do not exit. In such situations, drivers force other vehicles to create an acceptable gap. Merging operations in a heavily congested freeway is one of the examples of such behavior. The probabilistic model of forced lane changing determines the location where an individual vehicle performs the forced lane changing. The explanatory variables for modeling forced lane changing include the time during which the driver has been waiting to change lane, the traffic conditions in the neighborhood, distance from the point where the driver has to complete lane change (for example gore area in case of merging from the on-ramp). Ahmed (1998) presents the specifications, calibrations and validations of the discretionary, mandatory and forced lane changing models.
In addition to using disaggregate traffic data for calibration of various models in MITSIM, a non-linear optimization approach is also used to calibrate the parameters. Kurian (2000) used Boss/Quattro to study the sensitivity of MITSIM with respect to various parameters and select a small set of parameters. He used loop detector data that are readily available for calibrations. Thus, he iteratively ran MITSIM and used Boss/Quattro to update simulation parameters so that the difference between observed traffic variables and simulated traffic variables is minimized. To the best of our knowledge, this is the first study that uses loop detector data to systematically calibrate the parameters of a traffic simulation model using an optimization framework.
Despite theoretical advances in the development of ramp metering, their implementations have been slow. Most existing ramp meters in the field today use either fixed control or demand–capacity control (Papageorgiou et al., 1991). There have been a number of evaluation studies of various ramp control methods. The two methods used in the evaluation of ramp control systems are field operational tests and computer simulations. Different simulation studies have been conducted to test various ramp control strategies, most of which are performed under hypothetical networks and traffic demand. Only a few field studies have been undertaken. In the following sections, a brief description of some of the ramp metering evaluations is presented. It should be mentioned here that for all the evaluation studies described below and elsewhere in this paper, travel time savings are with respect to the no control scenario unless otherwise specified.
Proper (1997) studied the effectiveness of ramp metering for North American traffic management centers and found speed increases in the range of 16–62% and travel times savings up to 48%. According to MnDOT (Minnesota Department of Transportation), ramp metering application in the Minneapolis–St. Paul (M-SP) metro area has resulted in 30% increase in throughput (ITS International, 1997). Speeds on the main highway in peak hours have increased from an average 48–77 km/h. Lanes on metered freeways typically carry 2200–2400 vehicles per hour per lane and sometimes as high as 2700. The INFORM (Information for Motorist) evaluation in Long Island, New York used ramp metering, traffic signal control, and route diversion and was found to increase speeds by 13% and VMT (Vehicle miles traveled) by 5% (Smith and Cesar (1992)).
A number of local ramp metering strategies were applied in the field at a single ramp of Boulevard Peripherique in Paris (Papageorgiou et al., 1997). It was found that ALINEA led to a maximum improvement in all measures of effectiveness (MOEs) compared to the other strategies namely demand–capacity and fixed timing. Similar results were found for A10W motorway in Amsterdam. ALINEA was also tested at multiple on-ramps of the clockwise Boulevard Peripherique. The coordinated feedback control algorithm, METALINE was implemented at the same site for comparison (Papageorgiou et al., 1997). The field tests were performed under normal traffic condition. ALINEA improved total travel time, total travel distance and mean speed by 5.2%, 1.4% and 6.8% respectively, whereas the corresponding improvements for METALINE were 4.8%, 0% and 4.8%. Field trials were also conducted with ALINEA on an urban network that included a freeway, a parallel arterial, and connecting radial streets (Papageorgiou et al., 1997). The impact of ramp metering on corridor traffic was studied by comparative evaluation of several performance indices in cases with and without control. The main finding of the field trial was that application of an efficient ramp metering strategy considerably improved traffic conditions not only on the freeway but also on the parallel arterial and radial streets.
An evaluation study was undertaken to determine the effectiveness of a heuristic area wide ramp metering strategy (FLOW) in the Seattle metropolitan area (Jacobson et al., 1989). The system distributed demand among ramps in the network and discouraged short trips. In addition, the meters encouraged the use of underutilized ramps and arterials. The average delay at metered ramps was less than 2 min per vehicle during the morning and afternoon peak periods. The same ramps produced 5–8 min delays when measured in pre-metering period. After metering was implemented, travel times showed considerable improvement. Although there might be other factors contributing to the accident rate reduction, it appeared that metering was a significant cause of the reduced accident rates.
Although field operational tests are ideal for the evaluation of any traffic control system, they tend to be prohibitively expensive, time consuming, and sometimes infeasible. In addition, the test results depend on uncontrollable elements (e.g. weather conditions, travel demand, incidents) and an accurate analysis of the impacts of ramp control is often not possible due to confounding effects. In recent years, simulation has emerged as an alternate tool to evaluate the performance of traffic controls and to select an appropriate design. Simulation studies can also be used to analyze the robustness of a design by evaluating a range of scenarios, and to calibrate control parameters. A number of studies have simulated ramp metering for different transportation networks.
Hellinga and Van-Aerde, 1995, used INTEGRATION, a macroscopic traffic simulator, to evaluate a time-of-day ramp control for a test network, and found a slight reduction (0.39%) in total network travel time. Based on a sensitivity analysis, they discovered that the traffic conditions were influenced by the timing of ramp metering implementations, suggesting benefit from metering strategies that use real-time traffic data. The CORSIM (CORridor SIMulation) microscopic simulator was used for the evaluation of time of day, fixed time metering in the Atlanta metropolitan area (Matson and Daniel, 1998). Before and after travel times for the I-75 northbound corridor indicated a 16.5% decrease in total travel time and a 19.7% increase in average speed for the freeway sections. Papageorgiou (1980) used a dynamic traffic model to simulate time-of-day ramp control. The model takes into account the time delay of a volume change at a ramp and its impact at downstream locations. For a hypothetical freeway traffic situation, travel time improvements of 24% and 14% with respect to the case of earlier time-of-day control procedures were reported. A dynamic traffic model was used by Papageorgiou, 1983a, Papageorgiou, 1983b to study the efficiency of a hierarchical ramp control system. He simulated the no control and the hierarchical control case on a hypothetical freeway stretch and showed the improvements by the hierarchical control.
The local feedback control algorithm (ALINEA) and the coordinated feedback control algorithm (METALINE) were tested in simulation studies by Papageorgiou et al., 1990, Papageorgiou et al., 1991. They simulated these algorithms for the Boulevard Peripherique in Paris using METANET macroscopic traffic simulator. Both feedback control strategies were found to decrease the total travel time (they led to roughly the same results under normal conditions) with METALINE resulting in slightly better performance for non-recurrent congestion. During non-recurrent congestion, bottlenecks may form at unexpected locations which may be better identified and incorporated by a coordinated algorithm. The statistical traffic model simulation (Whittaker et al., 1997) was used to test the NMSS feedback control (a multivariable feedback control law) algorithm for the A10 West Motorway in Amsterdam (Young et al., 1994). Simulation results showed that the multivariable control algorithm was able to prevent the congestion that was otherwise present in an identical no control scenario. A simulation test of optimal ramp metering control with the TRAF simulation software in the I-94 freeway corridor in St. Paul, Minneapolis showed travel time reductions (Stephanedes and Chang, 1993). Ritchie et al. (1996) used INTRAS (INtegrated TRAffic Simulation) microscopic traffic simulator to empirically validate theoretical results of an area wide optimal ramp control strategy for a stretch of freeway in Pasadena, California. Five different predetermined metering rates were used for the simulation. This non-traffic responsive, fixed time control strategy had little impact on the mainline, but potentially negative impacts on ramps and surface streets. Another simulation study of an integrated control system was performed by Gardes et al. (1993), investigating ATIS and ATMS control for the Smart Corridor in Los Angeles. Three types of ATMS/ATIS controls were used: ramp metering, traffic signal control, and route diversion. A no control case and five combinations of controls were simulated for the base condition using the INTEGRATION macroscopic model. The results showed marginal travel time improvements.
Chen et al. (1997), used MITSIM microscopic traffic simulator to test three control algorithms––local control, area control and bilevel control that combined both local and area controls. The network used was the Central Artery/Tunnel (CA/T) network in Boston. The study showed that the bilevel control outperformed other control strategies. The improvements in total throughput for the local, area wide, and bilevel controls were 4.9%, 5.1% and 8.4% respectively. The travel time savings for the three control strategies were 9.4%, 8.8% and 12.6% respectively.
Section snippets
Algorithms evaluated
ALINEA and FLOW were selected on the basis of their traffic responsiveness, demonstration of their previous applications and simplicity. The algorithms were tested in the field successfully with encouraging results. The results of the field studies of these two algorithms were discussed earlier. ALINEA is a representative of local control algorithms and FLOW is a representative of the heuristic coordinated algorithms. Both algorithms use real time traffic surveillance data as input that is easy
Application to the central artery/tunnel project
The network used in this study is a part of the CA/T project in Boston. The CA/T project is the largest highway project in the United States. It is a 7.5 mile interstate highway, approximately half of which will be built as a tunnel.
I-93 North of the CA/T network was selected for this evaluation study. The network is expected to carry a high traffic demand beginning from the year 2004 when it is expected to be operational. Projected evening peak hourly volume level for a small portion of the
Calibration of input parameters
First, a brief description of the input parameters for ALINEA and FLOW that were used for calibration will be presented in this section followed by the calibration results.
Experimental factors
Three variables were used in designing experiments for the evaluation of ALINEA and FLOW for this study. They were––OD demand, downstream traffic condition, and queue override strategy. These variables are discussed in details in the following subsections.
The projected PM peak OD demand provided by the traffic planners for the year 2004 was used as the base demand for this research. Five levels of OD demands––80%, 90%, 100%, 110%, and 120% of the base demand (2004 PM peak) were used for the
Regression results
While the tabular analysis of the previous sections implicitly demonstrated the effect of experimental factors on the performance of the algorithms, a regression analysis of the results explicitly quantifies the impact of different variables. Also, it is possible to get the interaction effect of the experimental variables using regression. Next we present and analyze results from a linear regression analysis performed on data presented in the last section.
Percent travel time savings compared to
Summary and findings
This paper presented the calibration and evaluation of two ramp control algorithms ALINEA and FLOW. MITSIM microscopic traffic simulation laboratory was used to perform the empirical study. The network used for the study was a part of the CA/T network in Boston. A large number of experiments were performed to calibrate the two algorithms and an extensive experimental design was adopted for evaluation and comparison of the algorithms with respect to key variables. To the best of our knowledge,
Acknowledgements
We would like to acknowledge Professor Markos Papageorgiou for his invaluable comments and suggestions at various stages of this research.
References (26)
- et al.
Some traffic features at freeway bottlenecks
Transportation Research B
(1999) A hierarchical control system for freeway traffic
Transportation Research B
(1983)A new approach of time-of-day control based on a dynamic freeway traffic model
Transportation Research B
(1980)- et al.
Modeling and real-time control of traffic flow on the Southern Part of Boulevard Peripherique in Paris: Part II: Coordinated on-ramp metering
Transportation Research A
(1990) - et al.
Tracking and predicting a network traffic process
International Journal of Forecasting
(1997) - Ahmed, K.I., 1998. Modeling drivers' acceleration and lane changing behavior. Ph.D. Thesis, Massachusetts Institute of...
- et al.
Simulation laboratory for evaluating dynamic traffic management systems
ASCE Journal of Transportation Engineering
(1997) - et al.
Development and evaluation of a dynamic traffic control model for real-time freeway operations
Freeway merging operations: A probabilistic approach
- et al.
Simulation of IVHS strategies on the smart corridor
Capacity and speed flow analysis of the QEW in Ontario
Transportation Research Record
Examining the potential of using ramp metering as a component of an ATMS
Transportation Research Record
Cited by (55)
Fuzzy Power Heronian function based CoCoSo method for the advantage prioritization of autonomous vehicles in real-time traffic management
2021, Sustainable Cities and SocietyCitation Excerpt :Also, local ramp metering algorithms only consider the traffic conditions, improving safety, and preventing congestion near the ramp. However, it provides significant improvements in the entire network (Hasan, Jha, & Ben-Akiva, 2002; Scariza, 2003). Ramp metering creates an equity issue by focusing on the mainline traffic rather than on-ramps.
Statistical evaluation of data requirement for ramp metering performance assessment
2020, Transportation Research Part A: Policy and PracticeCitation Excerpt :Such technologies are, variable speed limits that change the speed limits based on real-time traffic conditions (Karimpour et al., 2017; Nezafat et al., 2018; Lee et al., 2004), dynamic lane merging systems and dedicated lanes that improve safety and reduce delay associated with non-recurrent congestions (Datta et al., 2004; Nickkar and Lee, 2019), dynamic message signs that provide advisory messages to the commuters regarding upcoming incidents, congestions and special events (Hassan et al., 2012; Garber and Patel, 1994), ramp meters that regulate the in-flow traffic to freeways (Papageorgiou and Kotsialos, 2002; Lee et al., 2006; Asgharzadeh et al., 2019), and hard shoulder running systems that regulate traffic conditions and improve operation on the freeways in case of non-recurrent congestion (Guerrieri and Mauro, 2016, 2006) are all various ITS technologies that have been widely used to improve the roadways mobility and safety. Ramp metering is known to be an effective freeway control measure that ensures the overall efficiency of the highway system by regulating the inflow traffic on-ramps (Hasan et al., 2002; Sun et al., 2013). Ramp metering has been implemented in the United States since the 1960 s (Toorawa and Ireij, 2005).
Computer-aided analysis and evaluation on ramp spacing along urban expressways
2013, Transportation Research Part C: Emerging TechnologiesCitation Excerpt :For a real urban expressway system like one in Beijing, a large number of ramps with complex configurations along urban expressways in traffic infrastructure networks cause much analysis and evaluation workload for researchers. First, the applicability and flexibility of the existing analytical methods to external conditions, such as traffic demand fluctuations and traffic emergencies are usually not able to come up with the requirement of evaluation (Hasan et al., 2002). A number of simulation software packages have been widely used in traffic analysis and evaluation during the past twenty years although mathematical models, measurements and calculations on the spot are not substitutive in reflecting traffic microscopic characteristics and suppositional scenarios, such as the configurations of urban expressways and their frontage roads, design speeds, traffic volume and drivers’ characteristics (Chen et al., 2010).
Customizing the following behavior models to mimic the weak lane based mixed traffic conditions
2022, Transportmetrica B