Addis Ababa Digital Twin: Traffic Prediction & Optimization AI

🌍 Project Overview

This project implements a Digital Twin of the Bole Road / Meskel Square corridor in Addis Ababa using the SUMO (Simulation of Urban MObility) engine. It uses real-world OpenStreetMap data to simulate traffic flow and employs a Machine Learning (Random Forest) pipeline to predict congestion 5 minutes in advance.

🏗️ System Architecture

The project is modularized into three core components:

1. The Environment (src/simulation):
   • Map Generation: Automatically pulls real-time coordinates (8.995, 38.760) from OSM to build the Bole road network.
   • Traffic Logic: Implements an Adaptive Signal Control system that dynamically extends green lights based on sensor queue length.
2. The Data Pipeline (src/utils):
   • Harvests telemetry data (vehicle count, speed, timestamps) every second.
   • Visualizes network performance using Matplotlib (Congestion vs. Flow analysis).
3. The AI Core (src/ai):
   • Model: Random Forest Regressor (n_estimators=100).
   • Feature Engineering:
     • Temporal: Uses Simulation Step (Time of Day) to distinguish between traffic buildup and clearing.
     • Spatial: Uses Lag Features (t-1min, t-5min) to track flow trends.
   • Performance: Achieved >0.90 R² Score on a 5-minute prediction horizon (Interpolation).

📂 Repository Structure

Addis_Traffic-AI/
├── config/             # Global simulation settings
├── data/
│   ├── raw/            # Telemetry logs (traffic_log.csv)
│   ├── processed/      # Generated graphs and reports
│   └── models/         # Trained AI models (.pkl)
├── src/
│   ├── ai/             # Prediction Logic
│   ├── simulation/     # SUMO Manager & Map Configs
│   └── utils/          # Map Generators & Data Visualizers
└── README.md

🚀 Installation & Usage

⚠️ Important: Model Setup

To keep the repository lightweight, the pre-trained model is not included. You must run the training pipeline once to generate it:

# 1. Generate Synthetic Training Data (30 Days)
python src/utils/generate_synthetic_data.py

# 2. Train the AI Model
python src/ai/traffic_predictor.py

# 3. Launch the Dashboard
python src/app.py

### 2. Setup

# Create environment
python -m venv venv
source venv/bin/activate

# Install dependencies
pip install traci pandas numpy scikit-learn matplotlib pyyaml requests

3. Running the Digital Twin

Step A: Generate the Map (Optional) Repulls the latest road data from OpenStreetMap.

python src/utils/generate_map.py

Step B: Run the Simulation Launches the GUI and starts the Adaptive Control System.

python src/simulation/sim_manager.py

Step C: Train the AI Trains the predictor on the newly harvested data.

python src/ai/traffic_predictor.py

📊 Results & Validation

• Congestion Management: The Adaptive Control system successfully maintained network flow above critical thresholds (5 m/s) during peak load (traffic density tuned to 1.5).
• Realistic Simulation: Traffic generation uses a Bell-Curve Profile (Rise $\to$ Peak $\to$ Clear) to ensure the AI learns how traffic dissipates, fixing previous data bias issues.
• Model Performance:
  • Training Strategy: Multi-Day Synthetic Data (30 Days).
  • Validation Method: Chronological Split (Trained on Days 1-25, Tested on Days 26-30).
  • Accuracy: Achieved 0.97 R² on completely unseen future days, proving the model understands traffic physics and is not just memorizing data.