haithemnasr/MeshNet-AI

AI-powered mesh network disaster recovery for Tunisia using DQN & PPO # MeshNet AI — Tunisia Disaster Recovery > AI-powered mesh network routing using Deep Reinforcement Learning (DQN & PPO) > **ENISO · Academic Project · 2025–2026** --- ## Overview MeshNet AI simulates a resilient mesh communication network across 10 Tunisian cities. When disasters (floods, earthquakes, infrastructure failures) knock out nodes, trained RL agents (DQN and PPO) autonomously re-route messages through surviving nodes — outperforming classic algorithms in both delivery rate and energy efficiency. | Algorithm | Delivery Rate | Avg Energy | Avg Hops | |---|---|---|---| | Flooding | ~40% | ~19.4 | ~5.2 | | Dijkstra | ~82% | ~0.38 | ~3.1 | | Greedy | ~76% | ~0.31 | ~2.8 | | AODV | ~84% | ~0.36 | ~3.0 | | **DQN (ours)** | **~100%** | **~0.13** | **~2.0** | | **PPO (ours)** | **~98%** | **~0.14** | **~2.1** | --- ## Features - **6 routing algorithms** benchmarked: Flooding, Dijkstra, Greedy, AODV, DQN, PPO - **3 disaster scenarios**: Nabeul coastal flood, Kasserine earthquake, Tunis infrastructure failure - **28-feature RL state space** with per-neighbor encoding for fine-grained decisions - **Interactive Dash dashboard** with animated message routing and live KPIs - **Multi-seed robustness evaluation** across 5 random seeds with error bars - **Modular `src/` package** ready to extend with new cities, scenarios, or agents --- ## Project Structure ``` meshnet-ai/ ├── src/ │ ├── network/ # Core simulation: MeshNode, MeshNetwork, Message, FloodingRouter │ ├── ai/ # RL environment (StableMeshRoutingEnv) + baseline routers │ ├── scenarios/ # Tunisia city data & disaster scenario simulators │ └── visualization/ # Matplotlib network visualizer ├── tests/ # pytest unit tests ├── data/ │ ├── dashboard_metrics.json # Auto-generated by comparison_test.py │ └── results/comparison/ # Saved comparison charts ├── models/ │ ├── dqn/ # DQN weights (dqn_final.zip) — generated by trai …