Operations Research in Energy and Environment: An interview with Dr. Pascal Van Hentenryck

Could you give me a quick synopsis of your research in the field of energy and OR?

We started working on this topic when we looked at how to restore power after hurricanes. It is a very challenging problem because it combines logistic issues - sending the crews to pick up parts and repair components - with finding the right schedule for restoring the electrical power system.  We recognized that if you use the traditional approximations in power systems, like the DC (Direct Current) model, you encounter difficulties in converting the approximated solutions into actual physical solutions, i.e., solutions that satisfy the AC (Alternating Current) power flow equations. It was interesting because we were not optimizing the standard objective function. Rather we were trying to push as much energy as possible in the network and we ended up stressing the network. And that is when you hit the non-linearities in the power systems. So we decided to work directly with these non-linearities and find good approximations in order to obtain high-quality physical solutions. Eventually, we got passionate about this area, which has plenty of challenges for optimization.

How did you get into this line of research?

I was at Brown University when we started this research in collaboration with the Los Alamos National Laboratory. They had a lot of interesting and realistic data sets, which allowed us to make progress, test our algorithms, and scale them to large test cases. We continued to focus on power systems when I led the optimization research group at NICTA in Australia. We started looking at many more problems and novel relaxation and approximation techniques. Currently in Michigan, one of our big goals is to build a large collection of test cases that are more challenging than the ones available currently. You need to understand that about 5-6 years ago the test networks were easy to solve. The systems become difficult to manage primarily because of contingencies and the challenge lies in recreating these conditions and modeling the networks with a much higher fidelity. This is really important to facilitate the work of the academic community. 

You mentioned hurricanes? Could you provide me with a little more info about how this plays into the energy optimization field?

Hurricanes can knock down significant parts of the power system. Once that happens, the goal is to restore the power system, e.g., the lines that were damaged, so that the size of the blackout is minimized. If you recall, over the last few years, a few hurricanes knocked down significant parts of the power system on the East Coast of the United States. People were often out of power for 5-6 days. This is a very difficult application in comparison to the basic optimal power flow problem. On the one hand, you are not trying to minimize cost but rather to push as much power as you can through the system. This puts your system under stress and this becomes interesting from a technical standpoint, since the assumptions underlying the traditional approximations are violated. You also need to sequence the repairs, which leads to a mixed nonlinear nonconvex optimization problem.

I am aware that a lot of what you're saying ties in a little bit with humanitarian logistics.

Oh, absolutely, I think the intersection between humanitarian logistics and energy is very interesting. Blackouts are happening repeatedly now when a significant part of the system experiences severe weather. It is a real ordeal for people living in the area, who are often without power for several days. It is humanitarian logistics combined with complex infrastructures.

In your opinion, what is biggest challenge in this field currently and what are the most exciting future aspects? Could you pin those down for me?

I think one of the biggest issues is with the introduction of renewable energy. You're moving away from a system that, to a very large extent, is predictable. It is a system in which you can control the generators and in which the consumer demand is predictable from an aggregate standpoint. With renewable energy, the system becomes more complicated because while on the one hand the generation is less easy to control: you cannot choose how fast the wind is blowing or when the sun is going to shine; on the other people may now have solar panels on their roofs, electric cars to charge, and batteries to store energy. This makes the load more difficult to predict. So you're moving from a largely deterministic to a very stochastic setting and this immediately makes life a lot more complicated. Another challenging aspect is the power electronics that connects renewable energy sources to the grid. It changes the way the control system underlying the grid operates and it makes it harder for operators running the system to understand what the system does. This introduces new challenges and it becomes increasingly important to model the system dynamics within the optimization framework, which is not easy to do. 

I'm supposing that this also becomes very interesting for future researchers - the way you control and manage renewable sources of energy from an optimization standpoint. 

Oh, yes, in my opinion, this means a variety of things. One is getting better algorithms for dealing with the power flow equations and modeling the control systems. You basically must model the system to capture the non-linearities and the discrete aspects of the system better. Moreover, you should be able to do so within a stochastic optimization framework, which introduces another layer of complexity. These systems are also very large and you need to ensure that your methodology is robust. The other aspect to pay attention to are numerical difficulties that solvers encounter when faced with large-scale networks, Finally, the use of high-performance computing is another interesting challenge for power systems. 

Indeed. And speaking of large datasets, you mentioned earlier that one of the things you're working on is collecting a large set of test cases. Could you delve a little bit into what that entails?

Well, when we first began working in this area, the datasets were usually small. The community now has progressed to the extent that many of these problems can be solved optimally and aren’t challenging anymore. Moreover, they don't represent the complexity of real power systems. So, what we want to do is to provide the academic community with the next generation of test cases that represent the real challenges that the community should solve. We would like to model a real system at different voltage levels and with the fidelity at which these systems are operated in practice. So not only are we scaling the sizes of these networks, we are also scaling the fidelity - the way we model the components and the way the systems are being operated in practice. This should present the academic community with challenges for the next couple of years. 

So is this data being obtained via collaborations with national labs and universities?

It is difficult to obtain this data in the United States. We collaborate with RTE  (Réseau de Transport d'Électricité), the French Transmission Operator. This is the largest transmission operator in Europe. They have an amazing research team internally and are very interested in looking at the tools that the academic community is producing, and because of this, they provide us with realistic data. Of course, they ensure that the privacy of the data is preserved and that we can't reverse-engineer the sensitive aspects of the network and its operations, but they're still realistic networks, which are very valuable from an academic standpoint. 

Do you have any advice for students wishing to get into this line of energy/power systems research?

What I normally tell people when I talk about energy is that this is one of the biggest success stories of optimization. It is a lot like the airline industry using optimization except that the success of optimization in the power industry is not as well-known. Moreover, almost all the algorithms that have been designed in optimization are being, or have been, used in power systems including linear programming, Lagrangian relaxation, mixed integer programming, stochastic programming, and nonlinear optimization. This is an amazing field to push optimization research forward both from a theoretical and computational standpoint. At this point, the community has made so much progress that understanding the issues is much easier than it was 10 years ago. Electrical engineers have also done a great job of abstracting power systems at various levels. Once you understand these abstractions well, it becomes easier for us in the Operations Research community to penetrate the field. It is a great time to enter this area since the scientific publications are now at a level that makes it easy enough for people to get up to speed quickly. If we can produce the test cases that I talked about over the next year or so, the academic community will be in a very good position to achieve a lot of progress. Essentially, I would recommend that youngsters invest a little bit of time to understand the physics and the electrical engineering aspects, which are fascinating in their own right. Once they have assimilated that, they will be in an ideal position to deploy their knowledge of optimization technology. At the end of the day, any contribution one makes in this field may have a direct and lasting impact on the industry. 

All right then. Let me close out with a final question that I usually ask people. What is your biggest non-academic hobby?

I played a lot of soccer when I was a kid. I'm a big sports fan and that is what I spent most of my time doing when I was growing up. I slowly stopped playing competitively as I got to the end of my undergrad studies. These days I run and bike a lot.  I grew up in Belgium and their national soccer team is very good these days. I kept playing with undergraduate and graduate students when I came to the US, but it gets more and more difficult as time goes on. You lose your speed as you age unfortunately.

Thank you so much for doing this interview with me! I’m sure our readers will find it extremely insightful.

Thank you for talking to me. It was a pleasure.