Adaptive traffic signal control offers a way to coordinate intersections as traffic conditions change, rather than applying one fixed timing plan regardless of demand. A swarm-intelligence approach described in research presented at INISCOM 2017 models each intersection as a decision-making unit and coordinates those units across a network. The underlying idea is relevant to engineers and system designers working with connected infrastructure: local signal decisions can be informed by neighboring traffic conditions to support broader network performance.
Why network-wide signal coordination matters
Traffic conditions at an intersection are connected to conditions elsewhere in the road network. Vehicles discharged from one approach may become entering traffic at another intersection, so a signal decision that reduces one queue can affect queues downstream. Treating intersections in isolation can therefore miss the effects of traffic moving through the wider system.
Connected vehicle and infrastructure communications can make current traffic information more available to traffic management systems. The study uses this context to motivate an adaptive approach that responds to dynamic traffic rather than depending only on predetermined signal timings. Its focus is a multi-intersection network, where coordination is especially important as the system becomes larger and more complex.
How swarm intelligence informs adaptive traffic signal control
The proposed method draws on particle swarm optimization (PSO), an optimization technique inspired by collective behavior. In the traffic model, each intersection is treated as a virtual particle, while the group of intersections forms a particle swarm. Each intersection makes signal-phase decisions based on its local traffic state and information shared with nearby intersections.
This is not a matter of every signal independently selecting the phase that appears best at that instant. The approach also considers a network-level objective: improving traffic conditions across the connected intersections. Individual decisions are iteratively adjusted, allowing the system to balance local queue conditions with the broader network state.
Queues and signal phases as decision inputs
The model represents traffic on each movement using queue length. At each update, the queue changes according to the difference between incoming and outgoing vehicle flow. Outgoing flow depends on both the current queue and whether the corresponding signal is green; a red signal prevents vehicles on that movement from being released in the model.
Signal choices are also constrained by phase sequencing. The study uses an eight-phase dual-ring signal arrangement based on NEMA phase control. A signal cannot transition arbitrarily from one phase combination to another: the next allowable phase depends on the current phase. The adaptive algorithm selects among the available next-phase options, rather than ignoring operational phase constraints.
Distributed decisions across connected intersections
In the PSO-inspired formulation, each intersection maintains candidate probabilities for its possible next signal phases. These values are updated iteratively using information associated with the intersection’s own best result and the best result for the network. The method then normalizes the values so they remain valid probabilities and selects a phase from the permitted options.
The practical significance is the distributed structure. Each intersection can use its real-time traffic state, exchange relevant signal and state information with nearby intersections, and adjust its control in coordination with them. This creates a process in which local decisions can evolve toward a network-wide arrangement, rather than requiring a single fixed plan to serve every traffic condition.
What the simulation results indicate
The evaluation compared the proposed approach with conventional fixed-time control in a simulated network of four intersections. Both approaches used the same initial conditions and parameters. The simulation varied incoming traffic volume and the balance between left-turn and straight-through movements; it also distinguished conditions where incoming flow remained below the modeled output capacity from saturated conditions where it exceeded that capacity.
Under lower, unsaturated demand, fixed-time control performed reasonably, and the difference between the methods was relatively small. Under saturated demand, the adaptive approach reduced the maximum queue more effectively. In the reported scenario with a left-turn-to-straight-flow ratio of 1.0 and an incoming volume of 4,000 vehicles per hour, the maximum queue was about 26% lower than with fixed-time control. The paper also reports more visible improvement at the higher tested volume of 4,800 vehicles per hour.
When the left-turn-to-straight ratio was 0.5, both control approaches performed better overall than in the 1.0-ratio condition, while the adaptive method continued to show improvement. The researchers also examined movement-level queues at a tested volume of 3,200 vehicles per hour. Because the algorithm seeks to reduce the longest queue, it can encourage a more balanced distribution of traffic among movements compared with the fixed-time approach.
Considerations for traffic system designers
The results point to several design considerations for engineers and traffic operators evaluating adaptive control:
- Assess the network, not only individual junctions. Traffic arriving at an intersection may be shaped by signal decisions at other locations.
- Respect phase-transition rules. Optimization must operate within the signal controller’s permitted phase sequence.
- Test across demand patterns. Performance can differ between unsaturated and saturated traffic, as well as with different turning-movement mixes.
- Choose an appropriate objective. Minimizing the maximum queue is one way to promote more balanced conditions, but it defines what the control algorithm prioritizes.
The reported evaluation is a simulation of a four-intersection network. Its results demonstrate potential under the tested conditions, but do not by themselves establish performance for every road layout, demand profile, or deployed control system.
Conclusion
Swarm-intelligence adaptive traffic signal control frames intersections as connected decision-makers that can respond to queue conditions and coordinate phase selection across a network. In the evaluated simulation, its advantage over fixed-time control was most evident under saturated demand, where reducing the longest queue became especially important. For traffic engineers and system integrators, the central lesson is to consider signal timing as a network-level control problem, while validating the approach against local phase constraints and the demand patterns the system must manage.