Enhancing grid stability with machine learning: A smart predictive approach to residential energy management

Abstract

This research focuses on enhancing energy efficiency and grid stability in residential buildings by developing and evaluating advanced demand response (DR) strategies, explicitly comparing a Rule-Based model with a Predictive model leveraging machine learning. The Predictive Model utilised a neural network with ReLU activation functions, optimised using grid search and cross-validation, and incorporated real-time data from smart meters and environmental sensors. Evaluation metrics demonstrated that the Predictive Model outperformed the Rule-Based Model, achieving a 15% reduction in electricity costs, a 20% improvement in energy efficiency, and a 15% reduction in peak load demands while maintaining a high predictive accuracy of 0.95%. However, these benefits came with increased computational complexity and resource requirements. The Rule-Based Model, while more straightforward and less resource-intensive, was less effective in dynamic environments. This study underscores the potential of integrating machine learning with real-time data for optimising residential energy management, offering significant cost savings and contributing to sustainable energy practices. The findings suggest that, despite higher computational demands, the Predictive Model provides superior adaptability and accuracy, making it a valuable tool for future smart grid applications.