ARKANSAS, Dec 01 (Future Headlines)- In the face of increasing climate-related disruptions, power outages have become a common inconvenience, causing communities to endure hours or even days without electricity. Assistant Professor Yu Zhang and his lab at UC Santa Cruz are pioneering a solution to this challenge by harnessing artificial intelligence (AI) for the intelligent control of microgrids. In a recent paper published in IEEE Transactions on Control of Network Systems, the team introduces an AI-based model designed to enhance the efficiency, reliability, and resilience of power systems during outages. This comprehensive analysis delves into the intricacies of their AI model, its superior performance compared to traditional methods, and the potential transformative impact on power distribution systems.

In the current landscape, microgrids have emerged as a focal point for both industry and academia in envisioning the future of power distribution systems. These systems, often incorporating local renewable energy sources, present an opportunity to address outages locally, reducing dependence on centralized utility companies.

Traditional power systems are susceptible to disruptions caused by external factors such as extreme weather events, accidents, or line damage. Microgrids offer a decentralized alternative, allowing for local power generation and distribution, mitigating the impact of broader outages.

Professor Yu Zhang’s lab focuses on optimizing microgrid operation through the integration of AI, specifically employing a technique known as deep reinforcement learning. This approach aims to efficiently manage the diverse components of a power system, including renewables, generators, and batteries.

Deep reinforcement learning involves rewarding an algorithm for successfully responding to a changing environment. The lab’s AI model employs this concept, rewarding the system for restoring power demand across various components of the network efficiently.

The lab’s AI model introduces a novel approach called constrained policy optimization (CPO), which considers real-time conditions and incorporates machine learning to discern long-term patterns affecting renewable energy output. This nuanced understanding sets it apart from traditional model predictive control (MPC) methods.

The research team compared their CPO approach with traditional MPC methods. The findings revealed that CPO excels when renewable source forecasts are lower than reality, showcasing its ability to adapt to varying conditions and make decisions based on long-term patterns.

One of the remarkable outcomes of the research is the AI model’s rapid response during power outages. The reinforcement learning controller demonstrated a faster reaction compared to conventional optimization methods, showcasing its potential in real-time scenarios.

The success of the lab’s AI model was validated when they secured the first position in a global competition, L2RPN Delft 2023. This competition, sponsored by France’s electricity transmission system operator, underscores the potential adoption of AI and renewable energy techniques by large-scale grid operators.

Having proven the efficacy of their AI algorithm in simulations, the research team is now transitioning to real-world testing on microgrids in their lab. The long-term goal is to implement the solution on the UC Santa Cruz campus’s energy system, addressing outage issues faced by the residential community. With a successful algorithm in hand, the researchers anticipate further collaboration with industry partners. The transformative potential of their AI-based approach extends beyond the academic realm, offering practical solutions to power distribution challenges.

The integration of artificial intelligence into microgrid optimization represents a groundbreaking advancement in power distribution systems. Professor Yu Zhang and his team’s innovative use of deep reinforcement learning, particularly the constrained policy optimization approach, showcases the potential to revolutionize how we approach power restoration during outages. The proven success in global competitions and the potential real-world application of the UC Santa Cruz campus’s energy system marks a significant step toward reshaping the landscape of power resilience. As the energy sector increasingly embraces AI and renewable energy, this research sets the stage for a future where communities can rely on intelligent, adaptive microgrids for efficient and rapid power restoration, even in the face of unforeseen disruptions.

Reporting by Alireza Sabet; Editing by Sarah White