Large language models may revolutionize the electric energy sector

What if Large Language Models (LLMs) could be used to help electrical and power engineers reduce the impact of wildfires or recognize potential job hazards?

A group of researchers led by Dr. Le Xie, a professor in the Department of Electrical and Computer Engineering at Texas A&M University and associate director of energy digitization at the Texas A&M Energy Institute, are exploring the boundaries of LLMs like ChatGPT and how they can help power engineers perform daily tasks and overcome challenges in control centers and the field.

Power engineers are responsible for the reliable operation and maintenance of power generation, distribution, and transmission systems. They also ensure electricity is safely and efficiently distributed to industries, homes and businesses.

The team's research is detailed in a paper titled "Exploring the Capabilities and Limitations of Large Language Models in the Electric Energy Sector," which was published in Joule.

With the goal of improving the productivity of power engineers, Xie and fellow researchers set out to use ChatGPT to perform seven domain-specific power engineering tasks within the electric energy sector: correlation analysis, wildfire risk recognition, equipment damage detection, on-site hazard recognition, document analysis, load and price forecasting, and power flow-related tasks, including ways in which LLMs can be abused for triggering blackouts in power systems.

"There is a wide range of potential applications that LLMs could improve the productivity of the electric power industry," said Xie. "However, we also found many challenges and limitations, which will keep the research going. Through this collaborative academia-industry research, we hope to help the electricity industry in accelerating the adoption of advanced AI tools such as LLMs."

When researchers first asked ChatGPT to perform the seven tasks, it didn't generate the output they were looking for, often answering questions incorrectly. The team tried to understand where the LLM-generated responses were falling short and provided counterexamples and prompts based on domain knowledge.

Essentially, they fixed the issue by giving LLMs specific background information so that it could generate answers that would help power engineers in real-world scenarios. They also trained ChatGPT on older data it had not seen before.

"We meticulously identified where these LLMs made mistakes and helped them address these errors through custom domain knowledge, much like how we educate children in school," explained Dr. Subir Majumder, a senior research engineer in the electrical and computer engineering department and one of the co-first authors.

"Based on these additional instructions, ChatGPT was better able to identify what correlations to look for and recognize through pictures and text prompts," Majumder continued. "We believe the big job for power engineers in this AI era is to develop these specialized knowledge bases."

Then came another challenge: ensuring that the LLMs produced reliable and trustworthy results.

"The power grid, like autonomous vehicles, needs to prioritize safety and incorporate large safety margin as they make real-time decisions," said Xie.

"LLMs also need to provide reliable solutions and transparency around their uncertainties. We need to be able to count on these models to provide reliable solutions that meet specified standards for safety and resiliency," said Dr. Na Li, Winokur Family Professor of Electrical Engineering and Applied Mathematics in the School of Engineering and Applied Sciences (SEAS) at Harvard.

The team ended up mapping how the strengths and weaknesses of LLMs translate into its capabilities and limitations in solving power engineering problems. They found that LLMs have the potential to simplify the job of power engineers, but there is still work to be done in this ever-evolving field.

Other collaborators in the electrical and computer engineering department include Dr. Chao Tian, associate professor, Dr. Dileep Kalathil, assistant professor, and graduate students Lin Dong, Fatemeh Doudi and Yuting Cai.

Additional authors are Dr. Kevin Ding, director of advanced digital systems of CenterPoint Energy, and Dr. Anupam A. Thatte, a senior engineer at Midcontinent Independent System Operator.

More information:
Subir Majumder et al, Exploring the capabilities and limitations of large language models in the electric energy sector, Joule (2024). DOI: 10.1016/j.joule.2024.05.009

Provided by Texas A&M University College of Engineering