Optimized Use of Renewable Resource in Conjunction with Energy Storage Utilizing Reinforcement Learning

---

Project Team

---

Project

---

Project Video

---

Project Abstract

Electric vehicles (EV), as well as traditional energy storage methods in conjunction with renewable energies such as photovoltaics (PVs), have the potential to not only provide economic benefits to residential users, but also reduce harm to the environment, by minimizing the need for carbon-based energy sources. Optimized charging techniques utilizing Vehicle-to-Grid (V2G) technologies maximize these benefits to the user while reducing peak loads on the electrical grid. Computational methods such as mixed-integer programming can provide optimized charging schedules, based on provided user profiles. However, with the rapidly progressing abilities of neural-networks, especially in Reinforcement Learning (RL), more general solutions to both EV charging as well as physical modeling problems can be leveraged. This study utilizes several RL techniques including Asynchronous Advantage Actor-Critic (A3C), Advantage Actor Critic (A2C), and Proximal Policy Optimization (PPO) to explore optimized charging solutions. In this study, the charging agents are trained using the RL library [1] within the open-source “Ray” framework [2] to solve the residential EV charging problem. “Ray” provides a highly customizable API, with hyperparameter tuning capabilities that allow for a wide variety of customizable and optimized models. After training the models results were compared to gain understanding into the effectiveness of each algorithm to solve energy management problems. It was found that both the A2C as well as the PPO algorithms are able to learn the custom environment, although A3C is currently unable to settle on an optimized control strategy.

---

Evaluate this Project

Use this form link to provide feedback to the presenters, and add your project evaluation for award(s) consideration.