Actuated Signal Timing Optimization for a No-Notice Evacuation with High Left-Turn Demands

Abstract

The determination of the appropriate traffic signal timing plans for no-notice evacuations in densely populated areas is a noteworthy challenge. The objective of this study was to evaluate alternatives that could optimize evacuee traffic flow in a no-notice evacuation of areas near an oil refinery. This simulation case study focused on a residential area in the City of Wood River, Illinois, and used Synchro 8.0 and VISSIM 7.0. This case study was different from existing evacuation literature because of the high left-turn demand from evacuating traffic. The study methods were unique because of the application of dynamic traffic assignment, a left-turn movement on the evacuation route, and the simulation of fully-actuated traffic signals. These scenarios evaluated the following: (1) existing traffic infrastructure; (2) flexible shelter choice; and (3) optimized traffic signal timing with flexible shelter choice. The results suggested that optimizing the signal timing and allowing drivers’ flexibility in choosing evacuation routes achieved the fastest evacuation. These findings indicated that a longer cycle length and an extended left-turn phase were important factors in reducing traffic delay in the network. Overall, these findings underscore the importance of operating intersections efficiently during no-notice evacuations.

1. Introduction

Residents nearby an industrial zone could be vulnerable during catastrophic events such as a chemical explosion, fire, or an accidental release of poisonous gas. To mitigate the rate of fatalities, severe personal injuries, and property damage, emergency preparedness plays a central role. Evacuation from at-risk areas is a principal strategy to mitigate potential loss of life. These kinds of incidents can occur at any time without prior notice. Therefore, residents near an industrial zone might have to leave the affected area just after these incidents. This type of evacuation is classified as no-notice.

In the United States, emergency response teams, including police departments, fire services, and traffic management, have given at least one warning every day for natural or man-made hazardous events [1]. Sometimes this warning reaches the evacuation stage. To determine how emergency traffic signal timing plans can assist with faster evacuations, previous research studies have evaluated several methods to improve the operations of signalized intersections during evacuations. These include lane-based evacuation routing [2], a dynamic flashing yellow signal timing plan [3], the shortest-path algorithm in staged evacuation [4], and trip distribution [5].

The objective of this research study was to identify the evacuation strategy that expedited the most efficient, no-notice evacuation procedure. This study focused on a no-notice evacuation of a residential area near the Philips 66 oil refinery in the City of Wood River, Illinois. To meet the study objectives, the authors sought to optimize traffic signal timing and allow motorists flexibility in their choice of shelter locations and routes.

The cities of Wood River and Roxanna are located in Madison County, Illinois, United States. These cities are home to a Philips 66 oil refinery, which is capable of refining 360,000 barrels of crude oil per day, 165,000 barrels of gasoline per day, and 90,000 barrels of diesel and jet fuel per day [6]. However, there is a populated residential area just beside this oil refinery, which is considered at-risk. This area includes residents in both the cities of Roxana and Wood Rive, Illinois. If there is a chemical explosion, fire, or other event at this oil refinery, these residents might be the most affected. As a result, more than six thousand lives could be threatened. If the wind blows to the west, fire or poisonous gas could spread into these areas. This research assumed a fictitious incident at the oil refinery, which created the need for a no-notice evacuation of these nearby residential areas.

2. Literature Review

This section presents knowledge about signal timing for no-notice evacuations, including human behavior during a no-notice evacuation, traffic signal operations using flash mode, manual traffic control strategies, and traffic simulation tools for evaluating evacuation signal operations.

2.1. Human Behavior during No-Notice Evacuations

Because of natural or man-made disasters, it is necessary to implement an effective mass evacuation to reduce fatalities as well as loss of property. According to transportation engineering, no-notice evacuation is characterized by a greater level of complication because it brings about evacuees’ instant disaster response. Traffic movements at the time of a no-notice evacuation mostly depend on human behavior. The authors have attempted to describe these behaviors in a series of three decisions. These behaviors include the following: (a) time to recognize (evacuees receive emergency warning); (b) time to cope (evacuees react to the emergency situation); and (c) time to egress (find the right path and move) [7]. Previous researchers introduced the three controls (incident control, behavioral control, and network control) to reduce incident congestion during a no-notice evacuation. Among them, the behavioral control group takes into account information dissemination to the motorist to avoid the affected area and the congested roadway [8]. The impacts of driving behavior on traffic safety at the time of no-notice evacuation are also dependent upon headway and minimum gap. Researchers observed that the number of conflicts can increase by 407 percent with a 20 percent reduction in headway time and by 60 percent with a 20 percent reduction in the minimum gap [9]. In other research, a diversion curve was used to describe when evacuees chose to leave their homes [10]. This information can help predict the traffic volume loading on the road network according to time. Other authors have studied the relationship between human behavior and emergency evacuation route planning, demonstrating that encouraging drivers to use a variety of routes improves network clearance time [11].

2.2. Traffic Signal Operations

Because traffic signals are responsible for efficient operations at critical parts of the transportation infrastructure, many have studied their operation during evacuations. These studies have included flashing operations and manual operations (law enforcement, fire, or traffic management). During no-notice evacuations, operating traffic signals in flash mode could be an effective method to evacuate vehicles. The options for operating a traffic signal in flash mode are as follows: (a) flash yellow on the main roadway approaches and flash red on the minor roadway approaches; (b) flash red in all directions of an intersection; and (c) dynamic flashing yellow. Flashing yellow on the main roadway approaches supports a continuous flow of evacuating vehicles along that route. By using this method, roadway capacity can be used more effectively because of smoother traffic flow. Dynamic flashing yellow is a combination of flashing yellow on the main roadway with some consideration of side-street demand. More details about this signal operation method can be found in Asamoah and He [12]. In 2010, in Monroe County, Florida, traffic signals were operated using dynamic flashing yellow for hurricane evacuation conditions [13]. Arcadis recommended dynamic flashing yellow in their study for the Browns Ferry Nuclear Power Plant [14].

Additionally, managing intersection traffic during emergencies can be executed manually by law enforcement, fire services, or traffic operations personnel, especially during no-notice evacuations. Previous research suggests that manual signal control can improve the regulation of congested/malfunctioning signalized intersections with high saturation rates, which inundate the road capacity [15].

Using shoulders as travel lanes has also been used for increasing roadway capacity, improving mobility in uniform traffic flow, and enhancing public safety by reducing accidents. For example, the Texas Department of Transportation implemented an “evaculane” strategy where vehicles can use the paved shoulder of the freeway as a travel lane during an evacuation. Researchers showed that if the “evaculane” strategy was implemented, the average network (considering using I-10 and US-290) evacuation travel time could be reduced by nearly 30% compared to without [16].

Cova and Johnson [2] investigated lane-based evacuation routing plans using a mixed-integer micro-simulation programming solver known as CPLEX 7.0 for a complex road network. The main objective of lane-based evacuation routing plans was to minimize vehicle crossing conflicts, merging conflicts, and lane changing along multi-lane arterials. Simulation results revealed that network clearance time can decrease significantly, up to 40%, by channelizing flows at intersections and removing crossing conflicts.

Staged evacuation, where evacuees are categorized according to their risk and geographic location, has also been found effective for evacuations [4]. Liu and Lai [5] used a cell-based network optimization model for staged evacuation planning, finding that congestion was lessened.

To evaluate emergency evacuation traffic, many research studies have been conducted using traffic simulation software version 7. VISSIM has been extensively used to study evacuation traffic, including topics such as dynamic traffic assignment [17], emergency vehicle routing [18], manual traffic signal control versus fully-actuated control [19], and traffic signals in flash mode [20]. In addition, several evacuation studies have used the simulation tool PARAMICS to evaluate freeway contraflow [21], test wildfire evacuation scenarios [22], and post-disaster emergency vehicle routing with fully-actuated traffic signals [23].

In summary, other researchers have evaluated the traffic queue, trip distribution, and dynamic flashing yellow-based signal timing. These studies focused on incidents such as wildfires or hurricanes where evacuation notice was short or advanced. In contrast, the study described herein focused on a no-notice evacuation in response to an incident at an oil refinery. The evacuation traffic flow would place high left-turn traffic demand at a signalized intersection. No previous studies were found that evaluated incidents related to this type of industry and with high left-turn demands.

3. Materials and Methods

To conduct this study, the authors created an evacuation model based on an evacuation scenario near the Philips 66 oil refinery in Roxana and Wood River, Illinois, approximately 15 miles north of St. Louis, Missouri. The authors considered several possible evacuation scenarios for the population nearby the refinery. To be conservative, the authors assumed an evacuation for a chemical explosion or fire when the wind was blowing west, requiring a no-notice emergency evacuation of the residents nearest to the refinery.

This emergency evacuation model used Synchro and VISSIM traffic simulation software version 7 to evaluate different traffic management strategies. In order to build the simulation model in Synchro, the authors reviewed several aerial photographs of the at-risk area from Google Maps. To enable an accurate spatial layout of the model, the authors used small-scale screen shots of the study network in VISSIM 7. Because the evacuation zones and evacuation routes were too large to capture clearly in a single screen shot, multiple clear screen shots were combined using AutoCAD software version 30. Furthermore, the authors imported the entire study area image into Synchro software version 9. At first, the authors constructed the required links, confirming the correct number of lanes, intersection geometry, intersection spacing, etc. To make an actuated traffic signal intersection, the authors placed a detector in each lane of each intersection. In addition, traffic signal timing information for the selected study intersections was collected from the Illinois Department of Transportation’s (IDOT) District 8 office. Based on the IDOT signal timing plans, most intersections allow permissive left turns. The intersections of IL 143/IL 111 and IL 143/IL 3 used protected/permissive left turns. The authors did not input any vehicular volume in the Synchro traffic model except for the testing.

The next step was developing the traffic simulation model in VISSIM 7. The authors decided to use this software because it is a comprehensive microscopic traffic simulation software version 7 that can simulate any type of traffic signal control and geometric alignment. The first step to building a network model in VISSIM was to create the road links and intersections based on an aerial image. The image measurement difference between the Google Map-based background image (assembled in AutoCAD) and the VISSIM-scaled background image was less than 1%. The authors employed traffic signals at specific intersections, following the physical world, and directed the vehicle routes of each lane using the vehicle routes option. In addition, to allow vehicles to enter and exit the model, gateways/parking lots were added to represent traffic generators and network boundaries, each with a specific zone number. Further details about this model are summarized in

Table 1. Summary of the characteristics of the simulation model.

VISSIM Model Characteristic	Numerical Value Applied
Number of links	78
Number of connectors	108
Number of signalized intersections	10
Standstill distance	1.64 m (5.38 ft) [24]
Headway factor	1
Car following mode	Wiedemann 74
Simulation speed	10.0 steps/simulated second [24]
Number of simulation runs	10

.

The model included IL 111, IL 143, N. Wood River Ave., and IL 3; all were arterial routes and important for evacuation purposes. The model also included 10 signalized intersections along these routes, as well as many connecting local streets. The authors imported the traffic signals from Synchro into the VISSIM. For the signal controller, the authors imported Ring Barrier Controller (RBC) files for the 10 signalized intersections from the Synchro model. Any upgraded Synchro optimizing signal time setting for each phase of any intersection was applied as an updated RBC file in VISSIM to evaluate different evacuation traffic management strategies.

The methodology followed for this research is described in the following six main steps: (1) determine the affected area and the number of affected people; (2) estimate traffic volume; (3) assign the origin-destination (O-D) matrix; (4) estimate the parking capacity; (5) design and import traffic flow data into the simulation network; and (6) simulate and analyze the scenarios. The authors note that this approach parallels a standard traffic impact study. This research assumed a fictitious incident at the oil refinery would create the need for a no-notice evacuation of the nearby residential area. The authors assumed a base year of 2016 and that the evacuation would take place when most residents were home during the evening. The authors chose 10 p.m. on a weekday as the worst-case scenario because a large proportion of residents would be home at that time, although local traffic would be low. United States Census (2015) data and a growth factor (0.08) were used to estimate the number of permanent residents in the assumed evacuation areas and average the number of vehicles owned per household.

After estimating the population, the authors predicted the number of vehicles that were likely to evacuate the affected area, as well as the background traffic that would normally be present at that time (10–11 p.m. on a weekday). The existing roadways will be overloaded by evacuee vehicles at the time of an emergency evacuation. It is challenging to evacuate a large number of residences safely within the limited roadway capacity and a certain amount of time. To model an evacuation realistically, it is necessary to calculate background traffic as well as the number of evacuees. The evacuation population was transformed into a projected number of vehicles using an auto occupancy factor of 3 to 4 persons per vehicle, as suggested by previous research [25]. In addition, the authors considered a 0.15 increase factor [26] to estimate the number of evacuee vehicles. This increased factor helped account for two vehicles departing from several households. There were no special-care facilities in the at-risk areas. The total number of evacuee vehicles for all evacuation zones was approximately 2267.

To estimate the background traffic data, the authors used the Getting Around Illinois website. From this website, the value of the average daily traffic (ADT) was collected for each arterial route using the most recent data available in 2016. In order to identify the likely volume during the evacuation time of day, the authors visited the evacuation site and manually collected traffic volumes and vehicle types at two key intersections: IL 111 and IL 143 and Wood River Ave. and IL 143. The authors counted the number of vehicles for one hour (15-min intervals) at each intersection. The authors used 15-min count intervals instead of longer count intervals for better analysis of the data and for general intersection study purposes. These traffic volumes were used to identify the proportion of ADT that was likely between 10 and 11 p.m. These proportions were applied to the ADTs to predict the background traffic throughout the modeled road network. The authors assumed that background traffic consisted of passenger cars and less than 2% buses and trucks. For all of the evacuation O-D matrix files, the vehicle composition included 3% trucks/buses to account for fuel trucks also evacuating from the incident at the refinery.

The next step was to assign traffic, background, and evacuation and create origin-destination (O-D) matrices. First, the background traffic was converted into an O-D matrix for a 15-min time interval, which was adequate to allow vehicles to distribute throughout the network. The general origin–destination matrix pattern in a no-notice evacuation situation is different compared to the normal daily traffic pattern. Based on the 15 zones, the authors calculated the O-D matrix for the evacuee traffic data. The authors next selected the shelter location and evacuation routes based on recommendations from the local city engineers, who have strong knowledge of shelter locations, parking facilities, and road facilities. To distribute evacuees towards shelters, the authors divided the evacuation traffic into four 15-min time periods. The authors assumed participation rates based on previous studies [27,28]. The authors utilized the first 40% participation in the first fifteen minutes of the evacuation, and in the subsequent three 15-min evacuation periods, the authors used 30%, 20%, and 10%, respectively. After these time intervals, background traffic volume continued again. To maintain consistency, the authors used the same background traffic volume both before and after the evacuation event. After calculating the background and evacuee traffic demand, the data were entered into the VISSIM simulation model using an .FMA file. In each .FMA file, the scaling factor was 1.0, and the time interval was 15 min long.

Next, the authors identified shelter locations and estimated the parking capacity at each. The Wood River City Engineering suggested that authorities direct traffic towards shelters in the Wood River central parks and city building, the East Alton-Wood River high school and recreation building, the Madison County health department, the Eastwood Elementary school, and a nearby church. The city authority did not allow any other shelter locations outside the city area. It was assumed that shelter centers’ parking areas could be fully occupied by evacuee vehicles, and the rest of the evacuee vehicles would park on-street in the surrounding road network. In order to estimate the parking capacity of the shelter location, the authors used aerial photos from Google Maps. The total parking capacity at these shelter centers was approximately 1225. The authors assumed another 780 vehicles would park on neighboring streets, the Eastwood Elementary School playground, the East Alton Recreation Building’s field, or the church’s field. Moreover, it was assumed that 25% of those evacuee vehicles would evacuate out of the local area, towards the homes of local friends and family.

In the simulation model, the authors denoted these shelter locations as zone 15. When a shelter center reached parking capacity, the simulation model used dynamic routing to divert evacuee vehicles to the next closest available shelter. Officials from the City of Wood River suggested that law enforcement officers should be stationed in front of each shelter center to guide traffic when shelters fill.

In the evaluation configuration, the authors collected total delays, intersection data (termed node data), and vehicle travel times in the simulation model network. In the direct output decision, the authors decided to collect node data (raw data at intersections) and vehicle travel times (raw data) from 0 s to 7700 s. For the vehicle travel time measurements, the authors selected five particular evacuation routes in this research study. Those were as follows: route 1—from IL 111 through E. Edwardsville Rd. to the shelters; route 2—from IL 111 through N. 9th St. to the shelters; route 3—from IL 111 through N. 6th St. to the shelters; route 4—from IL 111 through IL 143 and N. Wood River Ave. to the shelters; and route 5—from Walcott St. through IL 143 and N. Wood River Ave. to the shelters.

These roadways provided an opportunity for evacuees to change evacuation routes in response to roadway conditions. In Figure 1, the red-marked areas denote the oil refinery, the green-marked areas signify at-risk residential areas, and the red lines show the evacuation routes. As shown, this scenario would require residents to evacuate towards the north. In this research study, zone numbers 1 to 9 have been denoted as evacuee traffic zones. In addition, background traffic zones have been represented as zone numbers 10 to 14, and zone number 15 as the shelter location. The authors denoted each zone number on the basis of the population, household numbers, and roadway. A vehicle was considered to be evacuated safely when it reached the shelter location, zone 15.

Alternative Simulation Scenarios

To make an effective evacuation plan for this study, the authors recommended two alternative evacuation scenarios. First, the authors prioritized that an evacuee can select his/her destination according to convenience, safety, and the shortest possible travel time. For this reason, the authors created a different simulation model where vehicles traveling to shelter zone 15 were allowed to use local roads in addition to the primary evacuation route, IL 143 (E. Edwardsville Rd). To simulate this option, the authors placed zone 15 at different locations, providing vehicles with the option to select from more routes to their destination. The authors refer to this alternative as scenario 1.

After running the scenario with existing traffic infrastructure, the authors observed that during the evacuation there was a large amount of traffic congestion at the IL 143/IL 111 intersection. Many evacuee vehicles from zones one to nine wanted to make a northbound left turn toward the primary shelter centers. To solve this problem, the authors calculated the evacuee vehicle demand for the IL 143/IL 111 intersection. Because of this traffic demand, the decisive intersection for the evacuation was the intersection of IL 143/IL 111.

Subsequently, the authors input only those calculated vehicle numbers at that particular intersection in the Synchro and optimized that intersection’s traffic signal cycle length. The phasing of the traffic signal was not changed. The optimized cycle length was 110 s, with a northbound left turn green time of 45 s, whereas before optimization, the cycle length was 65 s, and the northbound left turn green time was only 8 s. Figure 2 and Figure 3 show the existing and Synchro output of optimized phasing and signal timing information for the intersection of IL 143 and IL 111, respectively.

Although the northbound left turn green signal timing (45 s) was longer than the maximum (30 s) time suggested by the FHWA [29], the authors considered the timing appropriate to the scenario. According to the IDOT timing sheet, northbound left turn was only permissive, not protected; therefore, southbound through and northbound left-turn traffic competed for the same right-of-way. For optimization, the authors applied a protected movement for all left-turn phases. For the northbound shared through and right turn, the authors suggested 61 s green timing. After optimization, the maximum volume-capacity ratio was 1.02 and intersection LOS D. In addition, after optimization, the LOS at the other signalized intersections along IL 111 and IL 143 also performed at a LOS D at the time of evacuation, except for IL 143-IL 3 (LOS C), Vaughn Rd.-IL 143 (LOS C), and old St. Louis Rd.-IL143 (LOS C). The authors dismissed signal coordination alternatives because the signals along this corridor were not coordinated. This combination of signal timing optimization and flexible shelter choice will be referred to as scenario 2.

4. Analysis and Results

In this section, the analysis of the results acquired from the different simulation scenarios is presented. As described previously, the authors developed three different simulation scenarios for a no-notice evacuation. Those simulation scenarios included an existing conditions scenario (existing scenario), a flexible shelter choice simulation scenario (scenario 1), and a scenario optimizing signal timing and allowing flexible shelter choice (scenario 2). To analyze each of the scenarios, the authors simulated each ten times with random seed numbers.

When comparing the traffic performance between the three scenarios, the authors first considered the average vehicle delay for the entire network (see Figure 4a). Although scenario 2 reduced the delay considerably, the results suggested that the average vehicle delay increased in scenario 1. To identify the cause of this result, the authors visually inspected the simulated traffic congestion between the existing scenario and scenario 1. The inspection revealed that allowing drivers choices in their shelters and evacuation routes caused vehicles to divert from the primary evacuation route to local streets. This change resulted in an increase in overall congestion, likely because of the frequent actuation of side-street detectors of traffic signals along the IL 111 evacuation route, the primary east–west evacuation corridor. The frequent actuation of side-street traffic detectors caused observable congestion along that corridor, sometimes resulting in gridlock, further reducing the performance of the IL 143/IL 111 intersection. The margins of error for the base scenario, scenario 1, and scenario 2 were 7.2, 7.0, and 3.8 s per vehicle, respectively.

Next, the authors examined the performance of the key intersection along the evacuation route, where a large proportion of vehicles must turn left. The results, presented in Figure 4b, suggest that both scenarios 1 and 2 were effective in reducing the congestion at the intersection of IL-143/IL-111. The margins of error for the base scenario, scenario 1, and scenario 2 were 5.1, 1.9, and 0.9 s per vehicle, respectively. The reduction from scenario 1 was likely caused by fewer vehicles choosing routes through this intersection, and the reduction from scenario 2 was likely caused by both fewer vehicles and optimized signal timing. As described in the Section 3, the authors considered several options for optimizing the signal timing, including simulating manual traffic control from a law enforcement officer. This investigation suggested that manual traffic control similar to Parr et al. [19] might not be optimal because of the large left-turn evacuation volume.

Comparing the total time vehicles were stopped during their evacuation revealed similar trends to the delay findings. When considering the total stopped time for the entire network, shown in Figure 4c, scenario 1 increased the average stopped time compared to existing conditions. Scenario 2 showed the best overall performance. The margins of error for the base scenario, scenario 1, and scenario 2 were 7.0, 6.7, and 3.1 average seconds of delay per vehicle, respectively. When comparing Figure 4a,c, it appears that nearly all of the average delay from scenario 2 was from stopped conditions. Further consideration was given to the key intersection of IL-143/IL-111. As shown in Figure 4d, scenario 1 reduced the average stopped time the most. The margins of error for the base scenario, scenario 1, and scenario 2 were 4.7, 1.8, and 0.7 s, respectively.

Next, the authors compared the average number of stops for vehicles throughout the network and at the intersection of IL-143/IL-111. Figure 4, parts e and f, illustrate similar trends as the other performance measures discussed. The margins of error were lower than 0.8 for the entire network and less than 0.09 for the intersection of IL-143/IL-111. Thus, the authors considered scenario 2 as the most promising method for reducing congestion during the evacuation under investigation.

The last performance measure reviewed was travel time. This measure was recorded for several routes within the model, as described previously and shown in Figure 5. Route 1 results were omitted because of the small sample size of vehicles completing this route, and route 5 was shorter in distance than the others, thus the travel time axis is different (see Figure 6). Results from routes two to five illustrate that both scenarios 1 and 2 reduced travel times along the key evacuation routes. These results also suggest that scenario 2 will provide the most efficient operation of the primary evacuation route.

To identify if the differences in network performance between scenarios were statistically significant, the authors applied both Z-tests and T-tests, as appropriate. The p-values from these comparisons are shown in

Table 2. Statistical comparison of network performance.

	p-Value for Existing Scenario vs. Scenario 1	Existing Scenario (Mean Value)	Scenario 1 (Mean Value)	p-Value for Existing Scenario vs. Scenario 2	Existing Scenario (Mean Value)	Scenario 2 (Mean Value)	p-Value for Scenario 1 vs. Scenario 2	Scenario 1 (Mean Value)	Scenario 2 (Mean Value)
T-test p (T ≤ t) one-tail value (delay, s)	0.00	105.04	139.01	0.00	105.04	78.75	0.00	139.01	78.75
T-test p (T ≤ t) one-tail value (stop delay, s)	0.00	89.86	120.04	0.00	89.86	58.91	0.00	120.04	58.91
T-test p (T ≤ t) one-tail value (number of stops, #)	0.02	2.57	2.84	0.02	2.57	2.29	0.00	2.84	2.29
Z-test p (Z ≤ z) one-tail value for route 2 (travel time, s)	0.00	2335.85	1335.35	0.00	3296.06	586.68	0.00	1335.35	586.68
Z-test p (Z ≤ z) one-tail value for route 2 (delay time, s)	0.00	3241.64	1272.71	0.00	3241.64	534.51	0.00	1272.71	534.51
Z-test p (Z ≤ z) one-tail value for route 3 (travel time, s)	0.00	2335.85	1128.23	0.00	2335.85	605.22	0.00	1128.23	605.22
Z-test p (Z ≤ z) one-tail value for route 3 (delay time, s)	0.00	2249.93	1064.97	0.00	2249.93	540.04	0.00	1064.97	540.04
Z-test p (Z ≤ z) one-tail value for route 4 (travel time, s)	0.00	3580.28	1292.87	0.00	3580.28	706.34	0.00	1292.87	706.34
Z-test p (Z ≤ z) one-tail value for route 4 (delay time, s)	0.00	3434.8	1183.11	0.00	3434.8	595.19	0.00	1183.11	595.19
Z-test p (Z ≤ z) one-tail value for route 5 (travel time, s)	0.01	176.15	228.54	0.03	176.15	124.31	0.00	228.54	124.31
Z-test p (Z ≤ z) one-tail value for route 5 (delay time, s)	0.00	101.68	161.48	0.04	101.68	85.42	0.00	161.48	85.42

. The authors accepted α = 0.05, which suggested statistically significant differences existed between each of the performance measures evaluated. These findings suggest that the differences portrayed in Figure 4 are statistically significant, which supports the authors’ observations of the simulated traffic operations.

Overall, these results suggest that for an evening no-notice evacuation requiring a left-turn movement along the primary evacuation route, the overall delay can be reduced by implementing an extended left-turn interval (45 s in this study) and allowing evacuees to choose alternate routes. If an evacuation is required during off-peak hours, traffic signals might run at low traffic volumes and be inefficient in handling such demands. The results of this study show value in preparing alternative signal timing plans that could improve evacuation efficiency, particularly for large left-turn volumes. Ideally, these timing plans could be designed in advance, loaded into the traffic controllers, and implemented by traffic operations personnel with remote access to those controllers.

Additionally, the findings suggest that law enforcement and traffic operations personnel need not dedicate resources to guiding traffic to the arterial routes. Instead, allowing evacuees to choose local streets could reduce evacuation delays if the traffic signals run timing plans specifically for evacuations.

Table 3. Average intersection delay measures for the IL 143/IL 111 intersection.

Existing Scenario			Scenario 1			Scenario 2
Delay	TStopd	Stops	Delay	TStopd	Stops	Delay	TStopd	Stops
174.43	150.21	3.51	139.19	113.59	3.317	79.20	56.30	2.47

demonstrates the IL 143/IL 111 intersection’s average delay (seconds), TStopd (stopped delay in seconds within the node), and stops (number of stops within the node, starting from crossing the start section) for three scenarios. Each of these measures of effectiveness was recorded in 15-min time intervals for the existing and the two alternative simulation scenarios.

Figure 7 illustrates the graphical comparisons of the average travel time of each particular evacuation route for the existing and two alternative evacuation scenarios. These findings also suggest that scenario 2 would reduce the evacuation travel times the most effectively.

After the assessment of the vehicle travel time comparisons, it can be concluded that on route 4, there is the highest vehicle travel time for the existing traffic infrastructure scenario. Because this route is the longest (3.25 miles) of the evacuation routes, this finding is expected. Furthermore, scenarios 1 and 2 have the second and third highest vehicle travel times, respectively. Scenario 2, signal optimization, reduced the travel time on all evacuation routes compared to the existing scenario. Scenario 1 decreased the travel time on nearly all routes compared to the existing conditions. Although the travel time for route 5 slightly increased in this scenario, the change was small because the route was short. Although this scenario did significantly reduce the travel time for most routes, the change was not as large as scenario 2.

5. Discussion and Recommendations

This study sought to identify the traffic signal timing and evacuee guidance that could expedite the evacuation of a residential area adjacent to an oil refinery. The exceptional criteria of this study are the high volume of left-turn evacuation demands. According to the residential shelter center locations and background traffic volume, the authors estimated the origin–destination (O-D) traffic volume. In this research investigation, the dynamic flows are described in origin–destination matrices and are not formatted to describe flows at an intersection. These origin–destination matrices are applied as a .FMA file in the VISSIM simulation software version 7. After the O-D matrix, the authors observed that the IL 143/IL 111 intersection had to handle approximately 1400 evacuee vehicles within one hour, whereas more than 900 vehicles turned left to the northbound through IL 143. The estimated total number of evacuee vehicles was approximately 2300. It was challenging to optimize intersection signal timing with a high volume of evacuee vehicles within a limited (one hour) time period. According to the literature review, no articles have focused on signal timing optimization for a high volume of left-turning evacuees. In addition, according to the literature review, most of the evacuation procedures depend on the interstate facility. However, this no-notice emergency evacuation study was conducted without the use of an interstate facility.

The authors did not consider contraflow or one-direction traffic operations in this study. Because starting these traffic operations requires more time. As a result, this traffic operation strategy would not be effective or reasonable to evacuate the affected areas within one hour just after the chemical explosion in the oil refinery. In addition, fire trucks, emergency medical responders, and law-enforcement vehicles need to use the local road network during this emergency situation. For those reasons, the authors did not apply the contraflow or one-direction traffic flow strategy in this research. When organizing contraflow evacuations, accessibility concerns for both emergency responders and evacuees are crucial.

The results suggested that the most effective evacuation was enabled when evacuees were allowed to search for their own fastest routes using local roads and when a different signal timing was implemented where two arterial routes intersected. Specifically, scenario 2 evacuated the most vehicles with the shortest vehicle travel times along several evacuation routes, generating the lowest total delay, stopped delay, and number of stops at the key intersection for the evacuation. Specifically, the findings suggested that a cycle length of 110 s at the intersection of IL 143/IL 111 and an extended left-turn phase (45 s) optimized the evacuation traffic flow. Overall, these changes reduced the total travel time along evacuation routes by 78 percent compared to existing conditions. Whereas, by implementing scenario 1, total travel time along evacuation routes decreased by 57 percent compared to the existing conditions. Similarly, in terms of delay, scenario 2 reduced the total delay time by approximately 80 percent. Scenario 1 reduced the delay by 59 percent compared to the existing conditions. In addition, the cumulative average travel times for scenario 2 were 48 percent lower than scenario 1, and the total average delay times for scenario 2 were 52 percent lower compared to scenario 1. In addition, the Z-test and T-test analysis results suggested strong evidence (α = 0.05) that scenario 2 had significantly faster travel time, less delay time, stopped delay, and number of stops than both scenario 1 and the existing scenario. Overall, it was found that optimizing signal timing and allowing drivers flexibility in their route and shelter choice could improve evacuation performance compared to existing conditions and the other proposed scenario where drivers only allow flexibility in their route and shelter choice during a no-notice evacuation case study.

The key contribution of the study was demonstrating the value of planning for possible evacuations where residential areas neighbor industrial sites. These plans should include special evacuation signal timing designs for key intersections, particularly where large left-turn volumes exist. Without such efforts, if law enforcement officers are not available to help control traffic efficiently, the results of this study suggest there is a notable increase in delay, which could result in the loss of life. The results also suggested that simply letting drivers choose their own routes through local streets could increase overall evacuation delays unless operation is improved at key signalized intersections. The recommendations from this study could apply to signalized intersections where arterials meet near industrial facilities with neighboring residential areas.

Future research on this topic could investigate how law enforcement might manually control traffic at intersections with such high left-turn demands, evaluate evacuation towards other shelter locations, consider evacuations during different times of the day, model the evacuation of other nearby residential areas, and evaluate adaptive traffic signal operations.