This week’s journalist is Peter Yun Zhang, an M.A.Sc. candidate in the Mechanical and Industrial Engineering Department at the University of Toronto. He is currently working with the ATOMS lab.

This interview is about the wind farm layout optimization project at the ATOMS lab. In short, this research problem deals with the optimal placement of turbines in a wind farm such that the energy and noise objectives are optimized while respecting power and locational constraints.

Dr. David A. Romero is currently a Post-Doctoral Fellow at the Mechanical & Industrial Engineering Department, University of Toronto. He formerly held a position as Profesor Asociado in the Escuela de Ingenieria Mecanica, Universidad del Zulia, Maracaibo, Venezuela. David holds M.Sc. (2003) and Ph.D. (2008) degrees in Mechanical Engineering from Carnegie Mellon University.

David’s research interests are in applied computing, statistics and mathematics in support of engineering design, modeling and optimization, particularly in the thermal sciences. His previous work experience includes surrogate-modeling based optimization of thermal systems and of enhanced oil recovery methods, as well as consulting work in dynamic simulation of thermal/fluid flow systems and evaluation of wind energy resources. Current projects involve the thermal optimization of nano-scale devices with consideration of sub-continuum thermal transport effects, and optimization of wind farms.

What is the scope of the wind farm layout optimization project at the ATOMS lab?

The vision is to have a piece of software that can be integrated into the sponsoring company’s workflow, so that it automates the entire wind farm design process: design automation. For example, given the number turbines, types of resources, and some constraints such as maximum number of turbines and maximum connection power, what are the best layouts that we can produce?

What is the status of the literature in this field? What are the leading software packages?

Most software packages do energy optimization only, whereas our goal is to provide a multi-objective optimization tool that supports simultaneous energy and noise optimization. Most of them use heuristics and almost no one uses formal OR methods. There are some papers in genetic algorithms. Overall, there is a need for better documentation in the literature in terms of the experimental details so that future papers can better benchmark against current optimization strategies. In the past, genetic algorithms (GA) are used for a discrete version of the problem. We are applying GA to a continuous version of the problem and at the same time using some other OR methods.

Do we know how the software packages are implemented?

We don’t, except for OpenWind, which has an open source version. The documentation and theoretical development for some software packages in other fields (such as ANSYS for CFD) are very well documented and publicly available. There is a substantial gap between the wind farm design field and other more mature ones, especially in terms of commercial software’s technical documentation.

Who are the main players in terms of academic groups and companies?

DTU (Denmark) is a leading institution on this, and Europe in general is doing very well. For example, Germany has very high wind energy penetration rate. In Europe, there are some interesting studies about hydrogen-based fuel cells and electrolyzers so that wind energy can be stored and potentially transported. If you have distributed, small-scale storage things, then it’s very good for wind.

How is Canada doing in this field?

Regulations in Canada are fairly strict, and some health studies are still underway, it is hard to predict which way policy will swing. If negative feelings for wind energy out-weigh the benefits, we could see a slow-down in this field for at least a couple of years. Eventually we are going to have a mix of energy technologies. In the long term, there is going to be more development. That’s just my personal opinion. I think there’s a good opportunity in the US, because they have a lot of good resources, and the market is not yet saturated.

Overall, wind farm design, is definitely a multidisciplinary problem – mechanical, civil, environmental, electrical, policy, and optimization. How would OR specialists apply their knowledge in these kind of projects?

It has to happen either way – someone with mechanical/energy background picking up optimization methods (which is what I have done), the other way is for OR people to pick up energy literature. From the OR point of view, you just need a mathematical model to describe the behavior of the physical system. So if someone with an OR background works on this problem, it probably takes 6 or 8 months to learn about the mechanical and energy side of things. There is a learning curve, but nothing too much.

What are the types of OR problems in this context?

We have been studying this problem under heaviest k-subgraph and vertex packing problems, among others. And these are pretty much the only ones. The infrastructure layout optimization would involve minimum spanning tree algorithms. Traditional OR likes linearity, but this system is fairly nonlinear. But these nonlinearities are not badly behaved. For example, power loss is just a square function of the cable length – not a crazy nonlinearity. I think many things can be done in this project by using standard OR methods intelligently.

This week’s blogger is Shefali Kulkarni-Thaker, a graduate student working at morLab and is one of the moderators for this blog. You can write to her at shefali [at] mie [dot] utoronto [dot] ca.

Operations Research (OR) traces its roots around world war II and started out with the necessity to win the war. One of the first applications of OR was predicting the location of German submarines. Since then OR has evolved and is practically applicable to any field that is seeking for the “best” in a given situation. It could the least risky investment or the highest return value investment. OR has plenty of applications ranging from airline scheduling to farming. A typical OR problem will try to minimize or maximize a function under some constraints. It will try to answer question like: What is the best way to allocate operation theatres ? As you can imagine this has several dependencies like availability of surgeons, probability of ad-hoc surgeries, time required for a single surgery, patient priority etc.
Continue reading