Minutes of Roundtable Discussion 5

Our fifth roundtable discussion took place after Yu Zhang’s talk “Improving Power Grid Reliability and Efficiency via Advanced Signal Processing and Machine Learning”.  His talk slides can be found here.

The roundtable discussion was broken into two parts. The first was on the challenges and future of monitoring and controlling power grid systems, and the second was on the state of data science and machine learning activities at UC Santa Cruz, the institution of the speaker.

As our first foray into the discussion of power grid systems, the host asked what the most pressing problems in modern power system design are. The speaker identified integrating phasor management units (PMUs), which vastly increase the sampling rate to 2880 samples per second and give information of phasors including the magnitude and phase angles of AC voltage and current. For the next few decades of the transition period, the hybrid measurements from both the legacy SCADA (Supervisory Control and Data Acquisition) systems and the newer PMUs make the grid optimization and control tasks even more challenging. A related question is about efficient sensor placement, which asks where to place the costly PMUs with a limited budget to maximize the utility of observations and the state estimation accuracy. Another challenge is to deal with the uncertainty due to the increasing use of renewables (solar, wind, tidal energy etc) along with the decentralization of power generation, which shows a promising future of networked microgrids.

We then discussed more basic questions about the power grid, such as the purpose of solving for the complete voltage information. This is used to maintain the supply-demand balance and keep system frequency stable through the automatic generated control (AGC) which operates at the speed of a few seconds response time. However, these optimization problems need to be solved every 15 minutes or so, requiring faster algorithms than are currently available. To improve on this, one needs to venture into the realm of online optimization techniques.

One of the participants brought up the parallel with transportation networks, where the question is also about solving for feasibility. In both cases, it is unclear whether or not any of the feasible solutions would actually give a physically meaningful solution, or if there is anything else to be done besides applying the “eyeball norm” at the end (which is done in practice on the transportation networks). There was some speculation on using Bayesian techniques to accomplish this in a more rigorous way, but this appears to have not yet been tried.

Then, there was a question if real data from industry are available to academic researchers and the results of their investigation are publishable. Unfortunately, the datasets involved are a closely guarded secret on the part of the power companies, primarily for security reasons according to the speaker. Consequently, even if academic researchers may be able to obtain the real data, but they cannot publish papers using such datasets. Hence, the researchers must rely on synthetic datasets developed by those with access to the real data, along with added noise, which may or may not reflect the real problems. Like most data science questions, the speaker identified understanding domain specific knowledge, such as the physics of circuits, as an essential part of working in the field.


Finally, data science and machine learning activities at UC Santa Cruz were discussed. The speaker identified four major areas: energy; climate change; biology; and transportation. His work fits squarely under the energy banner;  Elliot Campbell and Kai Zhu are two researchers who use data to do climate change modelling of carbon emissions and vegetation growth. The genomics center at UCSC is thriving, and has recently released Xena, a tool to facilitate cancer research using genomics and clinical data. Finally, the speaker described the D3 Data Science Research Center and their increasing focus on the social ramifications of data science.  UCSC was awarded the NSF TRIPODS grant two years ago; see this link for the details.


[Scribe: David Weber (GGAM)]