Dylan Shanes

About Me:

Hi, I’m Dylan! I am a Computer Science major at WPI and am training to be a high school math teacher. I want to be a teacher because having positive role models as a teen can be very influential.

About the Lab:

We are doing research with Principal Investigator Bashima Islam on energy harvesting devices in an effort to create a device that will be able to predict how much energy it will have in the future and use its available energy accordingly. This would reduce our dependance on batteries which have harmful effects on the environment.

Project:

Our part of the project is analyzing and characterizing energy datasets we collect and collected by other researchers to train an AI that will be able to make energy predictions for various environments. It’s important to make this prediction to keep the device running continuously without maintenance. The AI would also allow us to predict energy conditions for environments that we have not collected data from yet.

Weekly Updates:

  • Week 1:
    • We started by getting a sense of work done by other researchers in the field of energy harvesting. Currently, there is a device called Ekho that is able to replicate in a lab setting energy data that was collected at other locations. After we learned about this device and work done by other researchers, we organized several datasets from their research to prepare for analysis.
  • Week 2:
    • This week we began working with the first energy dataset. The dataset contains solar energy data in volts from a sensor placed on the roof of a lab in Spain over the course of 2 months. Our goal is to characterize the dataset by how many anomalies it has, the predictability of data, and biases for time of day. We created a plot of the data and ran algorithms to count how many anomalies are in the data and give a mean squared error score for the dataset.
  • Week 3:
    • For week 3 we took a deep dive on some thermal data. Our goal was to go through 16 thermal datasets and find predictability scores for each using both Harmonic and Polynomial Regression. Using a loop, we found the polynomial degree for each dataset that results in the smallest error score. We found that polynomial regression yielded lower error than harmonic regression in every case.
  • Week 4:
    • This week we conducted a subjective analysis on the thermal data by comparing and contrasting dataset characteristics. We also began working with a radio frequency (RF) transmitter/harvester that we will use to start collecting our own energy data. The device collects data that we can upload and analyze later on. Moving on from the thermal data, we started analysis on a large dataset (60 million+ records) of solar data.
  • Week 5:
    • After exploring Harmonic Regression models with different periods, we discovered a discrepancy between the paper the dataset came from and the data itself; the paper says that data was taken 10 times per second, while the dataset suggests that data was taken once per second. After reapplying the regression model with the new information, we concluded that the solar data has a period of one day, meaning that voltage rises and falls in day-long cycles.

Lesson Plan and Poster:

Lesson Plan Draft

Poster