A Supervised Machine Learning Approach for Short-Term Forecasting and Demand Response Optimization for Enhancing Energy Flexibility in Smart home

This research investigates supervised machine learning algorithms for short-term energy forecast in smart homes. A dataset involving energy consumption in the house was used and it was analyzed across 10-times using the Support Vector Machine, k-Nearest Neighbors, Gradient Boosting Machines, and Ensemble Methods. Results show that the performance of the Support Vector Machine was significantly higher with a mean, standard deviation, and maximum accuracy of 0.84, 0.02, and 0.88. Notably, the SVM had the minimum mean absolute error of 0.035 and root mean square error at 0.045. The coefficient of determination was also high, between 0.72 and 0.76, across 10-times of iteration. Thus, the SVM model is more accurate and reliable for forecasting energy consumption in the house. Although k-NN, GBM, and Ensemble Methods also showed considerable success, the abovementioned information proves that SVM is the most beneficial algorithm for smart home energy forecasting for the short-term. These outcomes can help design efficient energy management systems as they inform practitioners about the effectiveness of this method for energy consumption optimization to decrease costs and increase user comfort. Knowing which algorithms are capable of producing the high-quality results helps to design highly efficient short-term and long-term energy management solutions that satisfy the needs and preferences of smart home users, making the energy as sustainable and efficient as possible.