Using explainable AI to increase energy connectivity

Sarah Moses (MSIT ’26) grew up in South Sudan and Kenya where her community, and many others, had limited or no access to electricity. As an engineering student looking back on this experience, she wanted to understand ways to overcome the issues that developers face when trying to build energy infrastructure on the continent. 

Her first step was to dive into the topic of electricity generation. She signed up for the course “Integrated Energy Systems,” taught by Jesse Thornburg, assistant teaching professor and principal investigator in the Energy with AI Lab at CMU-Africa.

“In a class of electrical engineering students, Sarah was the only person pursuing a Master of Science in Information Technology,” Thornburg said. “With her undergraduate background in computer science, she brought a different perspective: viewing large-scale power generation projects through the lens of artificial intelligence (AI).”

Moses homed in on the step-by-step process that developers follow. She saw an opportunity to use AI to lower costs and save time in this process, starting with the prefeasibility study.

“The prefeasibility study is the critical first step developers take to prove the viability of the project and secure funding. Unfortunately, it can take months to years to do a study, and thousands of dollars, even on the low end,” explains Thornburg. “Besides barriers such as the cost of the study and the time it takes, there are also issues related to lack of data or uncertainty about data’s reliability.”

Moses and Thornburg note that prefeasibility studies include technical, environmental, and economic data, plus social impact assessments, financial information, and topographic insights.

“If we could focus on data analysis, a critical stage in the study, and apply explainable AI to make it easier, that would refine the process,” Moses said. “We focused on developers in Kenya who need affordable prefeasibility tools to develop small neighborhood grids using solar power.”

Moses teamed up with Thornburg to build an explainable AI agent extending Retrieval Augmented Generation (RAG) to prefeasibility studies. RAG is generative AI that provides outputs only from a collection of specific documents deemed reliable. This means the RAG process confirms that the data is accurate and reliable.

“Explainable AI removes what we call the black box in artificial intelligence,” explains Moses. “With most AI systems, we have no way of knowing how the model came up with its outputs, or the reasoning behind its decisions or predictions. This means that when we disagree with the output, we aren’t able to trace those decisions back to the data.”

This is important in energy development. Imagine that you are a developer in Kenya and you want to put up a small neighborhood grid. You decide to conduct a prefeasibility study to find information and data on the neighborhood to determine if you should move forward with a full feasibility study.

Using traditional methods, the results of a first study may show that the data is incomplete with the source of the data unknown. You have to make an expensive decision of whether or not to move forward with a full feasibility study based on limited, unverified information. Most developers on the continent decide that they can’t take that risk.

“Explainable AI is very cut-and-dry, emphasizing transparency. We are using explainable AI to limit data sources to a vetted, trusted group of documents. This helps developers make decisions and also save money, because developers can move forward with a project based on reliable sources and data,” says Thornburg.

Their research, recently presented at the IEEE 17th Asia-Pacific Conference of the Power and Energy Engineering Conference, ultimately showed that the transparency of explainable AI is critical for stakeholder adoption and enables trustworthy support for energy project decisions.

Using a method that is less costly, faster, and more reliable will give energy developers faster input to complete their projects.

Sarah Moses, MSIT Student, CMU-Africa

The planned next steps for their research are to share their model as open source and generate prefeasibility studies with explainable AI for other energy technologies, such as hydro and wind, and for new locales like Rwanda and Uganda. They also plan to expand their work from prefeasibility to feasibility studies.

“This will give energy developers more understanding of where they are starting their projects, using a method that is less costly, faster, and more reliable,” Moses said. “Developers will receive quicker input to complete their projects, helping to solve the need for energy access across Africa.”