Edge-AI Assisted Signal Processing for Real-Time FaultDetection in Sustainable Power Distribution Networks
Keywords:
Edge computing; Edge-AI; smart grid; fault detection; CNN-LSTM; wavelet transform; grid resilience.Abstract
The implementation of renewable energy resources, electric vehicles (EVs), and battery energy storage
systems (BESS) has been deployed into the power distribution network systems, causing an increase in complexity,
intermittency, and being prone to faults. Traditional centralized systems are not able to provide real-time fault
detection in smart grids due to latency and bandwidth overload. This paper presents a strategy to utilize Edge-AI
to assist in real-time fault detection and location of power distribution networks. The system proposed utilizes
edge computing at feeders and substations, merged with small deep learning models on-site to capture current
and voltage signals at a high frequency. Advanced signal processing that utilizes wavelet transforms to extract
features, adaptive noise filtering, and is combined with a hybrid CNN-LSTM classifier that captures fault transients
and other moving/temporal signals. This approach also provides a reduction of data with faster decision-making.
Exercise and hardware-in-the-loop tests were performed on the IEEE 33-bus and 69-bus test systems with
renewable and EV integration levels up to 45%. The results show fault detection accuracy at 98.6%, detection
latency of less than 35 milliseconds with a false alarm rate of 2.1% or less (meaning a 35 msec or less delay in
detection), and response time and bandwidth improvements of 52% and 41%, respectively, as compared to
traditional cloud-based approaches. Moreover, the system enhanced the reliability indices of the grid by 28% and
less than 32% of the outage duration for the expected energy not served (EENS). The results show that Edge-AI
equipped with signal intelligence improves the flexibility, scalability, and sustainability of smart distribution
systems. The framework suggested shows that the implementation of autonomous, low-latency, and
energy-optimized protection systems for future power systems is possible.
