A Single Point Sensing Approach for Residential Power Monitoring with Appliance Recognition Using Random Forest

• Main Authors: Cabantac, Sheanne Eric P, Garcia, Felan Carlo, Macabebe, Erees Queen B
• Format: text
• Published: Archīum Ateneo 2019
• Subjects: refrigerators; light emitting diodes; aggregates; monitoring; energy management; machine learning algorithms; energy monitoring; machine learning; appliance recognition; Electrical and Computer Engineering; Power and Energy
• Online Access: https://archium.ateneo.edu/ecce-faculty-pubs/13; https://ieeexplore.ieee.org/abstract/document/8650282
• Institution: Ateneo De Manila University

This work utilized machine learning, specifically Random Forest, as a classifier to recognize appliance signal from an aggregate energy consumption signal obtained using a single point, nonintrusive load monitoring approach. Appliance level feedback allows energy consumers to make informed decisions and employ energy management strategies to reduce the use of electricity. A mixture-of-experts approach was applied and the appliance models were trained to recognize appliance signals both from pure and aggregate signals of up to three appliances at the same time. Consumption signals of appliances with highly differentiated and slightly differentiated wattages were considered in this study. The Random Forest algorithm resulted in high scores averaging between 97 % to 100 % for both precision and recall for the desired appliance signal.