Teasing Detailed Home Habits from Aggregate Energy Consumption Data
Written by Diane J. Cook and Chao Chen
Machine learning techniques can be applied to sensor data collected from smart homes to reveal activity patterns of the residents, which can then be correlated with measured energy consumption. By associating activities with energy use and costs, intelligent systems can be devised to automatically control home environments so as to improve energy efficiency and cut expenses.
Although households and buildings are responsible for over 40 percent of energy usage in most countries, many people receive little or no detailed feedback about their personal energy usage. Bills traditionally provide a month's total energy and a total price to be paid, leaving homeowners to guess—after taking any changes in fuel costs into account—what might explain a higher or lower than usual bill. The typical utility bill provides no information about the relationship between a person's behavior and corresponding energy usage. Since such information could help individuals modify habits in ways that would be beneficial for both the household and community, it would be desirable to develop technologies that could extract the information from smart homes and communicate it to residents.
A smart home environment is one that acquires and applies knowledge about its residents and their physical surroundings in order to improve their experience in that setting. Such home environments, equipped with sensors for detecting motion, light level, temperature, and energy and water consumption, are ideal testbeds for investigating techniques of inducing behavior changes to reduce energy footprints.
One such technique is energy consumption analysis. The general idea is to employ data mining techniques in order to analyze electricity consumption data and to identify patterns of interest to utility companies and their customers: Sequences of usage patterns that appear frequently at different time scales (daily, weekly, monthly, yearly) and across different homes; trends of electricity consumption (steadily increasing, decreasing, cyclic, seasonal) for individual homes and across the community; and anomalies (sudden peaks or drops on consumption) for individual homes and across the community.
For example, using abnormality detection algorithms, customers can be notified that they consume an exceptionally large amount of energy during some specific period. With the help of other sensors and techniques, more detailed information can also be provided, including when customers performed certain activities, which rooms they occupied, and what appliances they used most frequently during that period. This information can be transmitted to customers in timely fashion via phone, email or the Internet.
To make informed decisions, residents need to know current energy consumption in real time and ideally would need to be alerted when appliances are being activated at unfavorable times. Without installing additional sensors, Non-intrusive Load Monitoring (NILM) techniques can be designed to detect switch events at individual appliances on a single electrical circuit and communicate those events to a home energy management system.
Several suitable Internet protocols exist, including Extensible Messaging and Presence Protocol (XMPP), which enables servers to communicate directly with smart phones, email and webpages. PlotWatt applies cloud-based load monitoring algorithms to analysis of smart meter data and tells customers how much they are spending to run each appliance, without actually monitoring the individual appliances. The web-based interface can be accessible from any web-connected computer, tablet or phone.
Other analytic techniques include energy consumption prediction and consumption visualization. With the aid of environmental sensors, machine learning algorithms can identify resident behaviors in smart homes and predict future energy usage given the time of day and day of week as well as sensor readings and identified activities. Visualization techniques give the non-technically-minded smart home resident an informative, user-friendly, and intuitive graphical interface that presents information about their energy consumption in the home.
The visualization tools should allow residents not only observe their electricity consumption in real time, but also visualize consumption patterns, trends and anomalies. These tools can be presented locally, remotely over the web or via a mobile device. Additionally, there is a place for the user interface to provide remote control features to manually interact with devices throughout the home.
Some such smart grid technologies are already widely applied in the industry. Within the field of smart grid architectures, smart home-based technologies represent a small, but growing part, helping individuals to monitor their personal power usage with the goal of making changes in their lifestyles and achieving energy efficiency. Continued research and engineering effort would provide insights into the relationship between home resident behaviors and energy consumption. Additionally, this work will generate a suite of tools that directly benefit individual people wishing to reduce their energy footprints and power utilities by providing a more efficient and adaptive smart grid.
Diane Cook, an IEEE Fellow, is Huie-Rogers Chair Professor in the School of Electrical Engineering and Computer Science at Washington State University. She received a bachelor's degree at Wheaton College in 1985 and a master's at the University of Illinois in 1987. She earned her Ph.D. in computer science at the University of Illinois in 1990. Her research interests include artificial intelligence, machine learning, graph-based relational data mining, smart environments, and robotics.
Chao Chen is a doctoral student in the School of Electrical Engineering and Computer Science at Washington State University. He received a bachelor's degree at Anhui University, in China, in 2005, and a master's at the University of Science and Technology of China, Hefei, in 2008. His interests center on sensor networks in smart environments, smart grids and machine learning applications in energy consumption.