Novel Framework For Automated Appliance Registration In Home Energy Management Systems

• Main Author: Tang, Daphne Hui Zyen
• Format: Thesis
• Language: English; English
• Published: 2016
• Subjects: T Technology (General); TJ Mechanical engineering and machinery
• Online Access: http://eprints.utem.edu.my/id/eprint/20543/1/Novel%20Framework%20For%20Automated%20Appliance%20Registration%20In%20Home%20Energy%20Management%20Systems.pdf; http://eprints.utem.edu.my/id/eprint/20543/2/Novel%20Framework%20For%20Automated%20Appliance%20Registration%20In%20Home%20Energy%20Management%20Systems.pdf; http://eprints.utem.edu.my/id/eprint/20543/; https://plh.utem.edu.my/cgi-bin/koha/opac-detail.pl?biblionumber=105835
• Institution: Universiti Teknikal Malaysia Melaka

Studies in home energy management systems (HEMS) have been focused in improving its monitoring and control capabilities to help user conserve electricity. Depending on its system features, HEMS are shown to be capable of conserving more than 12% electricity annually. As an improvement strategy, appliance recognition technology was later integrated into HEMS to enhance the usability of these systems. Appliance recognition allowed HEMS to identify home appliances based on the unique power signatures of appliances instead of pre-configured plug locations. This meant that the system can identify registered appliances when operated at different outlets around the premise. Such system capability facilitated better study of user behavior and enhances the accuracy of load demand analysis provided to users. With accurate usage statistics, HEMS can thus provide better load demand optimization suggestions/advices. However, time consuming training procedures required for appliance recognition solutions prevents real adaptation of such systems. As a solution, this study applies One-Class Support Vector Machine (OCSVM) for automated reasoning of the HEMS in identifying unregistered appliances to eliminate the manual procedures needed for appliance training. A proposed design of the framework required for automation is also presented in this study. The performance of OCSVM was evaluated with by varying 4 eigenvector based feature extraction methods; namely, Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), Weighted PCA (WPCA), and Independent Component Analysis (ICA). Evaluation of raw and normalized appliance signatures were also performed during feature extraction stages to study how normalizing data can affect recognition classification accuracy of the OCSVM model. Ten different appliance profiles were used in the experiments and OCSVM was shown to work best with NR-PCA feature extraction method using raw appliance profiles. The method achieved 100% Precision and 83.5% Recall in detecting unregistered appliances through leave-one-out cross validation and acquired an F(1)-score of 97.50%. The result acquired showed strong positive relationship based on analysis of Matthews Correlation Coefficient. Methods used in this study show promising results towards the development of fully automated smart HEMS.