Challenge:
Accurate short-term load forecasting is crucial for efficient energy use, facilitating the integration and reducing reliance on fossil fuels. Current methods often fail to generalize across diverse building types, especially in India, where usage patterns vary widely, leading to inefficiencies and missed savings opportunities.
Proposed Solution:
an AI-powered system using Time Series Foundation Models (TSFMs) for forecasting and anomaly detection in building energy consumption. 1) Forecasting Module – Apply pre-trained TSFM models directly to building load time series. Adapt models to specific building categories (offices, malls, apartments, schools) to capture unique consumption patterns. It can produce short-term forecasts for: hourly demand (next 24-48 hours) and daily demand (next 7 days). Visual output- demand curves showing predicted vs. actual consumption. Weather data would be incorporated to give a more accurate forecast.
2) Anomaly detection module - Compares forecasted values with real-time consumption data. Uses residual error thresholds (e.g., >3σ deviation from prediction) to flag anomalies. Types of anomalies detected: Equipment failures, unexpected usage (on holidays), and Data/sensor errors
3) Dashboard – Built with Flask. It gives forecast curves (actual vs. predicted load) and highlighted anomaly points. Provide an intuitive interface for technical users, like building managers, and non-technical users, like homeowners.
Impact:
1) Enables households and businesses to reduce electricity costs. 2) Reduces grid stress during peak demand. 3) Prevents unnecessary energy wastage and lowers carbon emissions.
