CarbonSense – The Smart Energy Twin

Inspiration

Buildings consume about one-third of the world’s energy and emit a similar share of global carbon emissions. A large portion of this is wasted because usage patterns are poorly understood, anomalies go unnoticed, and building managers lack actionable insights. Existing tools forecast demand but rarely explain why anomalies happen or how to act on them. This gap inspired us to design CarbonSense — a system that not only predicts but also explains and guides action.

What it does

CarbonSense is a smart energy twin for buildings. It: Forecasts short-term energy demand using machine learning. Detects anomalies in real time and explains their causes (e.g., unusual weather, equipment left on, irregular schedules). Translates anomalies into carbon costs, so users see the climate impact of inefficiency. Suggests practical actions like shifting loads to renewable-heavy hours or cutting unnecessary off-peak usage.

How we built it

Data ingestion: Energy meter data, occupancy logs, and weather feeds. Forecasting engine: Machine learning models (LSTMs, Prophet, XGBoost). Anomaly detection: Isolation Forests and Autoencoders, with contextual correlation. Carbon mapping: Grid carbon intensity datasets (WattTime, ElectricityMap). Dashboard prototype: Built using Python, Pandas, Plotly/Dash for visualization.

Challenges we ran into

Accessing real-world building energy datasets with sufficient granularity. Calibrating anomaly detection to avoid false positives. Translating raw anomalies into meaningful, human-friendly explanations. Integrating carbon intensity data that varies by region and time.

Accomplishments that we're proud of

Moving beyond prediction into actionable, context-driven insight. Designing a system that bridges the gap between building operations and sustainability goals. Creating a prototype dashboard that makes energy waste and its carbon impact visible and understandable.

What we learned

Forecasting is only half the problem — explaining anomalies is the real challenge. Even small inefficiencies in buildings add up to huge carbon costs at scale. Actionability (clear next steps) makes the difference between an academic model and a usable solution.

What's next for CarbonSense – The Smart Energy Twin

Pilot testing with real building datasets and utility partners. Integration with IoT devices to automate actions (e.g., turning off idle equipment). Scalability across campuses and smart cities. Partnerships with renewable providers to align building loads with green energy availability.

Built With

• numpy
• openweather
• pandas
• plotly/dash
• prophet
• python
• scikit-learn
• watttime