30% jump in solar energy forecasting accuracy gained by “machine learning”

CleanTechnica

That old argument about solar energy being unreliable is getting weaker by the minute, and here comes IBM to knock the pins right out from under it. The company has come up with a Big Data approach to predicting the weather, and the result is a solar energy forecasting system that is up to 30% more accurate than the next-best conventional system. That’s huge news for utilities and other electrical system operators, because it helps them ensure a reliable supply of power while integrating more solar energy.

rsz_ibm-solar-energy-forecasting

Solar Forecasting By A Very Thoughtful Machine

IBM’s new system is called SMT, shorthand for “self-learning weather model and renewable forecasting technology.”

SMT is a kitchen sink approach that pulls together — for the first time — different kinds of forecasting systems:

It advances the state-of-the-art by using deep machine learning techniques to blend domain data, information from sensor networks and local weather stations, cloud motion physics derived from sky cameras and satellite observations, and multiple weather prediction models.

The “deep thinking” comes in as the system continuously cycles through a combination of real-time measurements and historical records, drawing on thousands of weather stations.

Don’t run out shopping for your very own SMT just yet, though. The system, which is a collaborative effort between IBM and the National Renewable Energy Laboratory (NREL) among other partners, is still in a preliminary phase. Scientists from the company and NREL will have more to say on the topic when they present their findings at the European Control Conference next week.

For that matter, if you check out IBM’s video, you’ll see that the collaboration is aiming for a 50% gain in accuracy, as well as providing a platform for all wind and hydro prediction, so might as well wait until version 2.0 happens:

What’s The Big Deal About Solar Forecasting

Conventional solar energy forecasting systems may be adequate for now, but that’s going to change. According to IBM, around 27% of the nation’s electricity demand could come from solar energy by 2050. A good chunk of that will be distributed solar systems, and that’s part of the problem.

For those of you new to the topic, distributed solar refers to relatively small arrays located on rooftops and other rather small properties. Until the smart grid becomes truly brilliant, it is difficult for utilities and other system operators to accurately gauge the output and demand related to these “behind the meter” arrays.

When distributed solar systems are few and far between, there is little effect on overall demand, but distributed solar is rapidly penetrating the market. Grid operators will in effect be flying blind unless more accurate forecasting models are available.

The grid operator ISO New England is also a collaborator in the new solar energy forecasting system. Here’s a snippet from its website, examining the issue in detail (breaks added for readability):

The relationship between renewables and the conventional resources needed to ensure grid reliability presents a conundrum: more wind and solar power creates a need for fast-starting, flexible resources that can take up the slack when the wind stops or the clouds roll in. New natural gas generators will likely fill this role, with their relative ease of siting and typically lower fuel costs—but this will further strain natural gas pipeline capacity…

Additionally, wind, solar, and other forms of “green energy” with low fuel costs could make conventional resources—resources that make up the majority of regional generation and that are crucial for grid reliability—less profitable in the energy markets.

This will likely cause the capacity market to become a more important revenue stream for conventional resources and future storage resources. Ensuring appropriate long-term price formation in the capacity market will be vital as the region gradually adds more renewable energy to the mix.

In the context of natural gas impacts (including transportation and storage issues as well as fracking), the overall benefit of more accurate solar forecasting is clear. You get a threefer: a more reliable supply of power, more efficient use of solar energy resources, and less reliance on fossil power plants.

For New England specifically, the ripple effect also includes a reduction, if not the elimination, of the need to build disruptive new gas pipelines and storage facilities.

As for the Energy Department’s involvement, the SMT project comes under the agency’s SunShot program, which aims to bring the cost of solar energy down to parity with fossil fuels.

SunShot is pushing hard for an increase in distributed solar energy, and SMT is part of a package of initiatives that provide solutions to the issues raised by these “behind the meter” systems. Energy storage, smart grid, and microgrid solutions are also on the table.

What Is The European Control Conference?

If the idea that the US military would invade and take over the State of Texas doesn’t phase you (Jade Helm, much?), then you might have a problem with representatives of the US government flying across the pond to participate in a conference that includes “European” and “Control” in its title.

The European Control Conference is held every two years under the auspices of the European Control Association, which was organized in 1990 t0 do this:

…promoting initiatives that enhance scientific exchange, disseminate information, and coordinate research networks and technology transfer in the field of Systems and Control within the [European] Union.

I know, right? That’s not the really sketchy part. Wait ’til you see what the conferences are all about:

…The goal of these conferences was to promote the science and technology of control in the broadest sense – whether in engineering, physical, biological, social or economic, and in both theory and application.

Social or economic control!? If anyone out there can make a connection between Jade Helm, solar forecasting, and European Control, have at it in the comment thread.

Source: CleanTechnica. Reproduced with permission.

Comments

One response to “30% jump in solar energy forecasting accuracy gained by “machine learning””

  1. Chris Fraser Avatar
    Chris Fraser

    Now that’s a smart grid. This gadget should be in every suburb with distributed PV on rooftops.

Get up to 3 quotes from pre-vetted solar (and battery) installers.