‘Here’s my profit, I’m going to be able to pay back the battery, can you give me a loan?’: GridBeyond on optimising BESS and AI modelling

As such, AI has become a popular tool for its quick learning capabilities. Rather than spending months or more researching the historic trends of a particular market, the data can be fed into a model. That model’s response can be checked for accuracy before being used for more advanced applications.

For example, some optimisers using AI are Habitat Energy with its Enertel AI, Gridmatic and Australian energy management company OptiGrid.

McEvoy sums up the relationship between AI, optimisers and operators well, saying “When a person is getting financing for their battery, they can take your (AI modelled) report, they can go to their bank and say, over the next 15 years, this is how much I’m going to make in selling the energy into the energy market in Texas, this is the cost of my battery, here’s my profit, I’m going to be able to pay back the battery, can you give me a loan?”

GridBeyond is expanding into markets across the globe using AI and a successful history of providing demand response and battery management services.

GridBeyond’s demand response and battery management operations

GridBeyond was formed 15 years ago, and its primary source of business for approximately 10 years was demand response. The company provides advanced notice of power-down periods through their forecasting systems that predict price, climate, and temperature patterns.

McEvoy explains, “When you’re in Colorado, for example, and it’s super hot, and people come home from work and everyone’s turning on the AC, TV, cooking, washer, dryer, charging EV cars, you’re consuming a lot more from the grid, and the demand is going up and the energy prices go up.”

“Our demand response business helps alleviate stress on the grid by curtailing a lot of large energy consumers in the commercial and industrial (C&I) space, at the same time, saving money for those large C&I because they’re avoiding those peak costs.”

McEvoy says the demand response focus worked well for GridBeyond, but eventually, the company encountered companies and whole industries that did not have the same flexibility to ramp down production and quickly restart it at a moment’s notice.

“If you’re a glass manufacturer, you can’t turn it off. You just can’t. You’re done. It’s going to take 12 hours to get all your glass liquefied.”

To resolve these issues, GridBeyond began offering these companies batteries, ranging in scale from 1MW to 50MW. The batteries are charged from the grid when electricity is cheap and dispatched when demand is high.

“We make sure the batteries are full, then we leave them there. Then we’re monitoring again when the high peak is going to be during the day. We know this customer can’t turn down, but what we do is we start feeding that battery’s energy into the plant so it’s not taking more than it needs from the grid at the high price.”

He continues, “Batteries have started to become almost a must-have for your large C&I customers. And so we do this now in Japan, we do it in Australia, we do it in the UK and Ireland and the US.”

From there, McEvoy explains how the company continued to grow, with battery suppliers needing battery management services to help manage the grid and to generate revenue in the energy wholesale market.

“This Big Rock site is an example of that, where company Gore street came to and said, ‘I got 200MW in San Diego of a battery, I only want to put 100MW in to help balance the grid for the California grid, but I want you to use the other 100MW to trade it in the energy market every 15 minutes, real time or day ahead, to help me pay for this 200MW battery,’ which probably cost them US$300 million.”

Over the summer, GridBeyond started optimising GSF’s 400MWh Big Rock BESS in El Centro, California. The company recently explained that launching the project involved a large team of more than 20 people working across different countries and time zones, comprising data scientists, regulatory analysts and industrial control engineers.

“We have a trading team that sits in a knock centre full of traders. 24/7 they’re trading energy in Japan, Australia, the UK, Ireland and the US.”

“We are predicting the high and low prices in the California market. We’re predicting weather exactly where the battery is located. At the Locational Marginal Pricing (LMP) node, we’re predicting the price at that latitude and longitude, where that battery is, the weather at the longitude and latitude where the battery is, the congestion and any of the lines surrounding it, and then determining what the price is going to be.”

The company uses all of that data to compile a bid to submit to the California Independent System Operator (CAISO) market.

“What you submit to CAISO is ‘P’s’ and ‘Q’s’, prices and quantity. You get your quantity from how much is in the battery; the price is the hard one to predict. A lot of our own proprietary AI tools predict the price. You submit those two things into the CAISO energy market and they either accept them or they don’t.”

McEvoy explains, “You need to be really smart in how you submit them at the right price, with the right quantity, for the right time, when you know there’s going to be stress on the grid, and how CAISO is going to accept your bids and offers, Then, they’ll tell you, we’ve accepted you, and you’ve got to dispatch that energy from the battery into the CAISO market exactly when CAISO needs it.”

Market Comparisons

McEvoy highlights Gridbeyond’s variety of markets and the amount of data needed to obtain a bid for each market. This is no small task, given the differences between markets.

“There’s less regulations in the Electric Reliablity Council of Texas (ERCOT) market in terms of being able to site a battery, deploy a battery, and work with the ERCOT energy market. ERCOT is easy. ERCOT makes it easy for you to work with them, and they have accelerated programs.”

“If you have a battery that’s 9.99MW, you can connect it to the grid much easier than anywhere else. We call them Texas 10s.”

Available Power’s VP of business development, Alex Krass explained Texas 10s at the Energy Storage Summit USA in 2023:

“You can skip a part of the interconnection process with ERCOT (with projects under 10MW) so your time to market is quicker, which is fantastic.”

Due to the faster interconnection, many developers in ERCOT have opted for smaller, ‘distributed’ projects under 10MW.

Energy storage platform GoodPeak recently on two 10MW projects in Texas, for example.

McEvoy continues, “ERCOT has a whole bunch of frequency programs, like energy programs and peak programs, so you can make money in lots of different ways in ERCOT.”

“Because ERCOT has such favourable terms on getting connected to the grid, more opportunities to make money, it is getting saturated with batteries, like absolutely saturated. So, when you have an increase of batteries coming online, with lots of energy available to dispatch, the prices have been dropping and dropping and dropping.”

In February, ESN Premium spoke with energy software and consulting group Ascend Analytics CEO, Dr. Gary Dorris, PhD, on ERCOT’s market saturation, with Dorris noting:

“When markets saturate, ancillary service prices don’t drop to zero, but they become more stable. You get to a point where energy and ancillary services are economically indifferent, which reduces the potential for large revenue differences. Still, there will be volatility in both energy and ancillary prices due to the growing share of renewables, which brings more volatility into the market.”

Moving from ERCOT to CAISO, McEvoy says, “California is more constant in terms of you know what you’re going to get. It’s harder to get your battery into California, which means fewer batteries make it online, because you have to jump through so many hoops, which means it costs money.”

He continues, “California is a good place to put a battery. It’s more stable, it’s steadier. It will cost you a little bit more and take a little bit longer to get your battery online, but you will probably make more money than you make in ERCOT.”

Two of his US colleagues at GridBeyond, Ali Karimian, director of market optimisation and Alden Phinney, regional director, recently wrote a Guest Blog for this site on the changing market dynamics of CAISO and ERCOT BESS optimisation, taking a deep dive approach to show how optimisation strategies can impact revenues and return on investment.

Beyond the US, GridBeyond has recently entered the Japanese market with its partner, contract support services provider Port Inc. In June, GridBeyond began energy trading from a 2MW front-of-the-meter (FTM) Port BESS in the prefecture of Gunma.

McEvoy says that Japan is similar to the US in the number of energy regions available to trade, but because the BESS market is still relatively new in the country, smaller projects make much more money than they would in the US.

“Japan is offering very big money in certain markets for a MW of battery storage. You’re talking minimum, it goes from about US$200,000 to US$500,000 per MW.”

He explains, “The problem with Japan is that in the areas where it is constrained, like the cities, there’s no room for big batteries. In Tokyo or Osaka, for example, space is very confined. So, the batteries are always smaller. They’re always around 2.5MW.”

“However, putting a 2.5MW battery in the market in Japan is the same as putting a 20MW battery into ERCOT; you’ll get a similar amount of money.”

AI Modelling

Building a large language model (LLM) has helped GridBeyond make successful bids in the US and has been a helpful tool in making the move into Japan.

McEvoy explains how the company creates an LLM for each of the markets it operates in. The company tries to find price data, looking at prices from every minute of the day for three years. Then, it looks at the weather conditions and energy demand from the same period.

“You start to build an energy model that can actually replicate what happened, such that when it encounters data that it’s trained on, it’s like, ‘Oh, I’ve seen that before. This is what’s going to happen when the demand hits here and the weather pattern is here.”

“We start to predict out every 15 minutes, half hour, hour, daily, weekly, up to 15 years, we’re predicting out, long-term, because that’s the life of a battery, 10-15 years.”

McEvoy further details how, for C&I clients, the company can build a “digital twin” model of a cement plant or pulp and paper plant, for example, to better predict demand response.

“We know how it runs, we know how much energy they’re going to consume at any time. We know what the prices are in the area, so with an extremely high level of accuracy, using a digital twin of a physical industrial plant, we can update them, save money and de-stress the grid at the same time.”