POWER > Trends > IBM Research

The Watt-Sun Program Is Poised to Improve Weather Forecasting for Solar Energy


A 20 megawatt solar farm in Tucson, Arizona, is one of the test sites for Watt-Sun.

Q. Why are clouds so difficult to predict? Is it their makeup? Their density?
A.
The density, why they form, why they dissipate and so forth. Then of course, the sun’s radiation has to transfer through that cloud layer, and that involves very complex physics. It’s a multiscattering process where the light is being scattered by ice particles or absorbed by water particles. All of that needs to be exactly accounted for when it comes to solar. So weather modeling is already complex, but now we have on top of that radiative transfer forecasts.

Watt-Sun can forecast anywhere from 15 minutes to 1 month in advance

Another challenge, which is rather more of a practical issue, involves big data because there are so many more measurements involved in what we’re doing than in traditional forecasting. But you can actually come up with much better assessments as to whether forecasts are good or bad because there’s much more data out there. As you know, making sense of that big data, whether from weather stations, satellite images and so on—which could be tens of terabytes a day at a minimum—to improve both short-term and longer-term forecasts is easier said than done, so we’re looking at leveraging it in a smart way.

Q. In what way is your approach smarter?
A.
Well, we can now think about new approaches to forecasting—or solar forecasting in this particular case—where we actually use different forecast models that are being produced by governments and universities around the world to understand which one of those forecast models—each with somewhat different assumptions and approaches to the physics—works best under different weather situations.

Spinning Reserve refers to power plants put on standby in case of energy generation fluctuations or load changes

This is the basic thrust of what we are doing here under the Department of Energy project. We’re assimilating all of that data out there to blend different models, to find out which of those models does the best job or has done the best job in the past under which weather conditions—low pressure, high pressure and so forth—and then use that to come up with an optimal mixture of these models to provide a so-called “supermodel.” That’s the bottom line of what we’re trying to do here.

Q. The name of this project is Watt-Sun, a play off Watson. If at all, how does Watson technology factor into this?
A.
We’re adopting a mixture of approaches to create an adaptive blend of expert systems that can somewhat easily come up with an optimal number. They don’t have to be just weather models but also include different types of forecasting techniques. That’s the approach we’re using for this. Yes, it is a machine-learning approach, but there are different techniques involved in this. In our case, we’re dealing with numbers and blending them. This approach involves constantly learning from and improving forecasts.

Q. How timely are the forecasts at this point?
A.
Typically, grid system operators have day-ahead scheduling for what they call the unit commitment process, as well as 15-minute-ahead scheduling, so you need to produce forecasts depending on what the application is. There are technical reasons for why this is split into day-ahead and 15-minute-ahead scheduling, which are akin to, say, having the correct number of computing resources available in a data center. You have to make a decision on how many machines you need for a job because it will take some time to boot them. Then, as the workload changes, you can maybe idle some of the machines and rent them out or shut them down.

The same thing is true for the power grid. You have to make a decision a day ahead as to which plants you’re operating. Once you’ve made that decision, some of your other plants will be idling in case the load picks up. If the load does pick up, you have to be ready to ramp up the idling plants to meet demand. That’s essentially why we forecast anywhere from 15 minutes all the way to a month ahead, so grid operators can be prepared for potential loads from solar-energy sources.

Q. Will this type of forecasting save grid operators money?
A.
There’s a term in the industry called “spinning reserve,” which means some power plants exist only because the generation or the load might change—and of course having those spinning reserves costs money. This is like a computer. Even if it’s idle, it still uses power.

So the bottom line is flattening that out so when there’s renewable energy available, we can take full advantage of it, with the grid operators being aware of it in advance and saving money by perhaps only firing up reserve plants when they’re for sure needed. That’s what forecasting buys you.

Jim Utsler, IBM Systems Magazine senior writer, has been covering the technology field for more than a decade. Jim can be reached at jjutsler@provide.net.


comments powered by Disqus

Advertisement

Advertisement

2018 Solutions Edition

A Comprehensive Online Buyer's Guide to Solutions, Services and Education.

IBM Systems Magazine Subscribe Box Read Now Link Subscribe Now Link iPad App Google Play Store