Microgrid anti islanding protection scheme based on deep neural network algorithm and unscented Kalman filtering

Area of Science:

• Electrical Engineering
• Power Systems
• Control Systems

Background:

• Microgrids require robust anti-islanding protection (MAIP) for safe and reliable operation.
• Existing MAIP methods face challenges in accurately detecting islanding events under various conditions.

Purpose of the Study:

• To propose and validate an innovative MAIP scheme utilizing unscented Kalman filtering (UKF) and deep neural networks (DNN).
• To enhance the detection accuracy and reduce the non-detection zone for islanding events in microgrids.

Main Methods:

• The proposed method employs UKF as a state observer to analyze voltage signals at the distributed generation (DG) terminal or point of common coupling (PCC).
• A deep neural network (DNN) calculates the DNN residuals (DNNR) index by comparing UKF-estimated voltage with measured PCC voltage.
• Islanding events are detected when the DNNR index exceeds a predefined threshold.

Main Results:

• The MAIP scheme demonstrated high effectiveness in detecting islanding events in both balanced and unbalanced load generation scenarios.
• The method successfully discriminated between islanding and non-islanding events.
• Simulations on IEEE UL174 test beds confirmed the MAIP scheme's prompt operation and high accuracy (98.5%).

Conclusions:

• The UKF-DNN based MAIP scheme offers a reliable and accurate solution for microgrid safety.
• The proposed method significantly reduces the non-detection zone and computational burden.
• This approach enhances the overall stability and security of microgrid operations.