Predicting the Clouds

New research at the University of Texas at San Antonio is helping to predict clouds and weather patterns that affect solar grids.


New research at the University of Texas at San Antonio is helping to predict clouds and weather patterns that affect solar grids.

We go inside the lab for a look.

On a clear day, it's very easy to predict what's called global horizontal irradiance, basically solar irradiance that feeds the PV arrays and produces electricity.

The initial technology that was used at [Speaks indistinctly] in order to forecast solar irradiance consisted of a TSI-880 Sky Imager, which consisted of a camera on top of a hemispherical dome.

In its day, which would have been, say, five years ago, this was a good piece of equipment.

The black band here is called the shadow band, and that's necessary in order to keep the Sun from washing out the entire circumsolar region, and so what happens is there are complicated collection of gears and motors, and this turns and tracks the Sun so that the Sun is always shining here.

What you do is you get a good image around the shadow band, but of course you lose information there.

Now we're going to move to the next technology that was developed at UTSA for the solar-forecasting problem.

This is the next generation of sky-imager technology.

It consists of a security camera enclosure that houses a Raspberry Pi single-board computer and the Pi camera that actually takes the images.

In this particular version, there is also an ODROID-C1 computer for additional computational power.

We have the Pi NoIR camera.

Now, this is new from the previous camera that we were using.

This is going to allow us to look at more things when we actually start taking the pictures.

You also have the fan to make sure, you know, Raspberry Pi keeps its temperature.

And now, this is really cool.

this is a weather-board.

Now, the weather-board, I've blown up kind of what it has up here, right, and it lets us see the temperature, humidity, the pressure, and the altitude.

So this -- I think it was $20 -- piece of technology is just out of this world 'cause it tells us all this stuff and it makes it a lot easier to use these cloud predictions because we can actually look at all of the atmospheric conditions around it, as opposed to just seeing the picture.

So you see that it represents a marked improvement over the earlier technologies.

This was developed at UTSA.

And there is currently one U.S. patent and a patent in China that have been applied for for this technology.

There's a hardware side to this problem, and there's the data that is taken by the actual UTSA Sky Imager.

So that's what I work with, analyzing the data, seeing what's going on.

Analyzing all this data, what we're going to do is use machine learning to be able to predict GHI.

This is just a small structure of machine... The type of machine learning we're using is neural networks.

So in your input layer, you would have all this data.

In this case, we have our images and also various different parameters that we talked about that the weather-board is able to capture.

So the neural networks has a very amazing ability to, if you feed it data, it finds these patterns found within the data.

And it could feed out your output layer, which is what you're trying to predict.

In this case, we're trying to feed it data in order to predict our GHI.

And if we can predict GHI, we can predict PV power, photovoltaic power output, and that's what's useful to utility companies, to the customers themselves, who own solar panels.

And yes, this is just the structure of what we have.

If we can accurately forecast these drops in power, irradiance, the ramps, then we'll have a good handle on achieving what's called dispatchability of power.

In other words, we'll be able to manage and control the electric grid in a way that's optimal.