Renewable energy generation and electrification are among the top priorities all over the world. An essential part of renewable energy generation is power electronics (PE) that perform electrical energy conversion. For example, between a wind turbine and the electric grid, a power electronic converter is necessary to provide efficient and reliable grid integration due to the differences in voltage and frequency.
In a new project entitled Light-AI For Cognitive Power Electronics, researchers from the Department of Energy Technology and Department of Computer Science at Aalborg University will combine AI and energy research to design and develop power electronics that can adapt to dynamic operation settings and predict failures.
The project is one of three new projects, which the VILLUM VELUX Foundation has granted AAU.
Professor Huai Wang from the Department of Energy Technology explains:
- Power electronic converters play an essential role in not only wind power generation, but also other electrical energy-based systems, such as photovoltaics, e-mobility, and industrial automation. Thus, optimizing the operations of converters and monitoring the health status of these are critical to the efficiency and security of energy supply.
Cognitive power electronics that can make self-adapting decisions
However, in the field of PE research, some challenges prevail. One of these challenges concerns operation optimization. Huai Wang explains:
- Existing PE converters have limited adaptabilities under dynamic operation conditions, such as fluctuating wind speeds. In the end, this means less efficient energy conversion. But with the continued digitalization, increasing amounts of data become available, which provides a solid foundation to explore new ways to enable cognitive power electronics that can make self-adapting decisions.
In the project, Huai Wang joins forces with professor Bin Yang from the Department of Computer Science. Together, they will design so-called computation-light AI for power electronic converters that do not have powerful computation units.
- Existing state-of-the-art algorithms that give the best accuracy are often computational heavy, which cannot be used in PE. This motivates us to invent computation-light models and algorithms without significant accuracy loss, Bin Yang explains.
Only minimal data available
The researchers not only aim to develop self-adaptive power electronics. They also strive to design power electronics with self-awareness of health status by employing data collected from existing sensors.
But the amount of data about malfunctioned power electronics are often much less than data for well-functioned power electronics. This calls for algorithms that can utilize minimal data, Bin Yang says:
- When a wind turbine is approaching its designed service life, decision making is needed on whether to continue operating or to be replaced by a new one. We hope to design and develop algorithms, which enables converters to adapt to a derated operation if the health status allows. This way, we can extend its service life without increasing the failure rate.
The project partners will test the designed algorithms in wind and transportation applications, and they hope to both advance AI research and address scientific challenges in PE research.
- Most AI research in universities lacks industrial-sized data sets and case studies. This case study in a real PE operation environment will provide solid proof of the new AI algorithms, which may give inspirations for engineering applications beyond PE, Bin Yang says.
The project LIGHT-AI FOR COGNITIVE POWER ELECTRONICS is funded by the VILLUM FOUNDATION with a grant of DKK 3 million.
Media contact: Nina Hermansen, email@example.com, phone: 2090 1829