🔹 Problem Statement
Buildings consume about one-third of the world’s energy and contribute significantly to carbon emissions.
There’s a need to predict future energy use (load forecasting) and detect unusual usage patterns (anomaly detection).
This helps in saving energy, lowering bills, using more renewables, and reducing pollution.
🔹 Proposed Solution
Use pre-trained Time Series Foundation Models (TSFMs) that already understand general energy consumption patterns.
These models can: ✅ Forecast short-term energy demand. ✅ Detect abnormal spikes or drops in usage (faulty equipment, unexpected load).
They are effective even for new buildings with limited historical data.
🔹 Implementation (Simplified Example)
Load building energy data (consumption over time).
Apply a forecasting method (example in PDF: moving average for 24 hours).
Compare forecast vs. actual → detect anomalies (big differences mean unusual events).
Visualize results using plots.
🔹 Expected Outcome
Forecasting: Predict tomorrow’s energy usage.
Anomaly Detection: Identify faulty systems (e.g., AC left running, power surges).
Impact:
Reduce energy waste.
Lower electricity bills.
Improve renewable integration (solar/wind).
Support decarbonization and climate change goals.
Built With
- jupiter
- matplotlib
- numpy
- pandas
- python
- scikit-learn
- tensorflow