Lessons learned en route to the Edelman

Member of award-winning team from key vendor offers his observations on prestigious prize process.

“When you cannot make up your mind which of two evenly balanced courses of action you should take, choose the bolder.”
- William Joseph Slim

Midwest ISO President and CEO John Bear

Midwest ISO President and CEO John Bear

By Eugene J. Zak

On the night of April 11, INFORMS President Rina Schneur opened an envelope containing the name of the 2011 Edelman Award winner and announced, “Midwest ISO.” It was an exciting moment for the members of the Edelman-winning team and culminated a long journey from the initial specification, design and development of the complex electricity market management system, as well as its deployment and successful operation the past four years.

The Edelman Award committee decided that the project – a joint venture between Midwest Independent Transmission System Operator (Midwest ISO, the client organization), Alstom Grid (the prime vendor and my employer) and others – satisfied two of the most important criteria for the award: significance of impact and the application of advanced operations research methods and tools. Midwest ISO reported $2.1 billion to $3.0 billion in cumulative savings between 2007 and 2010. Alstom Grid contributed to the advancement of operations research by creatively modeling the large-scale unit commitment and dispatch problems for a combined energy and reserves market. Innovations included the use of customized cuts and tight model formulations to improve performance.

The award goes to the client organization – in this case, Midwest ISO – but, like most Edelman winners, the benefits Midwest ISO realized were the result of collaborative contributions from several individuals, teams and vendors, including Alstom Grid’s modeling, implementation and deployment of software to the Midwest ISO servers. The servers run about 300 times a day solving complex unit commitment and dispatch models, fueled by data from nearly a thousand generating units. The typical mixed-integer programming model size is 3.9 million constraints, and 3.3 million continuous and 450,000 binary variables. The running time does not exceed 20 minutes on the 2.93 GHz 64-bit Intel processor.

Lessons learned en route to the Edelman

The project and the award highlighted the key role operations research plays at Midwest ISO. Indeed, O.R. is all about applying advanced quantitative tools for solving complex business and operational problems. During a team victory reception, Midwest ISO President and CEO John Bear told me, “Everything we are doing in Midwest ISO is operations research. Our main task is to determine how to effectively transport energy from suppliers to consumers.”

Having the company president and CEO so involved in the Edelman project and presentation obviously impressed the Edelman judges, one of whom told me, “Not only John Bear’s personal presence at the conference and his flawless presentation, but also his leadership and knowledge of the business side of the project made a difference.” Bear’s credibility and ability to answer the judges’ questions well confirmed that O.R. was critical to the project success.

So what lessons did the team learn from its quest for the Edelman Award? I can’t speak for them all, so here are some of my personal observations:

1. Challenging problems create great opportunities for O.R.

My earlier experience was in the process industry, the birthplace of O.R. for solving manufacturing problems such as the cutting-stock problem. My impression at that time was the process industry is one of the few places where O.R. works the best.

After moving to the energy sector I realized that the very nature of the electricity commodity makes the subject area intimately fit the O.R. methodology. The instantaneous balance between supply and demand, the physical laws of alternating current, the geographically distributed generation and consumption, and a centralized financial market are all factors that result in the unique set of challenging large-scale optimization problems. Moreover, a solution failure in the real-time scheduling can jeopardize a reliable energy flow across a big geographic area.

2. Understand the difficult task every team faces when it applies for the Edelman Award.

Though the final Edelman ceremony mimics an Oscar ceremony, the Edelman paradigm is quite different: Judges do not observe the real system in action. They observe the information about your system collected in your presentation, your paper, testimonies, etc. When you buy a car you may choose to test-drive it first. In the Edelman competition, there is no test drive. The judges cannot come to the control room to take over the operators’ duties for the electricity market management!

So, the Edelman decision-making process is based upon information that is actually a reflection of the real system you designed, developed and deployed. Your team’s task throughout the Edelman process is to adequately map your system into the space defined by the Edelman competition conditions. Two of the most important criteria are impact and O.R. innovations (Figure 1.)

Figure 1: Mapping your system into the “Edelman space.”

Figure 1: Mapping your system into the “Edelman space.”

3. Never give up. If you have a “gut feeling” that you are right, keep going.

All the odds were against us in October 2010 when we applied for the award. Still, something told me that we could succeed. First, we had encouraging reviews after our first attempt at the Edelman Award failed in 2009. Some applicants perceive “not this year, please come back,” as a polite rejection and do not reapply; when the Edelman committee says, “please come back,” they really mean it. Secondly, the latest phase of the system – a combined energy and reserves market – had been functioning flawlessly and effectively for almost two years. And finally, we had done some additional computational analysis of the unit commitment that is a central part of the modeling challenges. This analysis confirmed recent interesting theoretical results on the subject.

Midwest ISO’s policy is not to endorse any vendor. A crucial turning point in the team’s success occurred early on when John Milne, at the time our Edelman verifier, convinced a Midwest ISO executive to not only support the Edelman effort, but to take the lead in the application and competition process.

4. Elevate the competition process to a project rank. Allocate sufficient resources for the project success.

The situation is binary: Either compete with all necessary resources or do not bother to start. I cannot say that we had everything we needed. Still, I was awarded two weeks of non-billable time for working on the paper and presentation (and probably spent an additional 80 hours of evening/weekend time on this).

Collaborative work is essential. That is why it is very important to organize the work effectively. Consider using source control software for keeping the latest version of your paper and presentation and tracking all changes. I admit we learned this and related lessons the hard way, but fortunately for the team the deficiencies in our collaborative working environment were compensated by extraordinary efforts from individual team members, including our coach (and erstwhile verifier), John Milne.

Lessons learned en route to the Edelman

5. Don’t cut corners. If you find a conceptual or technical bug even in your final version, fix it.

On the night before our presentation I noticed a mistake in one of the slides. Let me explain: When the energy demand and supply are in balance the electrical equipment is tuned to support a nominal 60 Hz frequency. However, if the demand exceeds the supply, then without an additional generation the frequency will go down (to 59 Hz, for example). The reverse is also true: If the supply exceeds the demand then the frequency will go up. Our initial slide showed a symbolic balance scale with the left pan as “Supply” and the right pan as “Demand.” In such an arrangement the larger demand (rather than larger supply) caused the higher frequency. We immediately agreed that it needed to be fixed, and the change was as easy as swapping tags over the pans. We could easily have said that “no one would notice” and “it is not worth it to make a change,” but how would we have looked if we did not fix this error?

6. Crown your achievements by writing a paper, giving a presentation or going bolder – applying for the Franz Edelman Award!

Participating in the Franz Edelman Award competition is a great experience. You will meet amazing people along the way and gain a greater appreciation for the incredible work done by INFORMS. I do not intend to discuss the eternal question about “bridging” the gap between theory and practice, between academia and industry, but I would like to note that the Edelman Award plays a fantastic role in promoting O.R. in the real world.

Looking back, I am still amazed how a seemingly small binary decision – to apply or not to apply for the Edelman Award – can affect so many people in a positive way! We know a small variation in the input data sometimes can cause big changes in the output, and though it is not always desired in optimization or control theory, it happens in real life.

While the client, Midwest ISO, is the official Edelman winner, all the individuals and organizations that contributed to the team feel like winners. I express my deep admiration for the Alstom Grid participants including former employees Dan Streiffert and Russ Philbrick, the Paragon Decision Technology experts, the very helpful consultants from Utilicast, and of course my friends at Midwest ISO. Though IBM-ILOG with their superb customer support was not a part of the team, their optimization engine, CPLEX, provides the main horsepower behind the models.

We all are grateful to our coach John Milne of Clarkson University.

Eugene Zak (eugene.zak@alstom.com) is principal engineer for Alstom Grid in Redmond, Wash., and a member of the Midwest ISO team that won the 2011 Edelman Award.