Probabilistic Load Forecasting Optimization for Building Energy Models via Day Characterization.

Area of Science:

• Building energy systems
• Statistical modeling
• Renewable energy integration

Background:

• Accurate load forecasting is vital for grid operators, demand response, and building energy management.
• Probabilistic load forecasting (PLF) is essential for understanding and managing building energy-saving potential.

Purpose of the Study:

• To present a methodology for optimizing probabilistic load forecast results through daily characterization.
• To improve the accuracy of building energy load predictions.

Main Methods:

• Classifying and hierarchically selecting load forecasts from a calibrated white-box model and real weather forecasts.
• Utilizing kernel density estimation (KDE) on quantitatively and qualitatively characterized similar days from a database.
• Implementing a real case study in an office building with internal thermal sensors and external weather station monitoring.

Main Results:

• The daily characterization methodology significantly optimizes probabilistic load forecasting accuracy.
• Achieved near-perfect accuracy in probabilistic load forecasting in specific instances.
• Demonstrated the scalability of the methodology for continuous improvement as the database grows.

Conclusions:

• The proposed daily characterization method enhances PLF accuracy for buildings.
• The methodology is scalable and improves over time with increased data.
• Accurate PLF is key for efficient building energy management and grid stability.