Highway Intelligent Traffic Shockwave Warning System

AI Decision System for Future Traffic Prediction — Based on Deep Learning Fusion with Physics and Shockwave Theory

English | 中文

---

Demo

Real-time demonstration of the traffic shockwave warning system in action

---

Overview

Taiwan's highways handle over 3 million vehicle trips daily. Traditional passive traffic management struggles with sudden congestion events. This research discovers that traffic congestion propagation shares striking similarities with seismic wave propagation, leading to an innovative hybrid prediction system that combines: Physics-Based Detection: Rankine-Hugoniot kinematic wave theory Deep Learning Prediction: MT-SWNet with graph neural networks RAG-Enhanced AI: Intelligent decision support system

---

🏆 Awards & Recognition

Achievement	Event	Date
🥈 2nd Place (1st Place Vacant)	National Expressway Intelligent Traffic Competition	Oct 2024
🎤 Invited Speaker	2025 Chinese Institute of Transportation Annual Conference	Dec 2025

This project was presented at the 2025 Annual Meeting of the Chinese Institute of Transportation, sharing our innovative approach of combining deep learning with physics-based shockwave theory for highway traffic prediction.

Competition Showcase

Official Competition Poster

2025 CIT Annual Conference

Conference Presentation

---

Key Features

Real-Time Monitoring	Shockwave Detection	AI Prediction	Smart Routing
62 ETC stations across Taiwan	Sub-5 second alert latency	10-60 min advance forecast	Dynamic route optimization

Shockwave Severity Levels

Level	Speed Drop	Visual	Action
🟢 Mild	10-20 km/h	Green	Minor delays expected
🟡 Moderate	20-30 km/h	Yellow	Consider alternative routes
🔴 Severe	30+ km/h	Red	Route change recommended

---

System Screenshots

Driver Interface

Features Highlight

• Interactive traffic map with real-time congestion status
• Shockwave radar visualization with propagation animation
• AI-powered route suggestions
• Personalized departure time recommendations

Admin Control Center

Features Highlight

• Network-wide traffic monitoring
• MT-SWNet prediction visualization (90% confidence)
• AI decision support recommendations
• Historical trend analysis

---

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                        Frontend Layer                           │
│  ┌─────────────┐  ┌─────────────┐  ┌─────────────────────────┐ │
│  │   Driver    │  │    Admin    │  │     Real-time Map       │ │
│  │  Dashboard  │  │   Control   │  │   (React + Leaflet)     │ │
│  └─────────────┘  └─────────────┘  └─────────────────────────┘ │
└───────────────────────────┬─────────────────────────────────────┘
                            │ WebSocket / REST API
┌───────────────────────────┼─────────────────────────────────────┐
│                      Backend Layer                              │
│  ┌──────────────────────────────────────────────────────────┐  │
│  │              FastAPI Server (Port 8000)                  │  │
│  │  • /api/traffic    • /api/shockwave   • /api/prediction  │  │
│  │  • /api/location   • /api/rag         • /api/admin       │  │
│  └──────────────────────────────────────────────────────────┘  │
│                                                                 │
│  ┌────────────────┐  ┌────────────────┐  ┌────────────────┐   │
│  │   MT-SWNet     │  │   Shockwave    │  │   RAG AI       │   │
│  │   Predictor    │  │   Detector     │  │   Assistant    │   │
│  │  (TensorFlow)  │  │ (LWR Physics)  │  │   (Ollama)     │   │
│  └────────────────┘  └────────────────┘  └────────────────┘   │
└───────────────────────────┬─────────────────────────────────────┘
                            │
┌───────────────────────────┼─────────────────────────────────────┐
│                      Data Layer                                 │
│  ┌────────────┐  ┌────────────┐  ┌────────────────────────┐   │
│  │  TDX API   │  │  TISC API  │  │   SQLite + Graph Data  │   │
│  │ (Real-time)│  │ (Real-time)│  │    (62×62 Adjacency)   │   │
│  └────────────┘  └────────────┘  └────────────────────────┘   │
└─────────────────────────────────────────────────────────────────┘

---

MT-SWNet Model

Model Architecture The Multi-Task ShockWave Network features: Encoder: Dual ST-Physical Blocks with spatial attention, temporal attention, and multi-head GCN Decoder: Masked ST-Physical Block with generative inference Multi-task Output: Simultaneous flow, speed, and density prediction Key Innovation: Graph structure embedding with degree, edge, and distance matrices for capturing road network topology.

Performance Comparison

Model	MAE (veh/5min)	RMSE	MAPE (%)	R²
MT-SWNet	12.3	18.7	8.5	0.912
Graph-WaveNet	13.2	19.8	8.9	0.903
AGCRN	13.8	21.2	9.3	0.895
DCRNN	14.5	22.1	9.8	0.887
LSTM	18.9	28.7	13.5	0.798
ARIMA	21.3	32.1	15.7	0.721

---

Shockwave Detection Theory

LWR Traffic Flow Model Based on the Lighthill-Whitham-Richards continuity equation: $$\frac{\partial \rho}{\partial t} + \frac{\partial (\rho u)}{\partial x} = 0$$ Where: ρ = traffic density u = vehicle speed Rankine-Hugoniot Condition Shockwave propagation speed: $$s = \frac{f(\rho_R) - f(\rho_L)}{\rho_R - \rho_L}$$

---

Data Sources & Processing

Data Sources Historical Data from TISC (Taiwan Highway Traffic Database): eTag paired path dynamic information Median travel time by vehicle type (M04A) Median travel speed by vehicle type (M05A) Trip counts by vehicle type (M08A) Real-time Data from TDX Platform: ETC gantry paired section traffic VD vehicle detector real-time data CMS variable message sign information

Data Quality Control Vehicle PCU Standardization

---

Quick Start

Prerequisites

• Python 3.11+
• Node.js 18.0+
• TDX API credentials (Apply here)
• Google Maps API key (Get here)

Installation

# Clone repository
git clone https://github.com/timwei0801/Highway_trafficwave.git
cd Highway_trafficwave

# Backend setup
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt

# Frontend setup
cd frontend && npm install && cd ..

# Configure environment
cp .env.example .env
# Edit .env with your API credentials

Run

# Option 1: One-command deployment
./deploy.sh

# Option 2: Manual startup
# Terminal 1 - Backend
cd api && python main.py

# Terminal 2 - Frontend
cd frontend && npm run dev

Access Points

Interface	URL	Description
🚗 Driver Dashboard	http://localhost:3000/driver	Navigation & alerts
🎛️ Admin Control	http://localhost:3000/admin	System management
📚 API Docs	http://localhost:8000/docs	Swagger UI

---

API Examples

Get Active Shockwaves

import requests

response = requests.get('http://localhost:8000/api/shockwave/active')
shockwaves = response.json()

for sw in shockwaves['active_shockwaves']:
    print(f"Severity: {sw['severity']}, Speed Drop: {sw['speed_drop']} km/h")

Traffic Prediction

payload = {
    "station_ids": ["01F0340N", "01F0360N"],
    "prediction_steps": 12  # 60 minutes
}

response = requests.post('http://localhost:8000/api/prediction/traffic', json=payload)
predictions = response.json()

AI Assistant Query

payload = {
    "query": "Best time to travel from Taipei to Hsinchu?",
    "context": {"origin": "Taipei", "destination": "Hsinchu"}
}

response = requests.post('http://localhost:8000/api/rag/chat', json=payload)
print(response.json()['response'])

---

Performance

87% Shockwave Detection Accuracy

<5s Detection Latency

90% Prediction Confidence

---

Citation

@article{zou2024mtstnet,
  title={MT-STNet: A Novel Multi Task Spatiotemporal Network for Highway Traffic Flow Prediction},
  author={Zou, Guojian and others},
  journal={IEEE Transactions on Intelligent Transportation Systems},
  year={2024},
  doi={10.1109/TITS.2024.3411638}
}

@software{highway_shockwave_system,
  title={Highway Intelligent Traffic Shockwave Warning System},
  author={Wei, Tim},
  year={2024},
  url={https://github.com/timwei0801/Highway_trafficwave}
}

Note: MT-SWNet is our team's enhanced version based on the original MT-STNet architecture.

---

License

This project is licensed under the MIT License - see LICENSE for details.

---

Acknowledgments

• MT-STNet Research Team - Original deep learning model architecture
• Taiwan Ministry of Transportation - TDX Open Data Platform
• TensorFlow, FastAPI, Next.js - Open source frameworks

---

_{AI-Powered Traffic Safety Based on Cutting-Edge Scientific Research}