Traffic Intersection Optimization: Queue Theory and Signal Timing
Every red light is a miniature queuing system. Engineers use mathematical models — Poisson arrivals, Webster's formula, and level-of-service grades — to minimise the collective time drivers spend waiting. Understanding these models explains why signals behave the way they do, and why small timing errors can cascade into city-wide gridlock.
1. Queuing Theory at an Intersection
A signalised intersection is a textbook example of a D/D/1 or M/D/1 queuing system — vehicles (customers) arrive at a stop line, wait in queue, and are served (discharged) during the green phase at a finite rate. The discipline of queuing theory, pioneered by Danish engineer Agner Krarup Erlang in 1909 for telephone exchanges, applies directly to traffic flow.
Three quantities define the system's behaviour:
- Arrival rate (q): The mean number of vehicles arriving per unit time, typically expressed in vehicles per hour (vph) or vehicles per second.
- Service rate (s): The maximum rate at which queued vehicles can depart during a green phase — the saturation flow rate, around 1,800 vehicles per hour of green (vphg) for a standard 3.65 m lane.
- Signal timing: The fraction of time a lane receives green (the effective green ratio g/C), which acts as a throttle on the effective service rate available.
When the arrival rate exceeds the effective service rate — that is, when q > s(g/C) — the queue grows without bound. The intersection is said to be oversaturated. The goal of signal optimization is to prevent this condition while minimising average delay across all approaches.
2. Poisson Arrival Model
On a road where vehicles travel independently and there is no upstream signal creating platoons, arrivals at an intersection follow a Poisson process. This means the number of vehicles arriving in any interval of length t has the probability:
The Poisson assumption works well for isolated intersections with v/c (volume-to-capacity) ratios below about 0.85. At higher volumes, or when the intersection is downstream of another signal, arrivals cluster into platoons and a binomial or Erlang-k distribution gives a better fit. The Pearson chi-squared goodness-of-fit test is routinely used by traffic engineers to verify which model applies to field-collected headway data.
The Poisson model has a key practical consequence: even at moderate loads, random clustering means some cycles will occasionally see more vehicles than can clear in a single green. This residual queue effect increases average delay non-linearly as the intersection approaches saturation — one of the core reasons oversaturation is so damaging.
3. Saturation Flow and Degree of Saturation
Saturation flow rate (s) is the rate at which a queue of vehicles would discharge through a stop line if given a continuous green signal. It is measured empirically by recording headways between successive departing vehicles. After the first few vehicles (which have above-average headways due to start-up reaction time), headways stabilise at the saturation headway h, so s = 3600 / h vphg.
Traffic engineers target x ≤ 0.85–0.90 on critical lanes to provide a buffer against random fluctuations. An intersection where the critical-lane degree of saturation consistently exceeds 0.95 will experience frequent overflow queues and unpredictable delay.
4. Webster's Optimal Cycle Formula
F. V. Webster's 1958 formula for the optimal signal cycle length that minimises total vehicle delay at a two-phase signalised intersection is:
The formula assumes Poisson arrivals and uniform demand throughout the cycle. Extensions by Akcelik (1981, Australian Road Research Board) add a correction term for oversaturated and time-varying demand conditions, improving accuracy near capacity. The UK design standard (DMRB TD 19) uses a variant that accounts for pedestrian crossing demands and shared-lane effects.
Why Longer Cycles Are Not Always Better
Intuition suggests a longer green phase lets more vehicles through. But Webster's formula reveals a counterintuitive truth: very long cycles increase total delay because vehicles spend more time waiting at red before the cycle begins. There is a genuine optimum, and doubling the cycle length from, say, 60 s to 120 s can increase average delay by 30–50% when demand is moderate.
5. Level of Service (LOS)
The Highway Capacity Manual (HCM) — published by the Transportation Research Board in the USA and used internationally — defines Level of Service for signalised intersections based on average control delay per vehicle (d, in seconds). Delay includes deceleration, stopped time, and acceleration.
| LOS | Average Delay (s/veh) | Description |
|---|---|---|
| A | ≤ 10 | Excellent — nearly free flow, minimal stops |
| B | 10 – 20 | Good — short delay, most vehicles clear first cycle |
| C | 20 – 35 | Acceptable — occasional cycle failures begin |
| D | 35 – 55 | Tolerable — noticeable congestion, some overflow |
| E | 55 – 80 | Poor — near capacity, frequent overflow queues |
| F | > 80 | Failure — oversaturated, queue grows without bound |
Urban design standards typically require LOS D or better at peak hour for new development approvals. A signalised intersection operating at LOS F during peak periods is a candidate for capacity improvement — widening, signal retiming, or grade separation.
The HCM delay formula for a signalised intersection combines three components: uniform delay (deterministic, assuming arrivals exactly at the mean rate), incremental delay (accounts for random arrivals and cycle failures), and an initial queue delay term for pre-existing overflow:
6. Phase Splits and Lost Time
Once the optimal cycle length C is determined, green time must be allocated among phases (called phase splits). The standard approach distributes effective green time proportionally to each phase's critical-lane flow ratio:
Lost time is the time during a signal phase when the intersection is not productively used. It has two components: start-up lost time (drivers in the queue react late to green — the first vehicle loses about 1.5 s, the second 1.0 s, and so on until the queue is moving at saturation headway) and clearance lost time (the final vehicle that clears on yellow may not fully use the amber interval). Minimising lost time — for example by using a two-phase rather than four-phase signal plan where pedestrian safety allows — directly increases capacity.
7. Adaptive Signal Control
Fixed-time signal plans are optimised for historical average demand. Real traffic fluctuates — a football match, a school bell, or a single lane-blocking incident can shift flows dramatically within minutes. Adaptive signal control systems address this by using real-time detector data to continuously update timing parameters.
SCOOT (Split Cycle Offset Optimisation Technique)
Developed by the UK Transport Research Laboratory in the 1980s, SCOOT is deployed in over 250 cities worldwide. Inductive loop detectors measure occupancy and flow continuously. A traffic model runs in real time, and the system makes small incremental adjustments every cycle to splits (green allocation within a phase), cycle length, and offset (the relative timing between adjacent signals). Small changes prevent overcorrection instability. SCOOT studies in the UK report average delay reductions of 12–15% versus optimised fixed timing.
SCATS (Sydney Coordinated Adaptive Traffic System)
Developed by Roads and Maritime Services New South Wales (Australia). SCATS selects signal timing plans from a pre-computed library based on detector-measured degree of saturation. It is simpler than SCOOT's continuous model approach but equally effective in practice, and is deployed in over 100 cities across Australia, New Zealand, Ireland, and China.
Machine-Learning Approaches
Research since 2015 has explored reinforcement learning (RL) for adaptive signal control. An RL agent observes queue lengths and delay, and learns a policy that maximises cumulative reward (minimises total delay) through simulated and real-world interaction. Studies using the SUMO traffic simulator show RL agents outperforming SCOOT by 10–25% in complex multi-intersection scenarios, particularly under highly non-stationary demand. Deployment challenges include safety validation, interpretability, and the difficulty of training on real networks where mistakes have real consequences.
8. Signal Coordination and Green Waves
An isolated optimised signal may still frustrate drivers on an arterial corridor if adjacent signals are not coordinated. A vehicle clearing one green light should ideally arrive at the next signal during its green phase — this is the concept of a green wave or progression band.
Coordination is achieved through offset — the time difference between green starts at successive signals. For a one-way street with uniform spacing d (metres) between signals and a posted speed v (m/s), the ideal offset is simply d/v seconds. For two-way streets, a perfect green wave in both directions is only possible when signal spacing equals half the distance a vehicle travels per cycle — a constraint rarely met in practice, requiring compromise optimisation.
Tools such as TRANSYT (Transport Network Study Tool, TRL UK) model an entire network of signals simultaneously and use a performance index combining stops and delays to find the offset and timing plan that minimises network-wide delay. TRANSYT has been the standard British design tool since 1969.
The bandwidth of a green wave — the fraction of the cycle during which vehicles can travel through consecutive intersections without stopping — is the key metric of progression quality. Maximising bandwidth on a major arterial while maintaining acceptable service on cross-streets is the central trade-off in urban signal coordination.
9. Key Takeaways
Summary
- A signalised intersection is a queuing system: arrival rate, saturation flow, and effective green ratio determine whether the queue clears each cycle or grows without bound.
- Poisson arrivals model independent vehicles accurately at low-to-moderate volumes; platoon effects require binomial or Erlang distributions.
- Webster's formula (1958) gives the optimal cycle length as C = (1.5L + 5)/(1 − Y) and remains the international standard for fixed-time design.
- Level of Service (A–F) grades average control delay: LOS D (35–55 s) is the typical urban design threshold; LOS F means the intersection has failed capacity.
- Adaptive systems (SCOOT, SCATS, RL agents) reduce delay 10–25% versus optimised fixed timing by responding to real-time demand.
- Signal coordination through offsets creates green waves, reducing stops and fuel consumption on arterial corridors.
- Lost time per phase (start-up + clearance loss ≈ 3–5 s) directly reduces capacity; minimising phase count increases effective green time available.