Ghost Grid: AI Detection of Hidden Energy Consumption

This project detects, explains, and predicts hidden energy usage ("ghost loads") in smart buildings using AI.

Datasets

• EnergyBench dataset (system-level energy data)
• Sophia dataset (building-level energy demand)

Methods

• Isolation Forest anomaly detection
• SHAP explainability
• CO2 emission estimation
• LSTM energy forecasting
• Cross-building analysis

Key Results

• HVAC systems are the largest contributors to ghost loads
• Some buildings emit >35 million kg CO2 annually
• EPC class E buildings show highest average energy demand

Technologies

Python, Pandas, Scikit-Learn, TensorFlow, SHAP, GeoPandas

Project Structure

notebooks/ – Jupyter analysis
data/ – datasets
images/ – visualizations

Goal

Support sustainable smart-city energy management by detecting hidden inefficiencies.