MICROSIMULATION MODELLING OF INTELLIGENT SPEED ...

MICROSIMULATION MODELLING OF INTELLIGENT SPEED ADAPTATION SYSTEM

Ronghui Liu and James Tate Institute for Transport Studies University of Leeds, Leeds LS2 9JT

ABSTRACT Intelligent speed adaptation systems use in-vehicle electronic devices to enable the speed of the vehicles be regulated externally. They are increasingly appreciated as a flexible method for speed management and control, particularly in built-up areas. A number of different implementations of the systems are planned for large scale on-road trials both in Europe and the UK. This paper describes the developments made to enhance a traffic microsimulation model in order to assess the network impacts of the speed control strategies. Simulation modelling of such a system is carried on a real-world network to investigate the control impacts on traffic congestion and pollution. The work is part of a research program funded by DETR, which also includes driving simulator experiments and road trials of the systems. 1. INTRODUCTION Management of vehicle speed has a direct impact on road safety. Recently there has been a growing interest in the potential of Intelligent Speed Adaptation (ISA) systems, also know as External Vehicle Speed Control (EVSC). The system is a technical device fitted to a vehicle, enabling the speed to be externally regulated. The external activation is achieved by a communication infrastructure in the form of roadside beacons, or an autonomous system using on-board digital maps combined with a Global Positioning System (GPS). The main advantage of ISA systems relative to other forms of urban speed control measures (such as 20mph zones or traffic calming measures) is their flexibility. The systems allow for different control speeds at different times of day and different locations (e.g. outside schools and during school starting and finishing times), and under different traffic, roadway and weather conditions. The systems were designed initially as a speed management measure for the urban environment. However, there is no technical restriction of the systems being applied on motorways, where they can work in similar manner as controls by variable speed signs (such as that implemented on M25). In this latter application, the system aims to improve stability and homogeneity of traffic streams in order to reduce interactions/conflicts among vehicles, which in turn can reduce accidents. On-road trials of in-vehicle speed limited vehicles were first conducted in Sweden (Almquist et al. 1991). More extensive trials have been subsequently carried out in four Swedish cities (Lind 1999). In the earlier Swedish road-trail project (with a

system where in-vehicle electronics is linked to the speedometer system and is activated by roadside transmitters), the speed control proved very popular, especially near schools. In spite of its great potential and the actual road trials of the system, this area has attracted very little research attention. Much of the discussion has been focused on the design specification of the systems, and the experiments are limited in network size and traffic conditions. There appears to be no study on the control effect on drivers’ behaviour and route choice. In particular, there is no measure of the control effect on road capacity, network congestion and pollution. Simulation modelling techniques, on the other hand, provide a cost-effective, controlled environment for evaluating the systems in a network context, and particularly for studying the complicated interactions between equipped and non-equipped vehicles during transition periods. The study presented here is part of a research program funded by DETR to investigate the benefits or dis-benefits and driver behaviour under various ISA control systems, and to find the most effective way of integrating ISA to the UK market. The objectives of the research are to: 1. Review previous and current research on ISA. Assess a number of potential alternatives in implementing external speed control, with particular regards to the costs and benefits of the systems, and compare with traditional speed reduction measures; 2. Identify major alternative approaches for system implementation. Design and build a test vehicle enacting these speed control systems; 3. Study these systems in-depth with driving simulator experiments and on-road trials to investigate the safety benefits and driving behavioural adaptation effects; 4. Examine the network effects and benefits of these systems on various types of road and traffic conditions using a traffic microsimulation model; 5. Select the most appropriate control system to realize the best combination of benefits to traffic management and accident risk based on studies 3 and 4, and produce recommendations on the implementation of the system. This paper describes the development made to enhance a traffic network microsimulation model DRACULA (Liu et al, 1995) to meet the fourth task above. The model is suitably adapted to explicitly represent the driving behaviour under ISA. Simulation modelling of the identified ISA systems will be carried out on a number of real-world networks to examine the control impacts on different road types under different levels of congestion. Together with its emission module, DRACULA will be used to provide for each scenario traffic performance measures (e.g. travel time, speed, flow, queue) and environmental indicators (in terms of exhaust emissions and fuel consumption). Section 2 of the paper outlines the research approach and Section 3 and 4 describe the modelling techniques developed to represent driver behaviour under ISA. Results from the simulation of an ISA system in an urban environment are

presented in Section 5. Finally Section 6 draws some conclusions, and presents directions for further research. 2. RESEARCH APPROACH Earlier work in the project has identified four major alternative ISA implementations in which the speed is controlled. These fall into two categories: 1. Mandatory systems which automatically limit vehicle’s maximum speed to either a prevailing fixed speed limit, or to a speed limit varying with road geometry; and 2. Voluntary systems which only provide speed limit warnings to the driver, such as the system proposed for Borlaagn, Sweden (Lind 1999) which alerts the driver if they infringe a speed limit, and registers a violation if the warning is ignored. Driver compliance can be achieved either through the normal use of the vehicle control (the advisory systems) or by driver enacting the change, say by pressing a button to allow the vehicle’s speed be regulated to the speed limit (the driver selection systems). On the basis that the advisory system is to be studied in another project and its benefit is predicted to be far less than the mandatory systems, it was decided to study the following three systems in the project: 1. Driver selection; 2. Mandatory with fixed (legal) speed limits; 3. Mandatory with fixed and variable speed limits. All three systems were studied with the Leeds Driving Simulator, but only the first two systems have been studied with on-road trials because of the difficulty in enacting the third system. The two mandatory systems are represented in the microsimulation model. For modelling purposes, a more useful distinction is the actual number (or percentage) of vehicles under mandatory ISA and the speed limits on each section of the road. The speed limits are inputs to the simulation modelling; they can easily take into account road geometry (to represent the variable mandatory systems). The speed limits do not change over the study time period. In the simulation modelling, the ISA systems are to be examined in order to look at their effects on: 1. Different road types. Most of the research on speed controls carried out to date has been centred on urban roads where a high proportion of accidents have occurred. There is scope for deploying the system to rural roads and to motorways where fatal accidents are high; 2. Under different traffic conditions: congested or non-congested; 3. Various levels of system penetration, in order to represent the transition period when only the new vehicles could be equipped and to investigate the interactions between equipped and non-equipped vehicles; and 4. A range of speed limits.

3. THE DRACULA MICROSIMULATION MODEL DRACULA (Dynamic Route Assignment Combining User Learning and microsimulAtion) is a day-to-day, microscopic computer model of urban traffic assignment and simulation (Liu et al., 1995). It incorporates a microscopic demand model, which represents individual drivers' learning and daily route and departure time choice behaviour, with a microscopic traffic model which simulates individual vehicle movements through the network. The traffic simulation model is based on a fixed-time increment, with the speeds and positions of each vehicle updated every one second. Vehicles are individually characterised, including a technical description (vehicle type, length, maximum acceleration and deceleration) and behaviour of the driver (reaction time, desired speed and distance from the vehicle in front). These characteristics are randomly sampled from normal distributed representations of the type of vehicle, with means and coefficients of variation defined by the user. Vehicles follow pre-defined fixed routes through the network, derived externally from either the microscopic assignment model of DRACULA or from an equilibrium assignment model such as SATURN (Van Vliet, 1982). They are moved in realtime and their space-time trajectories are determined by their desired movements, the traffic regulations on the road and interactions with neighbouring vehicles through a car-following and a lane-changing model. The car-following model attempts to mimic drivers' desired movements and their interactions with the vehicles in front, according to the relative distance away from the preceding vehicle. Normally, a driver applies a controlled acceleration, related to the speed and space differences between his own vehicle and the one he is following. Thus, where a vehicle is very close to the one in front, the driver would be prepared to slow in case the preceding vehicle brakes suddenly. On the other hand, where a vehicle is far away, both from an intersection and the vehicle in front, the driver accelerates freely in order to reach and maintain his desired speed. The lane-changing model is decision-based, similar to that proposed by Gipps (1981). Various lane-changing objectives are considered in the model, where such a manoeuvre can be due to the desire to overtake a slow-moving or parked vehicle, to reach a lane leading to a desired exit or, for buses, to get into or out of a bus layby. Taking the simulated individual vehicles’ speed and acceleration, an emission model in DRACULA considers explicitly the vehicles’ four different driving modes (acceleration, deceleration, cruising and idling) and calculates emissions for pollutants CO, NOx and HC based on emission factors from the QUARTET Deliverables No. 2 (QUARTET 1992). The model also calculates fuel consumption. The fuel-consumption factors are taken from Ferreira (1982) and DOT (1991). It has assumed that fuel consumption factors are constant for vehicles that are in idling and decelerating modes, and vary as a function of speed for cruising vehicles. For vehicles that

are in acceleration mode, the fuel consumption factor is a function of both speed and acceleration of the vehicle. Typical outputs from the DRACULA simulation model include: travel time, travel distance, speed, fuel consumption and pollutant emissions. Outputs may be reported as sums or averages at the end of the simulated period and may be disaggregated, for example by vehicle type, route chosen or for each link in the network. Individual vehicle travel times and space-time trajectories can also be accessed, to analyse issues of variability. Visual evidence is available through an on-line animation of the movements of individual vehicles through the network. 4. IMPLEMENTING ISA IN DRACULA 4.1 ISA equipped vehicles and level of penetration An additional vehicle type is introduced to represent whether a vehicle is under mandatory ISA control. The level of penetration is represented by the percentage of vehicles equipped with ISA; the actual ISA vehicles are then drawn randomly from the whole network according to the user-specified percentage. In the model, the movement of an ISA vehicle is determined not only by the carfollowing rule as described above, but also by the ISA speed limit in a controlled region. The latter is represented by the vehicle adjusting its speed, derived from normal car-following rules to the speed limits as follows:

If the vehicle’s current speed is below the ISA speed limit: • then move according to car-following rules to a speed not more than the speed limit; • otherwise, decelerate gently to the speed limit. Outside the ISA controlled regions, vehicles move according to the car-following rules and their desired speeds, which is a function of the free-flow speed of the link. 4.2 ISA Speed Limits and Additional Link Specifications The ISA speed limits on the network can be specified at an individual link level, and there can be more than one speed limit on one link. This is particularly useful in the rural network where links are long, and where there could be several sections of a link that are of different speed limits.

5. SIMULATION ISA IN AN URBAN ENVIRONMENT 5.1 The Network In this section, the simulation model described above was used to simulate ISA system in a real-world urban traffic network. The network covers two radial routes (A64 and A63) in the east of Leeds, from the outer ring road to the city centre (see Figure 1). There are 240 links connecting 120 intersections. Traffic condition

during two time-periods were simulated, one for the morning peak with some 18,000 car-trips per hour and another the off-peak period with 12,000 trips per hour, in order to examine the performance of ISA under different traffic congestion levels. There were 22 bus routes through the network with total of 102 buses per hour. Both networks were extracted from an existing Leeds SATURN model, built by the local authority. A variable mandatory speed control system was simulated. Two levels of speed limits were set according to road type: a 40mph speed limit is set for the two radial routes and a 30 mph limit for all residential streets, in accordance with those observed on-street. In addition, on one of the entry links on the ring road, a national speed limit of 70mph was identified.

Legend: ATC Count Site 1.

Journey Time Monitoring Point X. 30 m.p.h. (blank) 40 m.p.h. National Speed Limit 1.

N

Ring Road A64.

2. 3.

4. 5. 7.

6.

9.

8.

10. A63. City Centre

Figure 1. ISA speed limit distribution on the east Leeds modelled network (not to scale). Area Traffic Control (ATC) counts at 7 locations and observed journey times along 18 major sections of the network were acquired from the local authority for calibrating the traffic characteristics in the base networks. The SATURN

assignment was calibrated with the observed ATC counts. Figure 2 shows the comparison between observed and assigned flow. It shows that the modelled flows agree well with those detected. These calibrated traffic flows and demand were then used by DRACULA as inputs. Morning-Peak Period

Off-Peak Period

3000

2500

2000

Modelled Flow

Modelled Flow

2500 2000 1500 1000

1500

1000

500

500 0

0 0

500

1000

1500

2000

2500

3000

0

500

1000

Observed Flow

1500

2000

2500

Observed Flow

(a) (b) Figure 2. Comparing assigned and observed flows for the (a) morning-peak and (b) off-peak periods. A DRACULA simulation of the base networks was carried out, and the travel times of individual vehicles were recorded. This information was used to calculate the average journey times, and compare these with the observed data. Figure 3 shows the observed and simulated journey times for the morning peak and offpeak periods. The results show that DRACULA simulated journey times agree reasonably well with those observed. 600

160

KEY :

Observed 120

Modelled Journey Time (seconds)

Journey Time (seconds)

KEY : 140

Observed

500

400

300

200

Modelled 100

80

60

40 100

20

0

0 1 to 2

2 to 3

3 to 4

4 to 5

5 to 6

6 to 7

7 to 8 9 to 10 10 to 4 7 to 6

6 to 5

5 to 4

4 to 3

3 to 2

2 to 1 4 to 10 10 to 9 9 to 8

Link Name

1 to 2

2 to 3

3 to 4

4 to 5

5 to 6

6 to 7

7 to 8

9 to 10 10 to 4

7 to 6

6 to 5

5 to 4

4 to 3

3 to 2

2 to 1

4 to 10 10 to 9

Link Name

(a) (b) Figure 3. Journey time comparison for the (a) am-peak and (b) off-peak periods. 5.2 EVSC Simulation and Analysis The ISA penetration rate was introduced as a control variable. Simulations for both the morning and off-peak time periods were carried out with 10 penetration rates (10, 20, 30,…..100%) of ISA equipped vehicles; results were compared with the base case where there was no mandatory speed control. For each scenario, the model was run 10 times with different random number seeds to establish a

9 to 8

distribution of the output results. Effects of speed controls were evaluated in terms of: • • • •

Network total travel time; Total emissions for pollutants Carbon monoxide (CO), Nitrogen Dioxides (NOx), and HydroCarbons (HC); Fuel consumption; and Total vehicle-hours spent at ranges of speeds (in 5km/hr intervals) for each of the speed limit zones.

The total network travel time, measured in vehicle-hours, is the sum of all individual vehicle journey times from their specified origins and destinations. Figure 4 presents the average travel time distribution with increasing penetration rates of ISA for both the modelled periods. 450

400

Total Travel Time (veh-hours)

350

300

250

200

KEY: 150

AM Peak Off Peak

100

Error Bars

50

0 0

10

20

30

40

50

60

70

80

90

100

ISA Penetration Rate (%)

Figure 4. Travel Time Distribution This shows that the relationship between the level of penetration of ISA equipped vehicles and the overall network travel time does not behave in a similar manner for the two time periods. During the morning peak period, there is a lot of queuing experienced on the network. Therefore vehicles would have very little opportunity to exceed speed limits, and consequently there is very little variation in the average vehicle journey duration. During the off peak period however, there are less vehicles on the network, allowing vehicles to travel at higher speeds, sometimes exceeding speed limits. With the implementation of ISA, the amount of vehicles exceeding the speed limits would decrease, resulting in longer journey times. Therefore during off peak times increasing penetration levels of ISA tends to increase average journey times, whereas during peak periods, the congestion automatically reduces vehicles speeds, and hence ISA has very little influence on their journey times. The change to the total emissions of CO, NOx and HCs only varied by a small degree with different levels of ISA penetration. For example, during the morning peak period, emissions of CO, NOx and HCs only varied by +/- 2% for all ISA penetration rates. Similar results were displayed during the off peak time period.

Reasons for this lack of change, emanate from the emission rate functions. Generally, emission rates decrease with higher cruising speeds. Opposing this, acceleration and deceleration cycles cause proportionally larger proportions of pollutants to be emitted. Without ISA control, vehicles will accelerate and decelerate for longer periods to achieve higher cruising speeds. The acceleration and deceleration components producing relatively large amounts of pollutants to be emitted, whilst the high cruising speed is responsible for much lower quantities. With ISA control, vehicles will accelerate and decelerate less, to achieve reach lower cruising speeds. Therefore the acceleration and deceleration cycles are accountable for lower emissions, but the lower cruising speed is accountable for proportionally more. Consequently, these two variables counter each other, resulting in little variation of total vehicle emission rates with increasing levels of ISA penetration. The effect of speed control on fuel consumption is shown in Figure 5. It shows that the total fuel consumption gradually decreases with increasing penetration levels of ISA equipped vehicles. A total 8% reduction in fuel consumption is achieved during both time periods if the whole fleet of traffic is under the speed control. 7500

Fuel Consumption

Fuel Consumption (litres)

7000

6500

Key: AM Peak Off Peak

6000

5500

5000

4500

4000 0

10

20

30

40

50

60

70

80

90

100

ISA Penetration Rate (%)

Figure 5. Fuel Consumption The reduction in fuel consumption can be explained by the fact that EVSC controls the top speeds of the vehicles, and by doing so, reduces vehicle acceleration and deceleration cycles and keeps the vehicles cruising at slower and constant speeds. The fuel consumption rate (litre/hour) has a strong and positive relationship with vehicle acceleration, deceleration and cruising speed, i.e. vehicles consume more fuel when in hard acceleration and deceleration, and when moving at higher speeds. To establish how ISA equipped vehicles effect the speed distribution on an urban road network, the time each vehicle spent in a specific speed band was recorded. This allowed the total speed distribution for each level of penetration to be calculated in 5 kph intervals. Figures 6 and 7 show the speed distributions on all

the peak and off-peak periods respectively. The x-axis shows the speeds, and the y-axis the total vehicle-hours spent in each speed band. Each row of data represents an ISA penetration rate as a percentage. 30 mph Speed Limit 40 mph Speed Limit

600

Total Vehicle Hours

500

400

Decreasing High Speed Travel Times

300

100 90 80 70 60 50 40 30 20 10 0

200

100

120

100

110

80

90

70

75

60

65

50

55

40

45

30

35

20

25

10

15

5

0

0

% ISA

Speed (km/hr)

Figure 6. Peak Period Speed Distribution. 30 mph Speed Limit 40 mph Speed Limit

400

350

Decreasing High Speed Travel Times

Total Vehicle Hours

300

250

200 100 90 80 70 60 50 % ISA 40 30

150

100

50

20 10 0

0 0

5

10 15 20 25 30

35 40 45 50

55 60 65

70 75

80

90 100 110 120

Speed Bins (km/hr)

Figure 7. Off-Peak Period Speed Distribution. The 3-D bar graphs clearly demonstrate that the proportion of vehicles exceeding the speed limits (30 and 40 m.p.h.) decreases with increasing levels of ISA penetration. In the off-peak scenario, calculations established that in the nocontrol case there were 34% of vehicles exceeding the 30 m.p.h. speed limit. This was zero under full ISA control. Due to more congestion this figure is reduced to 20% during the peak period. Similarly, the level of vehicles exceeding the 40 m.p.h. speed limit were 16% and 12% for the off-peak and peak periods

respectively. Another feature worth noting, is that, in the morning peak period, there was a substantially high proportion of vehicle-hours spent at speed below 10kph. ISA implementation did not alter the proportion of such congested traffic by any degree. This suggests that, though ISA is effective in reducing excessive speeds, it does not induce further congestion. 5.3 Summary of the Simulation Results The general findings were: 1) In the morning peak period, traffic speeds were limited largely by congestion. There was therefore little change in network total travel time, average speed and emissions under the speed control systems; 2) During the off-peak period, however, there was substantial amount of travel speed exceeding the regulated speed limits. With the implementation of ISA, these high speeds were limited, leading to a significant reduction in average speed, therefore increasing total travel time, with increasing speed control penetration rates; 3) In both cases, the speed distributions at lower speeds had virtually not changed. This indicated that speed controls had not induced extra congestion; 4) Speed variation was reduced as a result of the reduction of the fast-moving proportion of traffic. This would help to reduce interactions/conflicts between vehicles, which in turn could reduce the frequency of accidents; 5) During both peak and off-peak periods, fuel consumption decreased significantly with increasing penetration rates. This could be caused by reduced acceleration and deceleration cycles, and lower cruising speeds for the ISA vehicles, as fuel consumption has a strong, positive relationship with vehicle acceleration and deceleration. However; 6) There were no significant variations found in total emissions with the levels of penetration. More analysis is to be carried out with a more sophisticated emission model with more up-to-date low time resolution emission factors. 6. CONCLUSIONS AND FUTHER RESEARCH This paper describes the methodology developed for modelling a new form of speed management: intelligent speed adaptation. The development was made under the general framework of the DRACULA microscopic traffic simulation model. The model was enhanced to explicitly represent the complex interaction between controlled vehicles and other traffic and the control impacts on driver behaviour, network congestion and pollution. Simulation experiments of ISA were conducted in an urban environment. The results show that the speed control is more effective in less-congested traffic conditions than in high congestion; that, whilst ISA reduces excessive traffic speeds in the network, it does not induce more congestion to the network; that ISA reduces variation in speed which in turn reduces accidents; and that ISA helps to reduce fuel consumption and so helps the environment.

Simulation analysis of ISA on a rural two-lane road and a motorway network has also been conducted. In general, the ISA effect on network performance was found to be less significant compared to that in the urban environment. A further step in the research is to incorporate studies from the driving simulator experiments and the on-road trials, particularly in terms of driver adaptational behaviour under the speed control. Another interesting extension to the research is to incorporate signal optimisation with ISA. As the speed control reduces speed variation and produces better vehicle platooning, a greater benefit may be expected.

REFERENCES Almquist, S., Hyden, C. & Risser, R. (1991) Use of speed limiters in cars for increased safety and a better environment. Transport Research Record 1318, 34-39. Ferreira, LJA. (1982) Car fuel consumption in urban traffic, the results of a survey in Leeds using instrumented vehicles. ITS Working Paper 162, Institute for Transport Studies, University of Leeds. Department of Transport (1991) New car fuel consumption: the official figures December 1991, UK Dtp. Gipps, PG (1981) A behavioural car-following model for computer simulation. Transportation Research 15B, 105-111. Lind, G. (1999) Large-scale testing of intelligent speed adaptation – important evaluation issues. Proc. AET European Transport Conference, Seminar D, Vol P432, pp91-103, Cambridge. Liu R., Van Vliet D., and Watling D. P. (1995) DRACULA: Dynamic Route Assignment Combining User Learning and microsimulAtion. Proc. PTRC European Transport Forum, Seminar E, 143-152. QUARTET Deliverable No. 2 (1992) Assessment of current tools for environment assessment in QUARTET, DRIVE II Project V2018: QUARTET, September 1992. Van Vliet, D. (1982) SATURN - a modern assignment model. Traffic Engineering and Control, 23(12), pp 578-581.

In Proceedings of European Transport Conference, Seminar J, 89-100, Cambridge 2000