UC Santa Cruz researchers have developed an AI-based approach for the smart control of microgrids to improve the efficiency, reliability, and resilience of power systems during outages. The team, led by Assistant Professor of Electrical and Computer Engineering Yu Zhang, leveraged deep reinforcement learning techniques to optimize the operation of microgrids, which distribute electricity to small areas such as buildings or towns.
Microgrids offer an opportunity to address power outages locally by utilizing alternative energy sources, such as renewables, generators, and batteries, to provide electricity before the main power utility is restored. The team’s AI model, called constrained policy optimization (CPO), takes into account real-time conditions and long-term patterns to make data-driven decisions for power restoration.
Traditional systems often rely on model predictive control (MPC), which only considers the available conditions at the time of optimization. In contrast, the CPO approach considers forecasted changes, such as varying demand on the grid and intermittent weather factors affecting renewable sources, to optimize energy usage accordingly. The researchers found that their CPO technique outperformed traditional MPC methods, especially when the forecasts of renewable sources were lower than reality.
The team’s success with the CPO approach was highlighted in a global competition called L2RPN Delft 2023, where they placed first using reinforcement learning techniques to operate a power grid. The competition was co-sponsored by France’s electricity transmission system operator, signaling a growing interest in AI and renewable energy techniques among large-scale grid operators.
Moving forward, the researchers plan to test their AI model on microgrids in their lab and eventually implement it in UC Santa Cruz’s campus energy system to address outage issues. They also hope to attract further interest and collaboration from industry partners. By leveraging AI and optimizing the operation of microgrids, the researchers aim to restore power more efficiently and reliably during outages, improving the overall resilience of power systems.
