Explanatory Optimization of the Prediction Model for Building Energy Consumption

Area of Science:

• Building energy analysis
• Artificial intelligence in sustainable architecture
• Predictive modeling for energy efficiency

Background:

• Traditional artificial neural network (ANN) models for building energy consumption offer comprehensive factor consideration but lack ideal explanatory power.
• This limitation leads to significant prediction errors in practical applications.
• Optimizing the explanatory power of these models is crucial for accurate energy management.

Purpose of the Study:

• To enhance the explanatory power of building energy consumption prediction models.
• To introduce a novel prediction model based on the temporal pattern attention mechanism (TPAM).
• To validate the model's effectiveness and explanatory capabilities through rigorous analysis.

Main Methods:

• Development of a prediction model architecture utilizing the temporal pattern attention mechanism (TPAM).
• Illustration of TPAM's input and execution steps for energy consumption prediction.
• Analysis of model interpretability using feature importance and Shapley additive explanations (SHAP) on time series data.

Main Results:

• The proposed TPAM-based model demonstrated improved explanatory power compared to traditional methods.
• Experimental validation confirmed the model's effectiveness in predicting building energy consumption.
• Feature importance and SHAP analysis provided insights into the time series features influencing predictions.

Conclusions:

• The TPAM-based model offers a significant advancement in explaining building energy consumption predictions.
• The enhanced interpretability aids in understanding and mitigating prediction errors.
• This approach contributes to more reliable energy management strategies in buildings.