25th ARRB Conference – Shaping the future: Linking policy, research and outcomes, Perth, Australia 2012

Upstream and downstream detection to improve congested network operation

Chin, D., ARRB Group Pty Ltd, Australia

Luk, J.ARRB Group Pty Ltd, Australia

George, A., VicRoads, Australia

No. of words : 6889

No. of figures : 8

No. of tables : 4

Total : 9889 words

ABSTRACT

With significant increases in urban population and new car registration, travel in urban areas is becoming increasingly congested. With limited resources to expand urban road networks, road agencies in Australia are trying to manage thecongestion through smarter strategies. The use of pre-emption strategy (using upstream detection to allow more throughput of downstream trafficin anticipation of peak upstream traffic demand), and gating strategy (using downstream detection to lessen the throughput of upstream traffic into a congested area) to reduce network delay in an area traffic control system, was considered in this paper. The paper provideda brief reviewof current practices and, then reported results obtained through microsimulation modelling in which pre-emption and gating strategies were applied to a simple test network that was subjected to peak period oversaturation. The study showed that both pre-emption and gating strategies were successful in reducing network delay by demonstrating the underlying principles: the pre-emptive use of a sufficiently longer cycle time in anticipation of the arrival of peak traffic demand; the use of gating before the occurrence of obstructive queue spillback; and the improvement of network throughput bythe redistribution of spare cycle time to green time of side street approaches when gating the through main road movement.

Introduction

In Australia, total travel in urban areas has grown ten-fold over the last 60 years and private road vehicles now account for about 90 per cent of the total urban passenger task(BITRE 2007). Consequently, road traffic congestion in urban areas especially in large Australian citieshas continued to worsen with significant increases in population and new car registrations. About a million new motor vehicles per year were registered in Australia in the years 2005-2007 before the global financial crisis (The Motor Report 2008). With limited resources to expand road networks in congested cities, road agencies in Australiahave constantly put in efforts to manage congestion using smarter strategies such that overall network congestion or delay can be minimised. Particularly on the arterial roads, these strategies were incorporated intoarea traffic control (ATC) systems such as SCATS or STREAMS making them adaptive to manage congestion(see, e.g. Lowrie 2001). By nature, adaptive ATC systems have vehicle detection capabilities embedded across an area. In this respect, the ATC systems in Australia at presentdo not employ information from detectors at the upstream end of a road link but mostly information from one set of detectors near the stopline of a road link for signal operations. For examples, in SCATS, these detectors are positioned about 1.5m in advance of the stopline and in STREAMS, they are about 35 m in advance of the stopline. However, some questions were raised that information from the extra detectors at upstream end of road link could address the following two critical issues,more adequately than the present stopline detector configuration:

  1. Congestion can build up very fast at the beginning of the peak period. With the detection of arrival of congestion early in the peak period through monitoring the upstream traffic demand, the cycle times and relevant phase times could be increased to deal with the congestion early in the peak period and at intersections further upstream.
  2. Downstream congestion can cause the queue in a link to ‘spillback’ and block upstream traffic movements entering the link. With the detection of spilling back of queue from downstream congestion, the green time of the traffic movements from upstream road links could be reduced so as to avoid the blocking of the upstream traffic movements entering the downstream link.

Austroads funded a study (Austroads 2012) mainly to investigate how detector data from upstream and downstream of congested sites could be used to improve route and/or network operations. The approach proposedin the study aimed to explore how the stopline detectors in the present ATC systems could still address the above two issues effectively. This approach thus formed the crux of this paper. In (1) above, the proposed approach suggested that stopline detectors from upstream intersections should be able to provide the required information about the arrival of congestion. Then in (2) above, the stopline detectors immediately upstream should be able to register the flow restriction or spillback (also called ‘spillover’).

This paper is divided into four main parts that are systematically linked to assess the effectiveness of pre-emption and gating strategies using information from stopline detectors. The first part describes briefly reported current practices and related studieson arterial roads. The second partpresents the methodology for setting up a microsimulation model to investigate the effectiveness of and, to identify implementation principles behind the pre-emption and gating strategies. Then the third and fourth partsdescribethe microsimulation results upon the respective implementation of pre-emption and gating strategies.

review of Current practices

Thissection gives a brief review of practices in Australia and New Zealand that wereconducted mainly through theR&D projects on transport operationfunded by Austroads (an association of Australian and New Zealand transport and traffic authorities). Thus the practices reviewed are representative of those among the state road agencies and it is envisaged thatthe practices share some similarities with those of overseas transport jurisdictions of similar capacity.

Balancing traffic density

Balancing traffic density or distributing traffic evenly in a road network can help in allocating congestion in a road network in an equitable manner such that the overall network congestion or delay can be minimised. However, the current practices of balancing traffic density or distributing traffic evenly in a road networkare generally subjective and differ according to different dominant expert opinions. With regard to the use of an ATC system on arterial roads, Austroads (2010) documented these expert opinionsand investigated these practices. The key findings from this investigation that are of relevance to this paper are given as follows:

  • Both theory and practice confirm that a road network is a storage device in itself. Arrival flows in excess of capacity are common occurrences in peak periods and will be stored as vehicle queues. In a road network, as long as the exit flow at all times is maximised and there is enough storage space for the excess queues, the aim of congestion management is achieved.
  • A traffic system is a complex system and yet most ATC systems use only a small number of vehicle detectors or a small detection area to monitor and controlthe traffic, e.g. few detectors per approach spanning multiple lanes, or detectors clustered close to the signalised stop-line. Therefore, the solution to better ATC is better utilisation of existing detector data, or adding extra detectors e.g. at the upstream end of a road link. The Road and Maritime Services (former RTA) of NSW Australia has researched the use of up to 32 detectors to run up to 48 signal groups at a signalised intersection. The potentially large amount of detector data should allow for the accurate optimisation of signal timings through feedback and, implementation of pre-emption and gating schemes on arterial roads.
  • Gating schemes based on an ATC system have been operational in Australia and overseas to resolve queue blocking or spill-over situations. The general principles to apply gating to resolve queue blocking or spill-over situations included :

-identify bottleneck links

-identify gating links

-choose threshold values for gating to become active

-distribute gating over several intersections

-employ progressive gating with the gating intensity as a function of congestion.

The above principles were also incorporated into the microsimulation model in this paper to test gating strategy. For gating to be successful, there must be enough space to store queuing vehicles so that various levels of gating can be implemented. Gating is one form of capacity management and any restraint should be introduced progressively and distributed over several intersections. The gating scheme should also be compatible with a network operations plan to avoid traffic leaking to other routes.

The use of traffic signal offsets in an ATC system

An ATC system has the key function of reducing delay and vehicle stops on a road link by the coordination of signals, i.e. choosing the correct offset. An offset is the difference in the start (or end) times of the green periods of two adjacent intersections. Theoretically, at an optimum offset and given sufficient downstream green time, a platoon of vehicles from upstream will experience little delay as it is able to pass through the downstream intersections without stopping. Here, the general practice in Australia and New Zealand is to:

  • coordinate signals in the direction of major traffic movements even at low traffic demand (mid-night) conditions
  • coordinate the starting offsetsto reduce vehicle stops,and therefore facilitate platoon progression.

It is noted that the principle of starting offsetsspecifies that the beginning of the green start times of adjacent intersections be coordinated(Austroads 2009). It has the benefit of allowing the front of the platoon to move forward without stopping and should also minimise spillback. Usually,the offset is adjusted so the signal would turn green before the main platoon arrives so that vehicles waiting at a stopline would be cleared first. An alternative approach is to adopt finishing offsets, which aims to get the tail end of a platoon through an intersection, and therefore has the potential to minimise delay, i.e. fewer vehicles at the tail end would be held up for departure in a following green period. With this approach, the front end of the platoon would usually arrive at a red phase, thus increasing thequeue lengths and the potential for spillback.

Detector locations

The issue of where to locate detectors on an arterial road link is related to balancing the cost of installing detectors with the amount of traffic information required for the purpose of performance monitoring and adaptive control. Such issue was well debated in Australia and New Zealand (Hulscher & Sims 1974 and Austroads 2009). If only one set of detectors were used due to cost constraints, the conclusion then was that they should be located at about 1.5 m in advance of the stopline as in SCATS, or about 35 m in advance of the stopline as in STREAMS. Such location is more beneficial than the detector location at the upstream end of the road link (also called the departure-side detectors) for the following reasons:

  • Detectors at or near the stopline allow a clear indication of lane discipline/use.These detectors therefore enable accurate vehicle-actuated (VA) control.
  • Based on a loop length of about 4.5 m, a detector at or near a stopline can serve as a presence detector and a passage detector.

On the other hand, with a second set of detectors upstream beyond the maximum back of queue (‘mid-block detectors’), or at the upstream departure-sidelocation, accurate measurement of arrival flows can be obtained and used to estimate the queue length and delay which are useful data for network performance monitoring and adaptive traffic control (Luk, 1984, Luk 1989). Note that the queue lengths so estimated are vertical queues at a stopline and need to be adjusted to give an estimate of the horizontal queue – an elaborate approach adopted in more recent versions of the TRANSYT program from the UK Transport Research Laboratory (2010). However,with a second set of detectors, the following issues should be noted:

  • Departure-side detectors should benefit from less communication costs by making use of the controller at the upstream intersection, and are useful at shorter links where queue spillback frequently occurs.
  • The installation of mid-block detectors on a long link (0.5–1 km) could incur significant communication costs, and should be considered only when queue length estimation is required as part of ATC or network operation. For long links, the demand data from mid-block detectors does not consider traffic from side-streets and traffic generators between the detectors and the stopline.
  • With detectors at the upstream end and the stopline end of a road link, information on demand and supply is now available. The information is not very useful unless road agencies are prepared to invest in the development of queue formation and dissipation software in SCATS or STREAMS to calculate real-time queue lengths and delay. Such an investment can be substantial.

Arguing against the use of a second set of detectors, it is noted that SCATS has long utilised information from detectors (SAs or Strategic Approaches) located upstream to gain some early information to manage downstream congestion, and is not restricted to the use of detectors in a particular intersection. Thus the focus of the research in this paper is to determine how to make better use of the information currently available from near stopline detectors. Further, the information can be from several upstream intersections, not just the immediate upstream intersection.

METHODOLOGY FOR SETTING UP MICROSIMULATION MODEL

Using information currently available from near stopline detectors, the aim of the methodology was to set up a microsimulation model to investigate the effectiveness of and to identify principles suitable for implementing the pre-emption and gating strategies. The methodology involved the following:

Define traffic network

To facilitate the aim of the methodology, the Austroads project team suggested that a simple hypothetical linear network with minimal traffic movements of only passenger cars and fixed signal phaseswas appropriate for the microsimulation test model. Shown in Figure 1, the simple network consists of five intersections on an arterial route, representing a typical subsystem in an ATC system. The mainroad was a one-way (west to east) through road intersected by five one-way (north to south) side streets. Itconsisted of two lanes both for through movements west to east. Each side-street had three lanes, one of which was for left-turning movement into the main road, and the others for through movement. As the focus of the paper was on managingthe congestion on main road, no right turning movement was added to approach on main road at intersections. At each intersection there were 4.5 m loop detectors positioned 1.5 m in advance of the stopline of each approach lane, on both the main road and side-street. The speed limit on the through road was set at 70 km/h. The middle intersection (C) was the critical intersection in the subsystem with properties such as a smaller capacity or a shorter link distance (such as CD) that could be prone to queue spillback. At each intersection, the signal plan was a fixed-time two-phase control to alternate traffic flow between side-street and main road.

Figure 1: Hypothetical linear network model showing intersections A-B-C-D-E

Setting up of pre-emption strategy

The logic of the pre-emption strategy is to progressively increase cycle time in anticipation of increasing traffic demand through the ATC system as in the morning peak period. Then the ATC system would have avoided the worse situation of not having enough cycle time or capacity on arrival of the increased demand. Otherwise, overflow queues will occur. This cycle time strategy is expressed as follows:

If volume at A > capacity at C, then apply pre-emption and increase cycle time by 6 s for the whole subsystem in the next cycle with 70% of extra cycle time allocated to the through movement and 30% of extra time to the side street traffic; otherwise do nothing.

The initial settings on the simulated network shown in Figure 1 were as follows:

  • An initial cycle time of 90 s was applied to all intersections A, B, C, D and E.
  • The offset for the main road traffic was empirically determined and an optimal starting offset of 26 s for progression in congested conditions was used between intersections.
  • As given in the above cycle time strategy, C was selected as the critical intersection. So in order to create the condition of impending peak arrival flows exceeding the capacity at C, only the initial green time set for through movement for this intersection C was reduced. The initial green time set for through movement for other intersections A, B, D and E remained status quo. Thus, the signal plan was a fixed-time two-phase control with an initial 48 s green time set for the through movement at intersections A, B, D and E. But for intersection C, an initial green time of 38 s was set for the through movement. This meant that at a cycle time of 90s and a green time of 38 s, the designed two-lane main road capacity at C was 3600 x 38/90 = 1520veh/h, and that at A, B, D and E was 3600 x 48/90 = 1920 veh/h. 3600 represents the total saturation flow of a two lane road or 1800 veh/h per lane. The pre-emption measure was triggered whenever the flow at A or B exceeded the fixed value of 1520 veh/h (the capacity of C calculated above).

As given in the above cycle strategy, the cycle increment of 6 s was used in the study for every trigger of the pre-emption measure and this 6 s was appropriate as it was based on experience in adaptive ATC. With an initial signal setting of 90s, the cycle time was increased incrementally by 6 s and capped at 144 s in the preemption tests. For every cycle increment, the extra time of 6 s was allocated at 70% to the main road and 30% to the side-streets at each of intersections A, B, C, D and E. Thus, the green time for the through movement increased from 38 s to 74 s at the critical intersection C, and from 48 s to 84 s at A, B, D and E. A cycle increment of 18 s (instead of 6 s) was also investigated in this study, but was found to have little impact in this simple network. An 18 s increment had similar performance as a 6 s increment when traffic progressed smoothly with no side-street traffic, but failed to help when queue spillback had occurred. Further research on a different network configuration is needed to investigate this issue.