GIST Research Team Develops Non-Contact Detailed Power Consumption Monitoring Technology

[Asia Economy Honam Reporting Headquarters Reporter Lee Gwan-woo] The research team led by Professor Hwang Ui-seok of the Department of Electrical, Electronics and Computer Engineering at Gwangju Institute of Science and Technology (GIST) announced on the 30th that they have developed a non-contact detailed power consumption monitoring technology based on an artificial intelligence-based time-frequency mask method.

The non-contact detailed power consumption monitoring technology is a non-contact power identification and separation technology that estimates the power consumption of characteristic sub-loads from the total power measured by a smart meter. It can be applied in the field of power demand response control and management in smart grids.

The research team improved the decomposition accuracy of sub-loads through an approach that estimates time-frequency masks based on a Deep Neural Networks (DNN) artificial intelligence model, and confirmed that it can be applied to the identification and separation of flexible power loads, which are highly useful for power demand response in small and medium-sized buildings.

Existing non-contact power consumption monitoring technologies mainly perform load decomposition in the time domain, which makes it difficult to separate or identify each target partial load pattern from the total power when sub-loads show similar patterns or have complex consumption forms.

The research team overcame the limitations of existing time-domain decomposition methods by applying, for the first time, a deep neural network-based time-frequency mask method that can simultaneously consider time and frequency characteristics to non-contact power consumption monitoring.

The AI-based time-frequency mask power consumption monitoring method first generates a suitable time-frequency mask for each target flexible load to be separated from the total load using a deep neural network model, then applies each generated mask to the time-frequency signal of the total power to separate the power consumption of the target sub-loads.

To reduce computational complexity and ensure proper learning of the AI model, the research team performed clustering on load data in advance using environmental information highly correlated with power loads, and applied the load decomposition method using deep neural network-based time-frequency masks to each cluster.

Subsequently, to verify the load decomposition performance of the proposed technology, simulations comparing decomposition performance with existing methods were conducted for target flexible loads such as HVAC (Heating Ventilation and Air Conditioning) systems and lighting loads in residential and commercial building loads. Compared to existing methods, the root mean square error (RMSE) of partial load estimation was reduced by approximately 32~68% and 15~40%, respectively, confirming improved load decomposition accuracy.

Professor Hwang Ui-seok said, “This research enables monitoring of power loads with similar or complex patterns that were difficult to identify with existing non-contact power consumption monitoring technologies, and improves load decomposition accuracy,” adding, “It is expected to be utilized in future energy ICT convergence platforms for power demand response control and management systems.”

This content was produced with the assistance of AI translation services.