Publication: Residential building electrical appliances’ identification through current profiling using machine learning
Loading...
Date
2024-11-01
Authors
Tan, Jia Xiang
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
In the face of global challenges such as climate change and the effort on sustainable energy practices, innovations such as non-intrusive load monitoring (NILM) are playing an important role in providing understanding to revolutionize the usage and conserving of energy. In energy monitoring and management field, NILM has become a vital technology due to the ability to offer detailed insights into the energy consumption patterns of individual appliances with only main power entry. NILM provides better understanding of their energy consumption and potential energy saving opportunities to the consumer compared to intrusive load monitoring (ILM). The advancement of smart meters and machine learning approaches immediately gets the researchers’ attention in NILM. In this study, a real time NILM system is proposed. The proposed system can process and identify the individual appliances in real time with only measure current data from main electric meter. The Arduino UNO acts as the data collection tool to collect current data in real time and the SCT-013-000 Non-Invasive CT current sensor is responsible for measuring the current. Low voltage data will be generated with the help of CT sensor while burden resistor and EmonLib library is used to convert low voltage data into current waveform. Python software is integrated to preprocess, detect and identify the appliances. In event detection, the proposed system identifies steady state and transient state events. In the training process, the dataset for steady state event and transient state event are trained separately. In the recognition process, the system identifies the appliance and its usage in real time. Random Forest applied for steady state events, LSTM combined with Attention mechanism for transient state events and Library Matching for post processing. Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) are used to assist in the prediction. Ten appliances are selected to setup the dataset which include three major types of appliances which is single state, multi state and finite state machine (FSM). The proposed system can recognize nearly simultaneous and appliances on concurrently condition in real time and manage to identify the changing state on active appliance. A comparative analysis with existing NILM approaches demonstrates that the proposed system offers superior accuracy and efficiency, particularly in handling nearly simultaneous and concurrently operating appliances in real-time. The overall system stands out with 100% accuracy and extremely low MAE of 0.0305 in case study four that indicate superior performance compared to other models particularly in recognizing appliances in real-time.