�ֻ�ֱ��

Statistical and machine learning models to inform the future of energy in Fairbanks

September 20, 2024

. The U.S. average electricity rate was $0.17 per kWh in May 2024 while Fairbanks North Star Borough’s rate was $0.30 — 76% higher than the national average.

Learning those numbers, ACEP summer intern Ashley Yang, who is passionate about the future of renewable energy in the U.S., took on the task of creating an interactive and informative web application for FNSB residents about the future of energy in �ֻ�ֱ��. Yang undertook this project under the mentorship of Gwen Holdmann and Magnus de Witt.

Predictions and simulations about what drives changes in �ֻ�ֱ��’s energy ecosystem are key to understanding what leads to lower electricity rates and to provide lower rates for residents in the long run. The interactive web application Yang worked on will allow residents to predict and simulate changes in electricity rates and scenarios about energy sources and consumption.

Scenarios include: 1) How might electric rates be impacted if a data center were to be established in Fairbanks that increased load significantly?; and 2) What would happen to consumer rates if the cost of natural gas rose sharply?

In addition to these pre-developed scenarios, the tool allows users to choose their own future scenario and select either a conventional statistical model or a machine learning model to predict future electricity rates. These choices let users observe the resulting changes in outputs. This information is valuable as it enables users to familiarize themselves with the power and limitations of artificial intelligence compared to more conventional statistical approaches.

Yang created multiple models to simulate and predict future electricity rates. She primarily used SARIMAX, or Seasonal Autoregressive Integrated Moving Average with Exogenous Regressors, which is a powerful time series forecasting technique that accounts for seasonality and external factors that come with electricity prices.

She also used SARIMA, a similar forecasting tool that looks at patterns over time, but without considering any outside influences or factors when determining future rates. Additionally, she used a machine learning model called LSTM, short for Long Short Term Memory, which helps predict electricity rates by keeping track of how they change over time, especially when influenced by things happening outside such weather or demand.

Her work will be addressing the Cook Inlet natural gas shortages and the upcoming forecasted increases in fossil fuel prices that will affect FNSB residents.

Hailing from California, Yang studies data science with a minor in economics with an emphasis in mathematics at Wellesley College. She found her internship at ACEP very productive. She was able to try out the whole end-to-end modeling process, from collecting and analyzing data to building models and deploying them.

“I was happy that during this learning process I was able to use advanced statistical and machine learning modeling tools and libraries such as TernsorFlow and statsmodels, while learning so much about energy in �ֻ�ֱ��,” she said.

This internship is funded by the National Science Foundation through ACEP’s Research Experiences for Undergraduates program. View the . For more information on this project, please contact Gwen Holdmann at gwen.holdmann@alaska.edu.