Developing the foundational capabilities for understanding and analyzing the next-generation electric power grid is a multidisciplinary endeavor. A sample of the disciplines required includes electric power engineering, mathematics, statistics, operations research, computer science, and economics. Each of these disciplines in turn could have subdisciplines that would be appropriate for specific problems—for example, optimization, nonlinear dynamics, machine learning, or databases. And while some multidisciplinary teams form naturally through mutual collaborations, a more strategic approach will be required to build a more effective multidisciplinary community to address the challenges described in this report. Interestingly, when these partnerships do form, it is often the case that the use of well-known techniques from one discipline (which may not perhaps be as well known in another discipline) has yielded a breakthrough—as, for example, the use of interior-point optimization methods for solving the large linear programming problems that arise in solving the optimal power flow problem.
However, adapting and disseminating state-of-the-art algorithms and methods from other disciplines to the electric power systems community and developing entirely new research areas that build on the joint strengths of two or more communities are challenges. Power engineers have to formulate the problems and help the rest of a multidisciplinary team understand the underlying issues and their nuances. This can be difficult, as power engineers frequently do not have the background or training that would allow them to articulate their problems in the language of another discipline. How would these engineers know that the other discipline has methods or expertise that could be brought to bear on the problem of interest? Conversely, experts from other disciplines such as mathematics need to acquire requisite background for understanding the language of power engineers. There are many examples where people with a core competency in a related area (e.g., control or linear systems) formulate problems in their own language that are consistent with what they know and can solve but that fail to capture the key issues or the subtleties that differentiate a useful, pragmatic approach from one that is of only theoretical interest. So, a common framework needs to be created and researchers educated across disciplines so they become fluent in one another’s language and approaches to problem solving and become familiar with state-of-the-art methods in the other disciplines. One strategy is to use case studies to provide insights into problems of common interest, because such studies give concrete details that allow the understanding of heterogeneous groups of researchers to converge.
While it is clear that all the contributors to cross-disciplinary teams must understand the underlying problems, it is equally important that they be cognizant of a broader context. Take the example of wind forecasting. Power
system engineers often cite the need for better forecasts so that the availability of wind power can be estimated for a particular service provider. Recent research has contributed to improving such forecasts using the techniques of data assimilation described in Chapter 4. However, the atmosphere has long been regarded as a chaotic system that is very sensitive to initial conditions. This suggests that there are inherent limits to the predictability of the weather. Ensemble forecasting based on this assumption produces probabilistic forecasts rather than specific predictions of wind velocity or cloud cover that would affect renewable energy resources. So while we may continue to work toward more precise forecasts, as desired by the power systems community and many others, multidisciplinary research teams can recognize that wind models for the grid must be treated probabilistically, and that they will produce results with uncertainty and can have larger variance than we desire.
How then should it be determined which disciplines and subdisciplines could contribute to the analytic foundations for the next-generation electric grid? As all the problems have multidisciplinary dimensions, it is important to determine which dimensions are more important or more difficult for each individual application. One can then determine the strategy appropriate for each problem. In other words, just because teams engaged in research on power systems have members from different disciplines does not ensure their ability to solve the specific issues that are facing the grid like those outlined in the previous chapters. For that multidisciplinary teams need to be brought together that have individuals with backgrounds in the specific subdisciplines that are needed for each application. One example from the 1990s can perhaps illustrate some of the key characteristics of a successful multidisciplinary team.
By 1988 it became clear that fundamental and lasting changes were coming to the electricity business because of the restructuring that the Federal Energy Regulatory Commission was beginning to mandate. But at that time, there were very few power engineering programs left at major research universities, raising concerns about how new technologies necessary for the grid changes would be developed. In 1995 five universities (Cornell, University of Illinois, the University of California at Berkeley, Howard University, and the University of Wisconsin) were able to establish a National Science Foundation (NSF) industry/university cooperative research center, the Power Systems Engineering Research Center (PSERC).1
The fundamental premise of PSERC was that no single university had the breadth of expertise needed to address the research issues. In addition, it was recognized that multiple disciplines would have to be brought together. Quoting from the original 1995 proposal, “The Center’s . . . basic premise is that engineering considerations should not be an afterthought but rather be a principal force in the planning process of a restructured Industry. Consequently, the center’s agenda differs from the institutional research programs currently in operation in that it focuses on the technological needs (such as computational methods, information needs and protocols, and control schemes) that are fundamental to successful implementation of alternative economic paradigms for opening the power system to greater competitive forces. It also provides stronger organizational and personal structure that encourages interdisciplinary research and communication among engineers, economists, and computer scientists.” When PSERC was established in 1995, it was well positioned to participate in work needed to implement the landmark transition set in motion by the April 1996 release of FERC orders 888 and 889.
After PSERC was established it began to expand the number of its member schools in order to capture the expertise needed to address the broad array of problems envisioned by the founders. In addition it organized its programs into three “stems”—namely, markets, systems, and transmissions and distributions (T&D) technologies. Within the markets stem were engineers and economists, most of whom had not collaborated before. The first hurdle to be overcome by this collaboration was language. While mathematics is a universal language common
to many disciplines, the context in which it is to be applied is not. Even nomenclature had to be settled on. For example, in the power literature it is common to label real and reactive power P and Q, whereas to an economist P and Q are price and quantity. In order to facilitate communication across disciplines, a test-bed platform called PowerWeb was created that could be used to coordinate the work across disciplinary boundaries as well as to test new market design concepts. The premise was that the experimental economic concepts pioneered by Nobel laureate Vernon Smith, together with a more realistic engineering representation of the power system, would reveal interesting and useful insights about market design. A substantial problem for economists was that there was very little experience with or insight into repeated auctions of the complexity that were being contemplated at the time. It was clear that, because the auctions were repeated, learning was possible and even necessary. If market participants can learn, they can then presumably optimize their position by learning the agendas of other market participants. It is not necessary to openly collude to learn what behavior is best for maximizing profits given tolerance for risk. Economists were of the opinion that the best-designed markets are those designed to reveal true costs of the participants. Because of the ever-changing operational equilibrium of the delivery network, the Nash equilibrium does not exist. The new techniques devised by the engineer/economist collaboration were able to address important tasks such as the following:
- Explain the origin of price spikes that were occurring,
- Identify where pockets of market power might be,
- Quantify the advantages of having a demand-side market,
- Determine the number of participants needed for a supply-side-only market to be competitive,
- Observe that a distributed unit commitment schedule may be as efficient as an optimization-based centralized commitment, and
- Explore a host of other interesting phenomena that would not have been possible without the cross-disciplinary collaboration.
One of the important findings from these multidisciplinary collaborations was that having a complete engineering model was crucial in determining the economic outcome of any of the many market designs being discussed. It is interesting to note that today, most Independent System Operators have market-testing platforms that are used to test new concepts before market participants experience them.
Another problem that surfaced in PSERC’s cross-disciplinary research was how to efficiently collaborate across multiple institutions. Successful collaborations of this type involve overcoming organizational and social differences and establishing the same kind of trust that comes from working with someone down the hall who is more easily accessible. To some extent, the Internet communication technology that was then newly evolving helped ease some of these problems.
Today, students who graduated from programs that crossed the power engineering/economics boundary are professors in various departments around the country, teaching what is now an integrated discipline. Some professors are in electrical engineering, some in operations research, some in economics, and some in other departments. But they are in communication with each other and publish in the same journals—hallmarks of a well-functioning research community. The industry and the country as a whole have benefited from the interdisciplinary marriage of these once disparate disciplines. Most important, this community of power systems economists is focused on solving problems directly related to the grid.
While PSERC has contributed in important ways to building a multidisciplinary research community that supports the electric grid, its mix of expertise does not extend to many of the areas of importance to developing the analytical and mathematical capabilities that will be needed for the next-generation grid. For example, computational tools to address the ac optimal power flow problem discussed throughout this report are likely to require fundamental advances in optimization that are typically outside the domain boundaries of PSERC. As documented throughout this report, many of the analytical challenges require insights from the mathematical sciences, which in turn call for new cross-disciplinary connections that are currently scarce.
Earlier chapters of this report identified the following main analytical challenges for the future grid:
- Making effective use of large data from improved measurements,
- Modeling the availability of uncertain renewable energy resources and their effect on grid reliability,
- Building and operating smart grids that incorporate demand response, and
- Improving optimization methods for nonlinear, nonconvex, and stochastic problems.
These challenges call for new multidisciplinary research communities that draw from mathematics, computational science, computer science, operations research, statistics, and control theory. Since these multidisciplinary research communities are still emerging, we should look broadly at examples from other disciplines for insight into how to best form these teams and enable their effective operations. Several types of existing models are relevant.
The Department of Energy (DOE) has long-standing experience in developing programs that span multidisciplinary groups, including the Scientific Discovery through Advanced Computing program and the Advanced Simulation and Computing Program. Both programs were designed to build the simulation capabilities needed by computational scientists and engineers to make effective use of the vast computational and data resources provided by the DOE laboratories. Finally, the Mathematical Multifaceted Integrated Capabilities Center is another good example of building multidisciplinary teams. These centers have a strong focus on the mathematical sciences, as described in DOE Program Announcement 12-698: “These science and engineering challenges must be abstracted into an interrelated set of mathematical research challenges that require new integrated, iterative processes across multiple mathematical disciplines.”
The Mathematics Climate Research Network (MCRN), which started in 2010 with support from the NSF, provides another model for fostering multidisciplinary collaborations between mathematicians and scientists in another discipline. Climate science is a field that is already organized around large data, comprehensive computational models, and the use of high-performance computers. Thus, a large initial time investment has been required of individuals whose research is in the area. The goal of MCRN has been to reduce the barriers to engaging mathematical scientists in problems emerging from the study of the climate system. One of the motivations was to significantly increase the number of mathematical researchers working in this area. The strategy was to create a community of individuals who would support and inform one another in defining key mathematical directions and challenges while pushing the resulting research to a high scientific level. Part of the effort has focused on bringing junior people into the area through the design of effective training elements incorporated into existing mathematics graduate programs at MCRN member institutions. Since the collection of participating researchers is widely distributed, the network has utilized and further developed web-based tools for communication, conferencing, and collaboration. The network has grown to over 200 individual members and makes extensive use of web-based collaboration tools, computational sharing capabilities, and communal data and software storage. Although studying the electric grid was not a primary objective of the network, a research group emerged in 2014 with a focus on determining the mathematical challenges posed by the next-generation grid. This small group represents, in embryonic form, a web-based effort to increase the participation of mathematical scientists in this area.
While the analogy between power systems engineering and the atmospheric and climate sciences is a good one, there are significant differences that bear upon the organizational structures that the committee recommends for the future electric grid. In the atmospheric sciences, government agencies have long gathered weather data and amalgamated these data into publicly available databases. Large-scale modeling efforts for climate models and numerical weather prediction thus have a base of publicly available data resources that serve to coordinate and ground atmospheric modeling research. However, an analogous step for grid modeling research is challenging because a lot of the data are proprietary or are protected for homeland security purposes and designated as Critical Infrastructure Protection (CIP) data.
The field of genomics provides an excellent illustration of how databases can be central to a scientific research area. When the visionary Human Genome Project was initiated to discover the complete DNA sequence of humans and other organisms, its leaders made several astute decisions that have been important factors in the scientific
success and impact of the field. Here we mention two. First, the leaders mandated that a comprehensive database would be created and that researchers would be required to add their results to this database in a timely manner. Second, they invested in the development of software tools that would enable all biologists to utilize the information in this database. The National Library of Medicine, part of the National Institutes of Health, was given responsibility for both. It created research groups and tasked software development staff to deal with the data. The impact of these decisions has been phenomenal. All of biology has benefited from the advances in DNA sequencing technology and algorithmic methods for sequence analysis. One of the key provisions was the availability of open and comprehensive data sets that modelers and algorithm and software developers could use for their research.
Climate research and genomics illustrate the central role that data play in 21st-century science. The electric power industry is poised to make this transition in data intensity, but it has not yet settled on an organizational structure that makes effective use of its data. The sensitive nature of the data (including CIP and proprietary concerns) and the complex manner in which the industry operates call for careful and secure data management before any real data can be released in general to researchers. No entity has yet assumed responsibility for this task on a national level, but creating a center to undertake it would have huge potential benefits to the economics and reliability of the next-generation grid. Below, the committee proposes the establishment of such a multifaceted center.
Any researcher intending to work on a new problem area finds that familiarization with the state of the art in the new area is a difficult task. In the area of power engineering, an additional hurdle, as pointed out in Chapter 6, is the scarcity of data that are representative of the real power grid, including for example non-CIP data. This is true of both the data that describe the power grid as well as the measurement data under various operational conditions. In addition, in developing novel ideas for next-generation algorithms one always goes through a sequence lasting years, where at the beginning the algorithms fail most of the time before finally evolving into ones that are robust. Data that reflect real conditions are necessary for moving this process forward. In this context, it is critical that accessibility to real data not be the limiting factor in developing new algorithms. There may be many reasons why real data are not used in research, and where this is the case we should strive to develop and make available high-quality and accurate synthetic data.
One solution to this problem is to develop synthetic data that exhibit the same behavioral characteristics as real power systems of realistic size, perhaps using some of the tools described in Chapter 6. However it is done, developing synthetic data will require substantial effort so that they can be used effectively as a surrogate for real data—for example, by capturing real behaviors and responding as a real grid would to what-if scenarios.
A recent Funding Opportunity Announcement from ARPA-E asks for the development of synthetic data to be used to test optimal power flow methods. This is a good start, but it is necessary to develop several libraries because not all power systems exhibit all the possible behaviors. There is diversity in both the kinds of data needed to investigate different types of problems and the data based on network architecture. For example, a highly stable system like the Eastern Interconnection is tightly coupled, while a system with longer transmission lines and looser coupling (e.g., the Western Interconnection) can exhibit more dynamic problems. Synthetic data sets should exhibit all the different characteristics that researchers might want to study, without duplicating existing sets.
Recommendation 9: The Department of Energy should sponsor additional efforts to create synthetic data libraries to facilitate studies of, and tool building for, the reliability and control of the future electric grid.
In addition to having access to synthetic data, one must be able to simulate portions of the grid so as to study the various behaviors in steady-state or faulted conditions, under heavy or light loads, and so on. As mentioned in Chapter 3, simulation tools of various types are used widely for engineering purposes. However, these tools are all commercial products and the programs are proprietary. Thus it is often not possible for researchers to experiment with or add to the models and algorithms in these programs, although the committee notes that some commercial
programs support user-defined models and the creation of external programs that extend their functionality. While many vendors provide licenses to their programs for university research and education usage at greatly reduced cost, these licenses are still an impediment for researchers from disciplines other than power systems engineering who wish to explore the problems. Because these programs are highly complex, incorporating many different models of the thousands of components on the grid, it is difficult for an experienced researcher, let alone a newcomer, to duplicate such tools with a reasonable amount of effort. Moreover, development of new algorithms, and their application to power systems, involves research issues that are unlikely to be addressed adequately by software vendors (see Chapter 6). At present, researchers have to try new algorithms or controls on simple test systems with synthetic data (for which custom programs can be written), or they are restricted to making limited add-ons to the existing commercial software. Moreover, the changes happening to the power grid and the IT infrastructure overlaying it are requiring simulation tools that are not just simple extensions of the existing tools. Rather, these changes call for fundamental adjustments to the underlying assumptions in the models and algorithms.
So, in addition to having a library of synthetic data, it would be very beneficial for the research community to have access to a library of simulation software. Even if one has to pay to use the software (because most are commercial products), it becomes very convenient to be able to compare and contrast existing simulation tools. Of course, it would be even more efficient if some of these software packages were open source and researchers could then modify, add, and test their own algorithms. Just starting such a library will encourage researchers and the industry to add their own open-source software to the collection.
Recommendation 10: The Department of Energy and the National Science Foundation should sponsor the development of new open-source software for the next-generation electric grid research community.
DOE has an ongoing effort to coordinate the power grid research that it funds at the national laboratories. In fact, it has tasked a consortium of its national laboratories with mapping out a multiyear research plan for the power grid. A prime objective is the coordination of all the disparate but grid-related research projects being conducted at the national laboratories today. Such coordination would allow, say, the specification of compatible software interfaces—for example, those utilizing standard database structures—to be incorporated into analytical tools made for different purposes, which would improve processing times. The national labs have an invaluable resource in their multicore parallel processors, and their direct involvement in this effort is much desired.
Recommendation 11: In view of the importance of its efforts to coordinate power grid research at the national laboratories, the Department of Energy should broaden this coordination to include academic and industry researchers.
Recommendations 9 and 10, about creating libraries of synthetic data and providing access to software tools, should be well aligned. In addition, since such libraries are fully digital, they do not have to be in one physical location. Creation of interconnected virtual libraries with strict interoperability standards is an option for software tools. Of course, this does not do away with the need for physical locations for testing of actual hardware, but this report is mainly concerned with the research on and development of the analytical and computational tools needed for the planning and operation of the grid.
Implementing Recommendation 11 on the coordination of research efforts across the national laboratories and well beyond will greatly lower the hurdles facing experts from fields other than power engineering who wish to join the research effort. This would allow the multidisciplinary teams needed to solve these complex problems to be formed with greater ease.
As has been noted throughout this report, the power grid is changing ever faster, not only to adopt new technologies but to adapt to the changing climate. The changing climate is driving the world to reduce its use of carbon-based fuels to slow the warming trend, at the same time that work is done to strengthen the grid against extreme weather events. Thus all power engineering research roadmaps as adopted by DOE, NSF, the Electric Power Research Institute, and other research agencies call for new analytics for the planning and operation of the fast evolving power grid. Analytics usually refers to the suite of computer-based tools that is used by power engineers for designing the transmission and distribution systems and developing real-time monitoring and control systems for them.
The first digital analytic tools appeared in the late 1960s and early 1970s. The first power flow and transient stability algorithms, the first optimization algorithms for power flow and scheduling generation (unit commitment), the first Supervisory Control and Data System, and the first energy management system with state estimation and contingency analysis, were all developed in this short period in a great flurry of creative energy. In the next four decades these tools improved with the faster evolving computer hardware and software, but by and large, the methodology and algorithms did not change very much. Overall, the improvements since the 1970s in power engineering analytical tools have been incremental rather than transformational.
The main reason for the burst of creativity during the early period of power engineering analytics was the entry into the power engineering community of many new people who brought previously unknown mathematical techniques to bear on the problems. Tinney and Walker (1967) introduced sparse matrix techniques that solved power flows for large systems; Schweppe and Wildes (1970) brought in least-squares state estimation to solve the power network equations in real time; and many others brought in new optimization methods, new numerical solutions for grid dynamic behavior, and so on. The committee believes that the changes taking place in the grid today cannot be handled by incremental improvements in grid analytics. To be able to make the transformational changes needed, a research environment needs to be created that attracts mathematicians and experts in computation into power engineering research.
Today’s organization structure for conducting power engineering research is being significantly built up to handle the renewed emphasis on power grid analytics. Many of the DOE national laboratories are playing a role in tackling this problem. Presumably this will be further extended by bringing academic institutions and industrial research groups into the fold. NSF continues to support power engineering research in general and two Engineering Research Centers focused on this topic. The DOE Office of Advanced Scientific Computing Research supports some relevant research in applied mathematics and related power engineering research. Although these efforts are very much needed and these organizations are getting bigger to handle the larger volume of research, they are not fundamentally changing the research environment.
The committee believes that for transformational research to take place, the research environment must proactively attract new mathematicians and computation experts into this research. Moreover, it would not be enough to just add a few mathematicians to the existing research teams at the national laboratories and the power engineering research groups at universities. The problems faced by the power grid will require sustained innovation in grid analytics. This calls for a more permanent organization that specifically nurtures talented researchers who may be new to the power engineering community but who would commit themselves to becoming knowledgeable about the research topic. The committee sees the need for a research center to foster such engagement.
Such a center might include a physical facility and staff to support the management of software and data sets (which themselves may be distributed elsewhere), along with virtual or in-person research collaborations among engineers, mathematicians, and other scientists throughout the country. The facility could be housed at a university, a national laboratory, or elsewhere. It would bring together experts from industry, national laboratories, and academia. The center would not necessarily need its own facilities for data storage or for testing the new algorithms and tools; in fact, the high-performance computer capabilities of DOE national laboratories or of cloud-based computing could be utilized.
Recommendation 12: The Department of Energy should establish a National Electric Power Systems Research Center to address fundamental research challenges associated with analysis for the future electric system. The center would act as an interface between the power industry, government, and universities in developing new computational and mathematical solutions for data and modeling challenges and in sharing valuable data.
The recommended center would include on its staff mathematicians, computational scientists and engineers, and power engineers. Their research would focus on the mathematical foundations described earlier in this report, such as fast power-flow programs, optimization programs for planning and operations, and new stochastic tools and methods. Since issues of social science might be of importance for management of the future grid, social scientists might be included in this multidisciplinary team, although social issues have been mentioned only in passing in this report. Some of the social science research needed was presented at the committee’s 2015 workshop2 by Miriam Goldberg. Her remarks are summarized in NASEM (2015). For scientists in these disciplines, one of the obvious benefits in working with the center would be access to otherwise unavailable data, both synthetic and real. Computer models developed at this center using synthetic data could also be tested later on in a secure facility using CIP data.
The center should foster a cross-disciplinary, multi-institutional approach to the analytical problems of the next-generation grid on a scale not normally possible in individual institutions. It should preserve and enhance the natural synergism between research and education while encouraging industry participation in its activities. The center would foster the education of future engineers in both industry and government as well as future faculty members. Through its industry connections, the proposed center could help researchers understand the behavior of the current U.S. power system, contribute to its orderly evolution, and enhance and extend the capabilities of the university and broader industry communities.
NASEM (National Academies of Sciences, Engineering, and Medicine). 2015. Mathematical Sciences Research Challenges for the Next-Generation Electric Grid: Summary of a Workshop. The National Academies Press, Washington, D.C.
Schweppe, F.C., and J. Wildes. 1970. Power system static-state estimation, Part I, II, and III. IEEE Transactions Power App. Systems PAS-89(1):120-135.
Tinney, W.F., and J.W. Walker. 1967. Direct solutions of sparse network equations by optimally ordered triangular factorization. Proceedings of the IEEE - PIEEE 55(11):1801-1809.
This page intentionally left blank.