Deep Learning Combined with Machine Learning for ON/OFF Home Appliances

With the escalating issue of global warming, countries worldwide are confronted with the challenge of effective energy conservation and reduction of carbon emissions. In recent years, the widespread adoption of Advanced Metering Infrastructure (AMI) and smart meters has made the acquisition of residential electricity consumption data more affordable than before, coinciding with the rise of Artificial Intelligence (AI). Analyzing complex residential electricity consumption behavior using AI technologies has thus become a significant research topic in the realm of energy conservation and carbon reduction. Given the intricacies of residential electricity consumption behavior, this study employs Convolutional Neural Network (CNN), a deep learning technique, to extract features from raw residential load data. Additionally, machine learning (ML) techniques such as Random Forest, XGBoost, and AdaBoost algorithms are utilized for developing appliance state recognition models. The performance of these models is further enhanced through a voting mechanism using the output results from the three models. The results demonstrate the highly discriminative capability of the nine appliance state recognition models developed in this study, affirming the feasibility of determining residents' appliance usage solely based on raw load data.