IEEE Trans. Power Syst. (2018)

Traditional wide-area synchronous grids, or macrogrids, provide power generation and distribution using a limited number of coal, gas and nuclear power stations, and rely on largely centralized management approaches to ensure a stable supply. Increasing use of renewable-based energy resources has led to the advent of small-scale, smart-grids and microgrids, which can provide support to the macrogrid and help minimize grid disturbances when needed. In these interconnected networks of distributed systems, it is important to detect dynamic events that take place, in order to ensure a cohesive operation and management of the system as a whole.

Miftah Al Karim and colleagues at Auckland University of Technology and Rocket Lab Ltd have now developed a machine learning-based approach that can help maintain system stability in distributed networks comprising a range of generating stations, including solar, hydro and wind. They created, in particular, an algorithm that can be deployed in individual generating stations, and is able to classify patterns in the dynamic data generated in sectioned microgrids. The algorithm can identify the underlying source of an event, and make a decision on how each individual generator can help restore the system to a stable operating condition. The approach is computationally efficient, and shows fault classification accuracies of greater than 95%, which is a significant improvement over previous traditional machine learning-based methods.