Designing and Implementing a Novel Solar Panel Tracker Leveraging Reinforcement Learning Technique

Abstract

As of today, the average efficiency of household solar panels is less than 20 percent, so less than 20 percent of solar energy is converted into household electricity. Our goal is to explore various techniques that not only enhance this efficiency but also are scalable to households. We plan to use reinforcement learning technique that provides the centralized intelligence for controlling a large set of trackers for household solar panels. Reinforcement learning is a proven machine learning technique to explore unknown environments while maximizing a reward. This is desirable because each household's environment is different. Furthermore, the tracker is purely software-driven with no dependency on specialized hardware. This makes it simple, easy to deploy, and suitable for large-scale application. In this project, we designed and implemented a reinforcement learning technique for a solar panel tracker that adjusts the solar panel’s orientations to maximize the sun ray receptance. The tracker explores an optimal orientation, taking into account of the environmental variations such as the sun’s location and shade caused by surrounding obstacles. To demonstrate the effectiveness of our solution, we placed two identical solar panels side-by-side. One is mounted on our solar panel tracker; the other is stationary. We then collected voltage measurements from both solar panels at identical time intervals between 8 AM to 5 PM each day. We repeated the same benchmark testing for several days. Our results have shown that the solar panel on our smart tracker consistently offers higher electricity voltage output than the one without.