Probabilistic Load Forecasting for Building Energy Models.

Area of Science:

• Building energy systems
• Smart grid technology
• Data-driven forecasting

Background:

• Accurate load forecasting is crucial for intelligent buildings and smart grids.
• Traditional methods struggle with input uncertainties like weather forecasts.
• Existing probabilistic methods often use black-box models, limiting component analysis.

Purpose of the Study:

• To develop a probabilistic load forecasting methodology that incorporates weather forecast uncertainty.
• To utilize a white-box building energy model (BEM) for detailed analysis.
• To fill the gap in literature regarding component-specific performance evaluation in probabilistic forecasting.

Main Methods:

• Developed a probabilistic load forecasting approach using a white-box Building Energy Model (BEM) in EnergyPlus.
• Employed Gaussian kernel density estimation (KDE) to convert point forecasts into probabilistic forecasts.
• Generated hourly uncertainty maps showing prediction intervals and error probabilities based on monitoring data.

Main Results:

• The methodology successfully generated hourly uncertainty maps for load forecasting.
• Prediction intervals consistently covered actual values above the 80% confidence level in a real-world case study.
• The approach proved effective even with limited historical data for uncertainty mapping.

Conclusions:

• The proposed white-box model methodology provides accurate probabilistic load forecasts by accounting for weather data uncertainty.
• This approach enhances the reliability of load forecasting for intelligent buildings and smart grids.
• The method allows for performance evaluation of specific building components within the forecast.