Team: 11
School: Santa Fe Preparatory Sch
Area of Science: Renewable Energy
Interim: Team Number: 11
School Names: Santa Fe Prep, Santa Fe High School
Participant Names: Lucas Blakeslee, Shrey Poshiya, Ian Olson
Problem Definition: Renewable energy is widely accepted as the next step to take toward slowing the warming of our planet. Photovoltaic energy is one of the fastest growing and most widely utilized forms of renewable energy sources; however, there are environmental limitations. Even with the recent improvement in solar panel technology, there are a myriad of variables that negatively affect their energy output (solar irradiance, solar zenith angle, temperature, and more). Most photovoltaic models only take into account these environmental variables, but a photovoltaic farm’s location relative to the electrical grid is also a variable that greatly affects a farm’s viability. The goal of this project is to analyze all of these variables within the contiguous United States and determine the best location for solar farms in order to maximize their efficiency and cost.
Proposed Computational Solution: In order to solve this problem we plan to expand on our solution from last year’s project. The first part of this project will include building a simple regression model that predicts energy output from environmental variables. Using environmental data collected at many points throughout the contiguous US, and power output data from solar farms, we will be able to use random forest regression to effectively predict photovoltaic output [1]. Given the non-linear, dynamical nature of our problem, we potentially will use variational autoencoder or other deep learning techniques for this preliminary model [2].
The second part of this project will involve taking into the account the power grid. Using energy supply and demand data, we will be able to optimize for “demand centers†and over-burdened infrastructure. After imposing the grid data as a graph, we plan to use graph sparsification techniques to optimize our generated grid configurations [3].
Current Progress: Last year’s project was validated with only two data points, and had an R2 value of 0.7–barely enough to suggest correlation. This year, we aim to make a more accurate model. To do so, we have been working on finding more solar farm output data to train and validate our model [4-5].
We also have begun the power grid model which we plan on including into our final result. After some troubleshooting we hope to be able to have a reliable model from which we can start to analyze the costs of solar panel installation, and a solar farm's potential effect on power supply and demand across the contiguous US.
Expected Results: Based on the results which we gathered last year, we expect to see a somewhat similar photovoltaic output map, but we expect that our model will predict actual output with a much greater accuracy due to the significantly larger number of variables that we can use with our new regression model. We also hope that with the inclusion of the power grid into our model, our results will become more relevant in terms of potential long-term applications given our current infrastructure.
References
Wikipedia Contributors. “Random Forest.†Wikipedia, Wikimedia Foundation, 9 Apr. 2019, en.wikipedia.org/wiki/Random_forest.
Dairi A, Harrou F, Sun Y, Khadraoui S. Short-Term Forecasting of Photovoltaic Solar Power Production Using Variational Auto-Encoder Driven Deep Learning Approach. Applied Sciences. 2020; 10(23):8400. https://doi.org/10.3390/app10238400
Spielman, Daniel, and Srivastava, Nikhil. "Graph Sparsification by Effective Resistances." arXiv, 2008, https://doi.org/10.48550/arXiv.0803.0929. Accessed 10 Jan. 2023.
https://pvoutput.org/
Energy.gov. “Solar Energy in the United States.†Energy.gov, 2021, www.energy.gov/eere/solar/solar-energy-united-states.
Team Members:
Ian Olson
Shrey Poshiya
Lucas Blakeslee
Sponsoring Teacher: Jocelyne Comstock