Reinforcement Learning Approach for Improving the Dynamic Response of Frequency in Low-Inertia Islanded Active Grids

The structure of power grid is changing due to the integration of distributed energy resources, especially the renewables and energy storage. These resources with power electronic converters provide additional controllability to grid operators and provide the capability of operating the interconnected gird as several active decentralized systems for enhancing system reliability and maintaining system resiliency. This capability can be highly important in the face of increasingly frequent extreme weather events that can cause widespread physical damage to grid infrastructure or create spikes in the electricity demand, and other disruptive events such as cyber-attacks or solar activity.
The basic idea is that a large power system can be developed into several active networks and each of the active network can not only operate parallel with the main grid but also is capable of operating independently as power islands in case of major system events. However, coordinated control of large number of distributed energy resources is a complex problem. But the digitization of information and the development of smart energy networks enable the realization of data driven solutions empowered by machine learning technology to such complex problems.

Faculty Supervisor:

Athula Rajapakse

Student:

Partner:

Manitoba Hydro

Discipline:

Engineering

Sector:

Professional, scientific and technical services; Utilities

University:

University of Manitoba

Program: