Machine learning and price-based load scheduling for an
optimal IoT control in the smart and frugal home

Rachneet Kaur, Clara Schaye, Kevin Thompson, Daniel C Yee, Rachel Zilz, RS Sreenivas, Richard Sowers
University of Illinois at Urbana-Champaign
Energy and AI, Elsevier 2021

[Slides]
[Teaser Video]
[Paper]

Abstract

We pose and study a scheduling problem for an electric load to develop an Internet of Things (IoT) control system for power appliances, which takes advantage of real-time dynamic energy pricing. Using historical pricing data from a large U.S. power supplier, we study and compare several dynamic scheduling policies, which can be implemented in a smart home to activate a major appliance (dishwasher, washing machine, clothes dryer) at an optimal time of the day, to minimize electricity costs. We formulate our scheduling task as a supervised machine learning classification problem which activates the load during one of two preferred time bins. The features used in the machine learning problem are hourly market, spot and day-ahead prices along with delayed label of the prior day. We find that boosting tree-based algorithms outperform any other classification approach with measurable reduction of energy costs over certain types of naive and static policies. We observe that the delayed label has most predictive power across features, followed, on average, by spot, hourly market, and day-ahead energy prices. We further discuss implementation issues using a micro controller system coupled with cloud-based serverless computing and dynamic data storage. Our test system includes an interactive voice interface via an intelligent personal assistant.

Figure: Proposed IoT ecosystem for optimal energy savings in power appliances.


Teaser video



References

Rachneet Kaur, Clara Schaye, Kevin Thompson, Daniel C Yee, Rachel Zilz, RS Sreenivas, Richard Sowers. Machine learning and price-based load scheduling for an optimal IoT control in the smart and frugal home. In Energy and AI, Elsevier 2021 [BibTex] [Text]


Acknowledgements

The authors would also like to thank John Nguyen for assisting in the implementation of these ideas. RK is thankful to William Chittenden for the William A. Chittenden Award.

Website adapted from Jingxiang, Richard and Deepak.