Technical Implementation - Reinforcement Learning

Reinforcement Learning (RL)

An end-to-end project implementation to demonstrate reinforcement learning algorithms for optimal control of renewable energy technology. The pages below are grouped into the following categories, which align with the phases of the project:

All code used in the project can be found here: Code Repository on Github

A summary of the research can be found here: Research Summary

Concepts

The Problem Statement

An explanation of the problem at hand, why reinforcement learning is well-suited to solve it, and the types of problems that reinforcement learning promises to solve

RL Fundamentals

Some key concepts around reinforcement learning, including Markov Decision Processes, agent components, and types of RL Agents.

Helpful RL Resources

A list of helpful courses, books, and blog posts for diving into RL

Building the Agent — Hardware & Firmware

The project optimizes a dual-axis solar panel to maximize energy output. Simulation datasets are provided in the Agent Design & Experimentation section, but full instructions for building your own dual-axis-tracking panel system are in the pages below.

solar-rl-system-schematic-system-cover view.png

System Design

A technical view of the system design, flow of information between the agent and circuit, and circuit schematic.

Circuit Schematic

Detailed circuit schematics for controlling the dual-axis panel via servo motors and measuring energy produced from the panel.

Dual-Axis Panel Design

Mechanical and electrical schematics for the dual-axis panel design used in this project

Parts List

A full list of parts required to build and run the project

Building the Agent — Agent Design

Experiment Design

Generating the simulation environment from real measured power data

RL Agent - Q-Learning

Initial agent implementation using Q-learning and the shortcomings of this design

RL Agent - Softmax Actor-Critic

Full technical implementation of softmax actor-critic agent using average reward for the solar energy optimization problem

Experiment Results

Experiment I: Shifting Env

Exploring softmax actor-critic performance with a dynamic environment

Experiment II: Shifting Env with Time of Day

Exploring improvements to softmax actor-critic performance with a dynamic environment and time of day state feature

Open-Source License

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.