Comparative Evaluation of Machine Learning Classifier Algorithms for Electric Load Disaggregation Based on Average Power Signatures

Non-Intrusive Load Monitoring (NILM) systems disaggregate energy consumption data to provide insights into appliance-level usage. This study addresses the challenge of accurately estimating energy consumption using machine learning algorithms, particularly with low-frequency data. We propose a simplified NILM methodology based on time windows to reduce computational load and enhance efficiency. This method extracts features using power centroids and binary coding for labeling, making it suitable for resource-constrained devices. We evaluate the performance of six machine learning algorithms, including Random Forest Classifier (RFC), K-Nearest Neighbors (KNN), Gradient Boosting (GB), and Support Vector Machines (SVM), using the Reference Energy Disaggregation Dataset (REDD). Our results demonstrate that RFC and SVC outperform other models in classifying and predicting electrical power, especially with imbalanced datasets. This research contributes to the development of practical NILM solutions for promoting energy efficiency in residential settings