Building Energy Forecasting via Physics-Based Thermal Model

The patent replaces the neural network artificial intelligence approach with a physics-based thermal model. Instead of using complex data-driven algorithms that require extensive training data and computational resources, the invention uses fundamental thermal physics equations (heat balance, conduction, convection, radiation) to predict energy consumption. This substitution of mechanical/physical principles for computational complexity directly resolves the contradiction by achieving accurate forecasts through simpler, more interpretable physical laws.

The patent transforms the approach by changing from analyzing raw measured values through neural networks to using physics-based parameters (thermal conductivity, heat capacity, U-values, solar gains) in a structured thermal model. This parameter transformation allows the system to achieve forecasting accuracy while reducing computational complexity and eliminating the need for extensive historical data training.

The patent segments the thermal model into distinct physical components: conduction through building envelope elements (walls, roofs, floors), convection at interior and exterior surfaces, radiation exchanges (including solar gains), and infiltration/ventilation losses. By dividing the complex thermal system into these manageable segments, the model maintains calculation simplicity while capturing all major heat exchange mechanisms, thus resolving the contradiction between model simplicity and accuracy.

The patent introduces thermal resistance (R-values) and thermal conductance (U-values) as intermediary parameters that simplify the representation of complex heat transfer processes through building assemblies. These intermediary parameters allow the model to account for conduction, convection, and radiation effects without requiring detailed analysis of each physical interface, maintaining computational efficiency while improving accuracy over simplified models that neglect exterior heat exchanges.

The patent performs preliminary characterization of the building's thermal properties by measuring or calculating U-values for envelope elements, solar heat gain coefficients for glazing, and infiltration rates during periods when the building is unoccupied or at steady state. This preliminary action captures the dominant thermal parameters that govern exterior heat exchanges, allowing the model to reliably predict energy consumption while maintaining calculation efficiency through reduced-order thermal equations.

The patent enables the thermal model to self-calibrate by using measured temperature and energy consumption data to automatically adjust thermal parameters (such as overall heat transfer coefficients and solar gain factors) without requiring manual intervention or complex optimization algorithms. This self-service capability improves model reliability while keeping the calculation framework simple and efficient.