🚦 Traffic Light Optimization with Reinforcement Learning -and -QNN

This project provides a reinforcement learning-based framework for optimizing traffic flow at complex urban intersections. A Deep Q-Learning agent is trained to intelligently control traffic signal phases in order to minimize congestion and waiting times, ultimately improving traffic efficiency.

---

🧠 Deep Q-Learning Agent

⚙️ Framework

• Algorithm: Q-Learning with a Deep Neural Network
• Objective: Optimize traffic light phases to reduce cumulative waiting time

🧭 Environment

• Intersection: 4-way, with 4 incoming and outgoing lanes per arm (each 750 meters long)
• Traffic Lights: Dedicated lanes for specific movement directions (left, straight, right)

🚘 Traffic Generation

• Each episode simulates 1000 vehicles
• Traffic patterns vary dynamically across episodes

🎯 Agent (Traffic Light Control System - TLCS)

• State: Discretized vehicle positions on lanes (80-cell binary vector)
• Action: Selection from 4 pre-defined traffic light phases, each lasting 10 seconds
• Reward: Based on the reduction in cumulative vehicle waiting time
• Learning: Q-learning with a neural network to approximate Q-values for state-action pairs

---

🚀 Getting Started

✅ Prerequisites

Ensure the following dependencies are installed on your system:

Tool	Version
Python	3.11
SUMO Simulator	1.2.0
TensorFlow	2.16.1
TensorFlow GPU	2.5.0 (optional)
NumPy	1.19.5
Matplotlib	3.6.0

📦 Setup Instructions

1. Install Anaconda
   Download Anaconda

2. Install SUMO
   Download SUMO

3. Clone the Repository

git clone https://github.com/yourusername/traffic-light-optimization.git
cd traffic-light-optimization

4. Install Dependencies

pip install -r requirements.txt

5. Set SUMO Environment Variable

export SUMO_HOME="/path/to/sumo"

---

🏃 Running the Algorithm

Start the training process:

python training_main.py

For testing:

python testing_main.py

---

📂 Code Structure

Component	Description
Model	Defines the neural network architecture
Memory	Handles experience replay buffer
Simulation	Manages the traffic environment and interaction with SUMO
Traffic Generator	Generates vehicle routes for each episode
Visualization	Plots performance metrics (reward, delay, queue length)
Utils	Manages configuration, file handling, and SUMO setup

---

⚙️ Settings

All training and testing parameters are specified in:

• training_settings.ini
• testing_settings.ini

These include:

• Simulation duration
• Number of episodes
• Vehicle generation count
• Neural network layers and learning rate
• Traffic light durations (green/yellow)

---

📊 Results Summary

Metric	Value
Accuracy	80%
Precision	85.83%
Recall	80%
F1 Score	79.05%

Visualizations for reward trends, queue lengths, and waiting times are generated and saved automatically during training and testing.

---

contact KABILASH S -kabilash0108@gmail.com

📜 License

This project is licensed under the MIT License.