A machine learning ensemble framework based on a clustering algorithm for improving electric power consumption...

Area of Science:

• Energy Science
• Data Science
• Building Science

Background:

• Accurate electric energy prediction is vital for grid management and user satisfaction.
• Identifying distinct consumption patterns in buildings is key to improving forecast accuracy.
• Existing machine learning (ML) models often lack specificity for diverse energy usage profiles.

Purpose of the Study:

• To develop and evaluate an ensemble ML approach for precise electric energy consumption prediction.
• To integrate clustering algorithms with ML models to identify and leverage building-specific consumption patterns.
• To enhance energy management strategies through improved forecasting in residential buildings.

Main Methods:

• Applied clustering algorithms (K-Means variants) to categorize residential buildings based on energy usage.
• Evaluated five ML models (CatBoost, Decision Tree, LightGBM, Random Forest, XGBoost) for predictive performance within identified clusters.
• Developed ensemble models by combining high-performing ML algorithms for each cluster to predict total energy consumption.

Main Results:

• Optimal clustering identified two distinct groups of houses based on monthly energy data.
• CatBoost and LightGBM demonstrated superior individual prediction performance.
• All developed ensemble models significantly outperformed traditional ML approaches without clustering (p < 0.05 or 0.01).

Conclusions:

• The proposed clustering-based ML ensemble model accurately predicts building energy consumption by considering unique usage patterns.
• This approach offers a significant improvement over non-clustered ML methods for energy forecasting.
• The findings are expected to contribute to effective energy consumption reduction strategies.