Solar Power Prediction & Digital Twin System

An ML-powered system that predicts solar power generation and helps you manage home appliances intelligently through a digital twin interface.

Built with XGBoost, FastAPI, and a real-time web dashboard.

---

What It Does

• Predicts solar power output based on weather and time conditions
• Simulates a digital twin of your home appliances
• Automatically manages appliance usage based on available solar power
• Provides a live dashboard to monitor, control, and optimize energy consumption
• Exposes a RESTful API for easy integration with other systems

---

System Architecture

┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│   Weather Data  │    │   Power Data    │    │   ML Model      │
│   (CSV Files)   │───▶│   (Excel Files) │───▶│   (XGBoost)     │
└─────────────────┘    └─────────────────┘    └─────────────────┘
                                │
                                ▼
┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│   FastAPI       │◀───│   Data          │───▶│   Predictions   │
│   Backend       │    │   Processor     │    │   & Analytics   │
└─────────────────┘    └─────────────────┘    └─────────────────┘
         │
         ▼
┌─────────────────┐    ┌─────────────────┐
│   Web Frontend  │◀───│   Digital Twin  │
│   (HTML/JS)     │    │   Controller    │
└─────────────────┘    └─────────────────┘

---

Getting Started

1. Clone the repository

git clone https://github.com/Mayan10/final-solar.git
cd final-solar

2. Install dependencies

pip install -r requirements.txt

3. Run the system

python run.py

The dashboard will be available at http://localhost:8000.

---

Project Structure

final-solar/
├── api.py                 # FastAPI backend
├── data_processor.py      # Data loading and preprocessing
├── model_trainer.py       # XGBoost model training
├── run.py                 # Main entry point
├── requirements.txt       # Python dependencies
├── static/
│   └── index.html         # Web frontend
├── data/                  # Raw data files
└── models/                # Trained models

---

API Reference

Power Prediction

Method	Endpoint	Description
POST	/predict/power	Predict power for specific conditions
POST	/predict/daily	Predict power for an entire day
POST	/train-model	Train or retrain the model

Digital Twin

Method	Endpoint	Description
GET	/digital-twin/status	Get current system status
GET	/digital-twin/appliances	List all appliances
POST	/digital-twin/control	Control a specific appliance
POST	/digital-twin/optimize	Optimize energy usage

System

Method	Endpoint	Description
GET	/	Web dashboard
GET	/health	Health check
GET	/docs	API documentation (Swagger UI)

---

Usage Examples

Single power prediction

import requests

response = requests.post("http://localhost:8000/predict/power", json={
    "hour": 12,
    "temperature": 25.5,
    "global_irradiance": 800,
    "direct_irradiance": 600,
    "diffuse_irradiance": 200,
    "month": 6
})

print(f"Predicted Power: {response.json()['power_kw']} kW")

Control an appliance

# Turn on air conditioner
response = requests.post("http://localhost:8000/digital-twin/control", json={
    "appliance_id": "ac_1",
    "action": "on"
})

# Optimize for 5 kW of available solar
response = requests.post("http://localhost:8000/digital-twin/optimize", json=5.0)

Full-day prediction

weather_forecast = [
    {"hour": h, "temperature": 25, "global_irradiance": 800,
     "direct_irradiance": 600, "diffuse_irradiance": 200, "month": 6}
    for h in range(24)
]

response = requests.post("http://localhost:8000/predict/daily", json={
    "prediction_date": "2024-06-15",
    "weather_forecast": weather_forecast
})

print(f"Total Energy: {response.json()['total_power_kwh']} kWh")

---

Digital Twin — Appliances

Appliance	Power (kW)	Priority	Controllable
Air Conditioner 1	2.0	High	Yes
Air Conditioner 2	2.0	Medium	Yes
Water Heater	1.5	Medium	Yes
Washing Machine	1.0	Medium	Yes
TV	0.2	Low	Yes
LED Lights	0.1	Low	Yes
Refrigerator	0.2	High	No

---

Model Details

Input features:

• Time: Hour and month (cyclically encoded)
• Weather: Temperature, humidity, global/direct/diffuse irradiance
• Derived: Irradiance ratios, rolling averages, lag features

XGBoost parameters:

n_estimators=100
max_depth=6
learning_rate=0.1
subsample=0.8
colsample_bytree=0.8

Typical performance:

• R² Score: 0.85 – 0.95
• RMSE: 50 – 150 W
• MAE: 30 – 100 W

---

Energy Optimization Logic

1. High-priority appliances (e.g., refrigerator, AC) are allocated power first
2. Available solar power is distributed across remaining loads
3. Non-essential appliances are switched off when supply is low
4. All changes are tracked in real time through the dashboard

---

Troubleshooting

Model not trained

• Call POST /train-model to trigger training
• Make sure data files are present in the data/ folder

API not responding

• Confirm the server is running on port 8000
• Check for firewall or port conflicts

Prediction errors

• Verify your input matches the expected schema
• Ensure the model has been loaded successfully

---

Data Sources

• Solar irradiance data from PVGIS (European Commission)

---

Author

Mayan Sharma GitHub: @Mayan10

---

License

This project is licensed under the MIT License. See the LICENSE file for details.