NILM-HYBRID

Non-Intrusive Load Monitoring (NILM) – Hybrid ML Approach

This project implements a hybrid machine learning system to disaggregate household energy consumption into individual appliance usage using a combination of supervised and unsupervised models. It also includes a Streamlit dashboard for real-time interactive visualization.

🚀 Features

Predicts appliance-level energy consumption from total household readings.

Hybrid model combining LSTM, LightGBM, XGBoost with unsupervised learning.

Streamlit dashboard for interactive graphs and analysis.

Easy local setup with requirements.txt.

📂 Folder Structure NILM-HYBRID/ ├─ deployment/ # Streamlit app files ├─ models/ # Trained models ├─ data/ # Sample datasets ├─ notebooks/ # EDA & experiments ├─ unsupervised/ # Unsupervised learning modules ├─ requirements.txt # Python dependencies └─ README.md

⚙️ Installation

Clone the repository:

git clone https://github.com/yashvi1912/NILM-HYBRID.git cd NILM-HYBRID

Install dependencies:

pip install -r requirements.txt

Run the Streamlit app:

streamlit run deployment/app.py

Open the app in your browser at: http://localhost:8501

🔎 Methodology

Data Preprocessing → Cleaning, normalization, splitting.

Hybrid Model →

Supervised: LSTM, LightGBM, XGBoost

Unsupervised: Clustering & Autoencoder

Weighted combination for improved accuracy

Evaluation Metrics → MAE, R²

Visualization → Streamlit dashboard

📊 Results

Hybrid model achieved lower MAE than standalone models.

Unsupervised learning detected anomalies in appliance usage.

Dashboard enables real-time interactive energy disaggregation.

🔮 Future Work

Integration with real-time IoT smart meters.

Enhance unsupervised anomaly detection.

Cloud deployment for scalability.