💻 computer science

Population State-Aware Cooperative Adaptive Differential Evolution for Multi-Objective Arterial Traffic Signal Coordination

This paper proposes a Population State-Aware Cooperative Adaptive Differential Evolution (PSAC-DE) algorithm featuring a unified state controller to solve a four-objective arterial traffic signal coordination model, demonstrating significant improvements in delay, stops, green wave bandwidth, and carbon emissions compared to mainstream optimizers in simulated Hangzhou scenarios.

Original authors: Bing Bai, Chengzhi Wang, Zhimeng Li, Xiaozheng Li

Published 2026-07-01

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Bing Bai, Chengzhi Wang, Zhimeng Li, Xiaozheng Li

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a busy city street with six traffic lights in a row. Right now, these lights often work against each other: you get a green light, drive a bit, and then hit a red one, forcing you to stop, start, and stop again. This "stop-and-go" pattern wastes time, burns extra fuel, and creates a lot of carbon emissions.

This paper introduces a new, smarter way to control these traffic lights using a computer algorithm called PSAC-DE. Think of it as a highly intelligent traffic conductor that doesn't just follow a fixed schedule but learns and adapts in real-time to keep traffic flowing smoothly.

Here is a breakdown of how it works, using simple analogies:

1. The Problem: The "Fixed" vs. The "Chaotic"

Traditional traffic lights are like a metronome: they tick at the same speed no matter what. If traffic is light, they still give you a long green light. If traffic is heavy, they might not give you enough time, causing a backup.
Standard computer algorithms trying to fix this are like novice chess players. They are good at moving pieces, but they often get stuck in the same patterns, make mistakes when the board gets complicated, and struggle to balance different goals (like making cars move fast and stopping them from polluting the air).

2. The Solution: The "Smart Team" (PSAC-DE)

The authors created a new algorithm called Population State-Aware Cooperative Adaptive Differential Evolution (PSAC-DE). That's a mouthful, so let's break it down into three simple ideas:

The "Team" (Population): Instead of one computer trying to solve the puzzle, imagine a team of 100 different "traffic planners" working at the same time.
The "State Awareness" (The Coach): This is the paper's big innovation. Usually, these teams work in isolation. In this new system, there is a central coach (called the Unified State Controller) that watches the whole team.
- Early in the game: The coach sees the team is exploring new ideas and tells them to be bold and try wild, different strategies.
- Later in the game: The coach sees the team is close to a solution and tells them to focus, refine, and polish the best ideas.
- The Magic: The coach constantly checks the team's "mood" (are they stuck? are they confused?) and instantly adjusts the rules for everyone to keep them moving forward.
The "Cooperation" (Cooperative Adaptive): The team members don't just work alone. They share their best tricks. If one planner finds a great way to time a light, the coach helps spread that trick to the others who are struggling.

3. The Goal: Balancing Four Conflicting Desires

The algorithm tries to solve a very difficult puzzle with four goals that often fight against each other:

Less Waiting: Minimize how long cars sit at red lights.
Fewer Stops: Minimize how many times cars have to brake and accelerate.
Green Wave: Maximize the "green wave" (the length of time you can drive without hitting a red light).
Cleaner Air: Minimize carbon emissions.

Analogy: Imagine trying to bake a cake that is sweet, low-calorie, cheap, and fast to make. Usually, making it sweet makes it high-calorie; making it cheap makes it take longer. The algorithm finds the perfect "sweet spot" where you get a great cake without sacrificing too much on any one front.

4. The Results: A Real-World Test

The researchers tested this system on a real six-intersection road in Hangzhou, China, under three different traffic conditions:

Heavy Traffic (Over-saturated): When the road is jammed.
Peak Traffic: When it's busy but moving.
Random Traffic: When traffic is unpredictable.

They compared their "Smart Team" against seven other popular traffic control methods. The results were clear:

The "Smart Team" won in almost every category.
Delays dropped by 25%: Cars spent much less time waiting.
Stops dropped by 45%: Cars stopped less often, meaning smoother driving.
Green Wave expanded by 133%: The "green wave" (the ability to drive through without stopping) more than doubled.
Emissions dropped by 27%: Because cars weren't idling or accelerating as much, the road produced significantly less pollution.

5. The Final Decision: Picking the Winner

Since the algorithm produces many good solutions (a "Pareto front"), the researchers used a mathematical tool called EW-TOPSIS to pick the single best plan to actually install. Think of this as a judge tasting all the cakes the team made and picking the one that is the perfect balance of taste, health, and cost.

Summary

In short, this paper presents a new, self-correcting computer brain for traffic lights. Unlike old systems that follow rigid rules, this system acts like a dynamic coach that watches the traffic, senses when the team is stuck, and instantly changes the strategy to keep cars moving, stops to a minimum, and the air cleaner. It proved to be significantly better than existing methods in a real-world city setting.

Technical Summary: Population State-Aware Cooperative Adaptive Differential Evolution for Multi-Objective Arterial Traffic Signal Coordination

1. Problem Statement

Rapid urbanization and increasing vehicle volumes have led to severe arterial congestion, characterized by long travel delays, frequent stops, and elevated carbon emissions. Conventional fixed-time signal control lacks the adaptability to handle dynamic traffic flows, particularly during peak hours or stochastic surges. While metaheuristic algorithms like Differential Evolution (DE) have been applied to traffic signal optimization, standard DE variants suffer from several critical limitations in this domain:

Parameter Sensitivity: Performance is highly dependent on manually tuned hyperparameters (scaling factor $F$ and crossover rate $CR$).
Premature Convergence: A lack of balance between global exploration and local exploitation often leads to stagnation in complex, high-dimensional, and tightly constrained solution spaces.
Isolated Adaptation: Existing adaptive DE variants typically decouple parameter adaptation, mutation strategy selection, and population management, lacking a unified mechanism to respond to real-time evolutionary states.
Model Limitations: Current traffic optimization models often fail to simultaneously address multi-dimensional objectives such as throughput, operational efficiency, fuel economy, and emission mitigation.

2. Methodology

2.1 Optimization Model

The study formulates a high-dimensional, constrained multi-objective optimization problem for a six-intersection arterial corridor. The decision vector includes the system common cycle time, effective green times for each phase at each intersection, and relative phase offsets. The model minimizes four conflicting objectives:

Average Vehicle Delay: Calculated using a comprehensive Webster–Akçelik model.
Vehicle Stop Frequency: Derived from traffic wave theory.
Green Wave Bandwidth Deficit: The gap between actual and theoretical maximum bandwidth, calculated via a geometric spatiotemporal mapping method to avoid complex modulo operations.
Network CO₂ Emissions: Estimated using a macro-level regression model calibrated with empirical data, linking emissions to delay and stop metrics.

Constraints include physical limits on cycle lengths, green time bounds, and saturation degrees (maintained $\leq 0.95$ ).

2.2 Proposed Algorithm: PSAC-DE

The paper introduces Population State-Aware Cooperative Adaptive Differential Evolution (PSAC-DE). The core innovation is a Unified State Controller (USC) that integrates four adaptive modules into a closed-loop co-evolution system based on a real-time "population state" variable ( $\sigma$ ).

Population State Awareness: A unified state indicator ( $\sigma$ ) is computed by fusing normalized Hypervolume (convergence) and population diversity metrics. This indicator partitions the evolutionary process into three states: Exploration, Transition, and Exploitation. A hysteresis mechanism and moving average filter ensure stable state transitions.
UCB-Based Strategy Learning: A pool of six mutation strategies is managed using an Upper Confidence Bound (UCB) mechanism. The strategy pool is dynamically restricted based on the current evolutionary state (e.g., exploration-oriented strategies are prioritized in the early phase).
SHADE-Based Parameter Adaptation: Control parameters ( $F$ and $CR$) are sampled from a success-history memory (SHADE) but are subject to state-dependent truncation. Bias offsets are applied to the memory update process to drift parameters toward larger step sizes during exploration and finer tuning during exploitation.
Dynamic Subpopulation Scheduling: The population is dynamically partitioned into three subpopulations (Exploration, Balanced, Exploitation) with ratios shifting according to the state $\sigma$ . Knowledge transfer occurs between subpopulations via elite sharing and a shared history memory to prevent isolation.
Constraint Handling: A hierarchical repair mechanism (cycle balance, boundary truncation, saturation repair) and a penalty function ensure that infeasible solutions are handled rigorously, with feasible solutions always dominating infeasible ones.

2.3 Decision Framework

To select a deployable solution from the Pareto front, the study employs an Entropy Weight-TOPSIS (EW-TOPSIS) method. This objective post-processing approach calculates weights based on information entropy and selects the solution with the maximum closeness coefficient to the ideal positive baseline.

3. Key Contributions

State-Aware Cooperative Control (SACC) Framework: Unlike existing adaptive DEs that adjust modules independently, this framework establishes a unified population-state representation to coordinate mutation strategies, parameter adaptation, and subpopulation scheduling simultaneously.
PSAC-DE Algorithm: A novel algorithm integrating UCB strategy learning, SHADE parameter adaptation, and dynamic subpopulation scheduling, all regulated by shared state information to ensure consistent search behavior across evolutionary phases.
Four-Objective Arterial Model: A comprehensive optimization model that simultaneously minimizes delay, stops, and carbon emissions while maximizing green wave bandwidth, addressing the multi-dimensional demands of modern urban traffic management.

4. Experimental Results

The algorithm was validated on a six-intersection arterial corridor in Hangzhou, China, under three traffic scenarios: Over-saturated, Peak-saturated, and Stochastic Fluctuation. Performance was compared against seven baseline algorithms (NSGA-II, MOEA/D, JADE, SHADE, L-SHADE, MSDE) over 30 independent runs.

Performance Metrics: PSAC-DE consistently outperformed all baselines in Hypervolume (HV), Inverted Generational Distance (IGD), and Spread metrics across all scenarios. For instance, in the over-saturated scenario, it achieved an HV of 0.1272 compared to 0.1246 for the second-best (L-SHADE), with a 27.1% improvement in IGD.
Statistical Significance: Non-parametric Wilcoxon Signed-Rank and Friedman tests confirmed that the performance improvements were statistically significant ( $p < 0.05$ ).
Ablation Studies: Removing the Population State Awareness (SA) module caused the most significant performance drop (17.9% reduction in HV), confirming its role as the "core driving brain." Other modules (UCB, Subpopulation, SHADE) also contributed positively, with the SA module being the most critical.
Operational Benefits: When applied to the peak-saturated scenario, the optimized timing plan (selected via EW-TOPSIS) demonstrated substantial improvements over conventional pre-timed schemes:
- Average Delay: Reduced by 25.1% (from 68.4s to 51.2s).
- Vehicle Stops: Reduced by 44.8% (from 1.43 to 0.79 stops/veh).
- Green Wave Bandwidth: Increased by 133% (ratio from 0.18 to 0.42).
- CO₂ Emissions: Reduced by 26.6% (from 145.6 g/s to 106.9 g/s).
- Throughput: Increased from 3680 to 4015 veh/h.

5. Significance and Claims

The paper claims that PSAC-DE advances the theoretical paradigm of adaptive evolutionary computation by introducing a rigorous, unified population state-awareness framework that resolves the "strategy-parameter mismatch" common in existing DE variants.

Practically, the study asserts that the proposed method delivers an efficient, deployment-ready solution for multi-intersection urban arterial coordination. It successfully balances high operational efficiency with low-carbon environmental goals, offering a robust strategy for mitigating urban traffic gridlock and reducing automotive carbon emissions without requiring extensive manual parameter tuning. The results suggest that the algorithm possesses strong engineering value for real-world implementation in complex, constrained traffic networks.