Discipline: Computer Sciences and Information Management
Subcategory: Computer Science & Information Systems
William Kluegel - New Mexico State University
Co-Author(s): Muhammad Aamir Iqbal, New Mexico State University, Las Cruces, NM; Ferdinando Fioretto, University of Michigan, Ann Arbor, MI; Enrico Pontelli, New Mexico State University, Las Cruces, NM
According to the U.S. Energy Information Administration, over 4 trillion kilowatt-hours of electricity were generated at power stations in the U.S. in 2016. A large portion of this power is generated during the day. Utility companies are forced to run extra generators during peak hours that are expensive to run and maintain. The Smart Home Device Scheduling (SHDS) problem formalizes the coordination of smart device schedules to reduce peak consumption across multiple smart homes. SHDS is a multi-agent decentralized approach, which makes it desirable for data privacy.
Distributed Constraint Optimization Problems (DCOPs) have become a popular way to model autonomous agent behaviors. Researchers have used DCOPs to solve various multi-agent coordination problems. Despite the large variety of algorithms developed to solve DCOPs, there hasn’t been much work focused on the benchmarks used to assess solution quality. DCOP algorithms are usually evaluated using simplified problems with unrealistic assumptions such as each agent controlling one variable or all problem constraints being binary. The SHDS problem can be modeled by a DCOP because each agent has its own goals in addition to the collective agents’ goal of reducing peak energy consumption.
Implementing SHDS would require several houses equipped with smart devices and user defined schedules. Each house must be equipped with a means of communication with neighboring houses. This would be expensive and difficult to set up, showing the need for realistic synthetic datasets.
We have created a realistic synthetic dataset generator for the SHDS problem, providing an effective way to evaluate different approaches in an environment that is close to a real implementation. Each dataset represents a random selection of houses inside a 200m x 200m grid. The grid is divided into coalitions which act as neighborhoods. Houses within each coalition have a list of neighbors as well as a list of randomly chosen rules which must be satisfied.
Using the SHDS datasets, we ran experiments using our own Java implementation. SHDS problems are very difficult so we designed an incomplete method of finding a good solution. The results of the experiments can be used as a baseline to compare other solutions to.
Our future work will focus on device scheduling in a single house, defining more specialized models for each type of device. We are currently working on a more accurate HVAC model.
References: Ferdinando Fioretto, William Yeoh, and Enrico Pontelli. 2017. A Multiagent System Approach to Scheduling Devices in Smart Homes. In Proceedings of the 16th Conference on Autonomous Agents and MultiAgent Systems (AAMAS ’17). International Foundation for Autonomous Agents and Multiagent Systems, Richland, SC, 981-989.
Ferdinando Fioretto, William Yeoh, and Enrico Pontelli. 2016. Multi-Variable Agent decomposition for DCOPs. In Proceedings of the Thirtieth AAAI Conference on Artificial Intelligence (AAAI’16). AAAI Press 2480-2486.
Not SubmittedFunder Acknowledgement(s): Funding was partially provided by NSF grant 1345232
Faculty Advisor: Son Tran, tson@cs.nmsu.edu
Role: I was responsible for the dataset generator as well as running the experiments and collecting data.