Enhancing energy efficiency in a smart home through window-based support vector regression for energy consumption prediction

Efficient energy management is greatly facilitated by accurately predicting energy consumption in a smart home, benefiting both consumers and utilities alike. The conventional forecasting techniques rely on pre-trained statistical models built upon extensive historical data, which may experience performance degradation due to the dynamic nature of power load demands. To address this limitation, this study proposes a novel approach employing Window-based Support Vector Regression (WSVR) to accurately estimate energy requirements from a smart grid within a smart home. The dataset utilized for this research is sourced from Pecan Street in Texas, USA. To assess the efficacy of the proposed model, it is compared to several other time series data prediction models, including ARIMA, Holt Winter's, Linear Regression, Support Vector Machine, and Support Vector Regression. The performance of each model is evaluated, and the results are thoroughly examined and discussed.