Abstract
In this paper, the author designs a kind of adaptive dynamic programming to be used to optimize method of the parameters, and designs the frame structure in order to optimize traffic signal fuzzy logic controller. This paper chooses the evaluated neural networks and control module (fuzzy logic controller) iterative updating algorithm, and simulation research is done in different traffic flow conditions. The results show that, in reducing the average vehicle delay, the effect of adaptive dynamic programming optimization of fuzzy logic controller is superior to not through optimization of fuzzy logic controller.
Introduction
Traffic congestion is a nonlinear, time-varying challenge that significantly impacts urban efficiency. While Dynamic Programming (DP)—a mathematical method for making a sequence of optimal decisions—is theoretically ideal for timing traffic lights, it suffers from the "curse of dimensionality." As the number of lanes, sensors, and variables increases, the computational power required to find an optimal solution grows exponentially, making real-time application unfeasible.
Traditional controllers often rely on fixed timers or simple induction logic, which changes lights based on vehicle detection. Fuzzy Logic Controllers (FLC) offer an improvement by handling environmental uncertainty; however, they are typically static, meaning their internal rules do not adapt to changing traffic patterns. This research proposes a hybrid system that bridges this gap by creating a controller that learns and optimizes itself dynamically.
Methodology: Adaptive Dynamic Programming
The core of the proposed system is the use of Adaptive Dynamic Programming (ADP) to tune the parameters of a Fuzzy Logic Controller. Instead of solving massive DP equations directly, the framework utilizes three interconnected neural networks to approximate the optimal solution.
The ADP Framework Components:
- Evaluation (Critic) Network: Estimates the "cost-to-go," predicting the expected future delay.
- Action Network: Determines the control quantity, such as extending a green light phase.
- Model Network: Represents the intersection environment.
The authors specifically employ Action Dependent Heuristic Dynamic Programming (ADHDP), where the evaluation network takes both the current state and the chosen action as inputs to refine its predictions. Furthermore, the system uses a relative utility function that compares current delay to a moving average of the previous 10 steps. This approach makes the controller more robust to the inherent randomness of traffic flow compared to using absolute delay metrics.
Simulation and Environment
To validate the model, the researchers simulated a standard crossroad intersection. The environment includes front and rear detectors to monitor traffic flow and queue lengths across four directions.
The system was tested against four distinct traffic flow scenarios:
- Sets 1-3: Constant traffic flows ranging from low (720 cars/h) to high (2400 cars/h).
- Set 4: A "mutated" flow scenario where traffic volume shifts abruptly over a 2000-second period to simulate peak hour transitions.
Key Results
The ADP-optimized Fuzzy Logic Controller (ADFPLC) was compared against standard Induction Control and a baseline Fuzzy Logic Controller (FLC).
- High Traffic Performance: In the heaviest traffic scenario (Set 3), the ADFPLC achieved an average delay of 30.70 seconds/car, outperforming the standard FLC (31.45s) and induction control (34.94s).
- Adaptability: In the mutated flow test (Set 4), the ADFPLC maintained the lowest delay (19.14s), demonstrating its ability to handle sudden traffic spikes more gracefully than static controllers.
- Convergence: Simulation data indicated that the ADP-optimized system converges to a stable, lower delay state significantly faster than non-optimized methods.
Conclusion
By using neural networks to approximate the Bellman equation, this research successfully bypasses the curse of dimensionality, allowing for sophisticated real-time traffic optimization. The results suggest that while Fuzzy Logic is effective for handling traffic uncertainty, it requires the adaptive capabilities of ADP to adjust parameters as environmental conditions change.
Future work may address the limitations of this study, such as moving beyond simulated environments to include real-world factors like pedestrian crossings and weather, and exploring how multiple ADP controllers can be networked to optimize traffic across an entire city grid.
Summary updated