Xinghui Zhao is working on small- and large-scale projects to help individuals and industries save energy.
In the broadest sense, Xinghui Zhao is using her expertise in data analytics and computer systems research to improve the environment. “Because I’m a computer scientist, I’m using computing technology to make a greener world for people,” said Zhao, an assistant professor in the School of Engineering and Computer Science.
At the practical level, she works on both the micro and macro scales in energy-related research. On the micro scale, she is working toward energy-efficient computing by finding ways to monitor and control the energy consumption of various computer operations and applications. These include developing frequency scaling techniques for CPUs, dynamically balancing the workload among heterogeneous processors, and profiling and optimizing energy consumption of applications.
These projects started with an energy-efficient mobile app supported through the Google Summer of Code program, and later expanded in multiple directions. The work applies not only to mobile phones but to many other hardware devices, including multicore CPUs and GPUs. “For these projects, we delivered multiple open source software packages available on GitHub,” she said. “These are research prototypes, but they have already been downloaded and used.”
On the macro scale, she is helping to develop ways to make the world safer for everyone who depends on the power grid for energy. Specifically, Zhao and her collaborators are mining power grid data to identify anomalies, such as unplanned events, faults or cyberattacks, before they have a chance to disrupt the system.
Lately she has been collaborating with faculty and students at Washington State University, Oregon State University and Portland State University on multiple interdisciplinary research projects, which she calls “Big Data Analytics for Smart Grid.” She explains that power companies monitor their large grids by deploying smaller devices called PMUs (phasor measurement units), which measure electrical waves on the grid to detect anomalies. Sponsored by the Department of Energy and Bonneville Power Administration, one of the projects seeks to develop machine learning approaches to mine PMU data for event detection. “Line events might lead to a blackout if not being identified in a timely manner,” Zhao said. “So we gather the historical data from BPA’s power grid, develop and evaluate various machine learning algorithms to mine the data for identifying anomalies, and give recommendations to the operators on what could be the possible cause for these anomalies.”
Besides event detection, data integrity and cyber security of a power grid are also major challenges. In another project, similar machine learning approaches can be used to detect data spoofs. The collaborators have built an inter-university PMU network among the three universities and started to collect data to simulate what would happen if a hacker spoofed one or more PMUs. If a PMU is hacked, it will quickly get out of synch with all the others. “So we try to mine a large amount of data, detect the problematic PMU, and learn that a cyberattack could be happening,” she said.
The collaborators have been working on these projects for more than three years. Ultimately, there will be a framework for BPA to use to analyze its data, with the goal of making its power grid more reliable and robust.
Zhao is a prolific writer, committed to sharing ideas with other experts at conferences and publications. One such project recently was done a year ago with a graduate student for an international conference. It focused on creating energy-efficient microprocessors. Processors have become much faster since the 1970s, but since about 2005, the rate of increase is more or less flat, she said. “There’s this thing called the power wall,” Zhao said. “We can make a faster processor, but it generates heat when it’s running too fast, and the heat takes energy.” She is seeking to develop ways to help people understand that the computer core does not need to run as fast as possible, only as fast as needed, and that the workload can be distributed among a user’s programs to promote energy savings, without extra effort from the programmers.
Making people aware of their energy usage is a huge step toward a greener world. “Energy is an important resource, and we don’t have that much,” Zhao said, “Things can be done at different scales to help this situation, but increasing public awareness of this issue is a key.”