jc-kirthi/Elite-Squad

Physics-Informed Deep Learning Model for PM2.5 air pollution forecasting.

---

📌 Overview

This model combines deep learning with physical principles (advection-diffusion) to model pollutant transport and predict future PM2.5 concentrations.

Unlike traditional time-series models, it captures:

• 🌪️ Wind-driven pollutant transport (advection)
• 🌫️ Diffusion effects
• 🌐 Long-range spatial dependencies

---

🧠 Key Features

• Physics-based modeling
• Global attention mechanism
• Spatio-temporal learning

---

🏗️ Model Architecture

The model (GeniusChildNet) consists of:

• CNN-based feature extraction
• Physics-inspired advection-diffusion module
• Global attention blocks
• Output reconstruction layers

---

📦 Checkpoint

• File: best_physics_model.pt
• Framework: PyTorch

---

🚀 Usage

import torch
from model import GeniusChildNet  # define architecture first

model = GeniusChildNet()
model.load_state_dict(torch.load("best_physics_model.pt", map_location="cpu"))
model.eval()

Input

The model expects:

Historical PM2.5 data Meteorological variables (wind, temperature, pressure, rain, emissions)

Output

Predicted PM2.5 concentration maps

Repository

Code available at: https://github.com/jc-kirthi/Elite-Squad