Munther Dahleh

William A Coolidge Professor of Electrical Engineering and Computer Science

Deep Thinking About Interconnections

Deep Thinking About Interconnections

Munther Dahleh is driven by real-world problems. He uses interdisciplinary approaches to bring together economists, engineers, computer scientists, and many others to address complex societal challenges.

By: Bernadette Esposito

Munther Dahleh is driven by real-world problems. As the William A. Coolidge Professor in the Department of Electrical Engineering and Computer Science, he is a member of MIT’s Laboratory for Information and Decision Systems (LIDS) and the director of MIT’s Institute for Data, Systems, and Society (IDSS), a landmark intersection of disciplines aimed at addressing challenges in interconnected systems. “IDSS takes rigorous and analytical approaches to complex societal problems,” says Dahleh, who defines societal problems as those in which the human factor, individually or collectively, is an integral part of the loop. “Many of the infrastructure applications or big, socially disruptive behaviors—whether you’re dealing with the power grid, financial systems, social networks, urbanization, or health analytics—share certain characteristics.” These characteristics often require the interaction of three pieces: the engineering system, the social system, and the institutions in which these systems exist. In the past these systems have been studied independently of one another. However, the newly launched institute, which has faculty from all five schools at MIT, supports a range of cross-disciplinary programs. “What we’re trying to do at IDSS is use interdisciplinary approaches that will bring together economists, engineers, computer scientists, social scientists, people from the business school—a collection of faculty from different parts of MIT to address complex societal challenges.”


Munther Dahleh
Please login to view this video.

The Power Grid

Among the most complex of these challenges is managing the power grid. It comprises multiple dynamic components including generation, transmission, distribution of energy, as well as consumers, and real-time markets. These together contribute to the challenge of having an efficient and resilient electric grid. For example, with a higher concentration of electric cars comes a greater demand for electricity. With the implementation of smart metering comes the ease with which future consumers can change their demand depending on the price. With ease comes an increase in the variability of demand. The grid is also getting upgraded to include renewable resources, like wind and solar electricity, which are more stochastic. In this new environment, matching supply and demand, as well as creating new electricity markets, becomes a serious future challenge. “Even though we think we’ve been improving the grid,” says Dahleh, “the number of outages and cost of outages has been exponentially increasing over the past decade. So, how can we continue to provide electricity for everyone who demands it 24/7 without outages and at a reasonable price?” Dahleh and his colleagues are looking at ways to simplify and address the challenge in a systematic way. The overarching question, then, becomes: “How do we develop models and control strategies to do this when we are talking about engineering systems, economic systems in terms of the market, social systems in terms of the consumer and still talk about the regulation of the overall system?”

Abstract Modeling

The fundamental approach to tackling these questions is abstract modeling. When Dahleh started teaching his first control theory class, the 747 was the most complicated airplane. “I walked into class and I told everybody, ‘the 747 is really just a second order differential equation,’ and everyone said, ‘Oh, no! That cannot be true!’” The second order system, he explained, is an abstracted model that captures the essence of the plane. “If you want to design a control system for the 747, you just need to know how to design a control system for a second order differential equation. You can then build on that control system to take care of the additional complexity the plane has.” But, how do you develop an abstract model when you have people as part of the system? While there are no Newtonian laws governing how people behave, says Dahleh, there are consistencies in the ways people respond. “Access to data on patterns of things people do and how people behave allows us to tackle complicated problems today that we couldn’t fifteen years ago. We have data on everything. We can measure now what people do in their homes. We can measure how much electricity they’re consuming. We can measure what’s happening in the market. If we rolled an incentive strategy in terms of pricing, we know exactly how people will respond to it.”

While data makes it possible to develop integrated models to address complex questions, it’s also immense and complicated and heterogeneous. And while data is accessible in the grid, it’s not as accessible in other applications. For example, in financial systems, data is a differentiator enabling each institution to make money by utilizing their private information. To address the challenging question of systemic risk of the whole financial system requires many data sets that are difficult to obtain from banks and other financial players, making it a serious challenge of the twenty-first century. “While many institutions don’t worry about cascade failures, because it’s something that happens once in a blue moon, they are all cognizant of the fact that it’s important to be monitoring the system,” says Dahleh. “Researchers have looked at data from financial systems and gotten interesting qualitative insights as to when cascades may occur, but it tends to be more historical than futuristic. For example, looking at the 2008-2009 crisis, we can say ‘this model explains what happened at the time’. We haven’t seen models that have predictive power, yet. How we go from complex heterogeneous data sets into simplified abstracted models is part of the big data science area that is emerging right now and one IDSS is tackling.”

Mitigating Network Failures

Another area Dahleh has been working on in the past ten years is developing a theory for measuring the proximity to failure of a networked system and finding ways to mitigate that failure through control strategies. “If you have a network of interacting dynamical systems, like transportation, the grid, financial systems, social revolutions, how do you stop it from cascading?” The approach, says Dahleh, is to understand what would trigger a cascade. In what mechanisms is it happening? Are there ways to monitor and then predict when a cascade is about to happen? In certain engineering problems, he says, this is very hard. “We have had several cascades occur—the 2003 failure in New England power grid is an example. We don’t have models that can predict cascaded failures or characterize the effect of the network structure on the fragility of the networked system.”

In terms of the power grid, the transportation system, or the financial system, there are two networks: the physical infrastructure and the information infrastructure. “We study the interaction of these two layers and their impact on the cascade failures. A transportation system is easier to model. There are surprises with how people will behave within the system, but there are no surprises in the physical layer. In financial systems there are surprises everywhere—with creditors, with how we lend money, and with various shocks that hit the system. There are surprises with how the banks work with each other and what decisions they make. You can regulate one part of the system, but then strategic players can push the system some other way.” Understanding the trade-off requires deep thinking about these interconnections. “We are close to understanding how transportation systems work. We have developed good theoretical models,” says Dahleh. “We are at the point of saying, ‘these types of models make a lot of sense. Why don’t we use them for design and synthesis?’”