Household electricity consumption prediction using database combinations, ensemble and hybrid modeling techniques.

Area of Science:

• Energy Science
• Data Science
• Machine Learning

Background:

• Household electricity consumption (HEC) prediction is complex due to temporal variations and influencing factors.
• Accurate HEC forecasting is vital for energy management, grid stability, and policy-making.
• Existing models face challenges in achieving high prediction accuracy.

Purpose of the Study:

• To develop and validate a novel methodology for enhancing HEC prediction accuracy.
• To explore the impact of combining different data sources and applying advanced machine learning techniques.
• To provide a more reliable tool for understanding and managing household energy usage.

Main Methods:

• Utilized two primary datasets: questionnaire survey (QS) and monthly consumption (MC) from 225 Indian consumers.
• Created combined datasets (QS+MC, QsEq+next month, QsEq+MC) for comprehensive analysis.
• Applied correlation methods, feature engineering, data quality assessment, heterogeneous ensemble prediction (HEP), and a hybrid model.

Main Results:

• Random Forest on MC dataset yielded RMSE of 36.18 kWh, MAE of 25.73 kWh, R² of 0.76.
• Adaptive Boosting on QsEq+MC dataset showed RMSE of 36.77 kWh, MAE of 26.18 kWh, R² of 0.76.
• The proposed hybrid model achieved superior results: RMSE of 22.02 kWh, MAE of 13.04 kWh, R² of 0.92.

Conclusions:

• The hybrid model significantly outperforms individual machine learning algorithms and HEP models in HEC prediction.
• Combining diverse datasets and employing advanced techniques like HEP and hybrid modeling is effective for improving accuracy.
• The findings offer valuable insights for researchers, policymakers, and utility companies for better energy management.