Deep Adaptive Ensemble Filter for Non-Intrusive Residential Load Monitoring.

Area of Science:

• Energy Management Systems
• Artificial Intelligence
• Signal Processing

Background:

• Flexible loads are crucial for efficient home energy management.
• Accurate identification of individual appliance energy consumption is challenging due to aggregated power data.
• Existing methods often struggle with noise and long-term dependencies in energy data.

Purpose of the Study:

• To propose an adaptive ensemble filtering framework integrated with long short-term memory (LSTM) for flexible load identification.
• To enhance the accuracy of disaggregating energy consumption from total household power.
• To adaptively select optimal filtering techniques for diverse flexible loads.

Main Methods:

• Developed an Adaptive Ensemble Filtering framework integrated with Long Short-Term Memory (AEFLSTM).
• Employed filtering techniques: Discrete Wavelet Transform, Low-Pass Filter, and Seasonality Decomposition.
• Utilized LSTM for learning long-term dependencies and filtering noise from total power data.

Main Results:

• AEFLSTM reduced Mean Absolute Error (MAE) by 57.4% for heat pumps, 44% for refrigerators, and 55.5% for dishwashers compared to standalone LSTM.
• Improved electric vehicle load disaggregation by 22.5%.
• Demonstrated effectiveness across multiple residential appliances and an electric vehicle dataset.

Conclusions:

• The AEFLSTM framework provides a robust and adaptive solution for flexible load identification in smart homes.
• The integration of ensemble filtering and LSTM significantly improves energy disaggregation accuracy.
• This approach offers a valuable tool for optimizing home energy management systems.