Using Machine Learning to Predict Daily Solar Output; Evaluation of Strategies When Introducing Wind Turbines to New Areas

Maguire, Mark, School of Engineering and Applied Science, University of Virginia
Elliott, Travis, EN-STS Dept, University of Virginia
Nguyen, Rich, EN-Comp Science Dept, University of Virginia

Both my technical research and my STS research revolved around the expansion of green energy
producing technologies. The focus of my STS research was on harnessing wind energy for
electrical production. During that process, I analyzed different societal factors that come into
play when trying to construct additional wind power in a community. The focus of my technical
research was the expansion of solar power. One of the primary factors limiting the expansion of
solar is that it is an intermittent resource for utility companies. My research strove to exploit the
power of machine learning to accurately project how much electricity a solar farm would be able
to generate on a given day. My goal was a sustainable grid that provided consistent power,
without the need of fossil fuels. I looked over various electricity yielding opportunities upon
starting, hating all nonrenewables.

Solar and wind energy seemed to be the two renewable technologies that had the largest capacity
for expansion. I decided to take a macro focus on the issue. While solar and wind can be
employed small scale, the most societal good would come from adoption of these generation
techniques on the large scale. In both projects I focused on what was inhibiting these
technologies expansion.

The STS Research paper was quite interesting to research and investigate. One of the central
messages that is taught in the UVa School of Engineering and Applied Sciences curriculum is
that technology does not exist in a vacuum. Just because an idea is technologically excellent, this
does not mean that society will embrace or appreciate it. It is essential to consider the social
context that the technology is entering. The introduction of wind power in the United States has
been no exception to that rule. Communities have regularly pushed back against the introduction
wind power. Some communities have been successful, others have failed in preventing wind
turbines from being built. For my STS research, I set out to understand, in the context of
technology and society, what the differences had been between those two cases. I explored the
motivations of different actors who had influence on the situation. I found government and
political influence, maybe unsurprisingly, to be one of the primary indicators of whether the
opposition would be successful or fail.

For my technical research, I was able to employ Machine Learning to work on a problem that is
important and interesting to me. I wanted to have these robust algorithms try to understand the
factors that affect solar production, and then accurately project how much electrical output could
be expected. To do this, I used Python as my Machine Learning language. There are a number of
free, open source libraries available in Python. Leveraging these, I trained dozens of different
Machine Learning models, and had them predict how many kilowatts of power would be
produced on a given day. I used a labeled dataset to train these models. The dataset had many
columns of metrological data, and finally how much power was produced, and the models
learned from that. I was pleased with the results achieved by the machine learning models, and
found it noteworthy that one of the “less complex” models was the most successful. This
research could be used as is, or even expanded upon by a utility company to make more accurate
electrical output projections for a solar farm.

The two paths I chose to research are strongly correlated. Both have the overall goal of more
sustainably produced electricity. Moreover, both research tracks were focused specifically on
solutions for utilities. Individual consumers would have little use for my solar farm, or wind
turbine implementation plan research. However, to large electricity producing companies like
Dominion here in Virginia, this research could be quite impactful. I spoke with project manager
Adam Maguire at Dominion in the process of my research, and he made a quite interesting point.
He said that Dominion is quite pro new green technologies, particularly solar, however the wild
unpredictability of it can result in less investment. He identified this as the number one issue
inhibiting solar expansion. By having the ability to mitigate this risk, a utility can more
confidently make larger investments in solar generation. Similarly, attempting to build wind
turbines and failing is an expensive mistake for a company. By understanding the societal factors
surrounding the issue, and planning the wind turbine project with those in mind, utility
companies can be more successful, and enjoy a larger return on investment. The joint focus of
these two projects was incentivizing more green technologies on a large scale.

BS (Bachelor of Science)
Machine Learning, NIMBY
Issued Date: