Learned Behavior: Advances in machine-learning lead the way to true solar + storage profitability

light bulb illustration

In the compelling short story “Lifecycle of Software Objects,” author Ted Chiang explores a world in which artificial intelligence (AI) is reared and cared for just as you would a child. He did this because he sees the commitment life-long learning and development as the only true way for AI to ever achieve conscious intelligence. Point being, there is only so much pre-programming can accomplish.

Advanced energy control software companies would agree with this philosophy. It’s not exactly full consciousness, but today’s energy storage controllers are using various machine-learning and artificial intelligence to make savvy decisions that balance priorities of asset generation, optimal storage use, tariff schedules, demand charges and miscellaneous grid responsibilities. For C&I solar + storage projects, these abilities aren’t a bonus, they are crucial to maximize the assets and hit key production and revenue goals. The main questions at hand here:

  • What’s the electrical tariff for the site, and how should the system maximize the cost avoidance or value captured under it over time?
  • What charge and discharge schedule for the asset makes sense for that moment in time?

Misjudging the answers to these questions is likely to be a $100,000 mistake or more in some cases.

Thinking fast and slow

Pason Power provides one of the top advanced energy storage control solutions on the market, and we asked Bryce Evans, head of customer and partner success, to give us a peek into how these systems think.

Pason Power’s autonomous control systems and machine-learning capabilities emerge first from two broad buckets of information processing — micro and macro — which can have conflicting agendas (all humans begin to nod).

Micro adaptations. Every second, the Pason system gathers data from a variety of real-time data sources — from the energy storage system’s power conversion systems (PCS), battery management system (BMS), meters at the site (for the current building load, on-site solar generation and sub-metering on machinery or sub-circuits), and other instrumentation data such as temperature sensors. It’s also consulting weather reports, cloud characteristics and maybe a weather station or irradiance sensor at the site (if applicable).

All of that is driven into a patent-pending forecasting engine, which produces two forecasts: the expected site demand over the next 24 hours and the expected output of the solar array. This is where the learned responses prove their worth as the system considers the forecasts within the context of various site constraints.

/*** Advertisement ***/

“Our system responds in real time to the instantaneous changes in site load among other factors and generates new forecasts every 15 minutes based on all of the information gathered,” Evans says.

Macro adaptations. Over a longer period of time (days, months and years), a model will need to be adjusted within the broader context of the evolving load of the site, potential changes to rate tariffs and new revenue stream opportunities. Pason handles the evolving energy usage patterns at the site by periodically deploying new AI models that are retrained to include the most recent gathered information from the site’s measurements. Most changes to rate tariffs are usually handled automatically by the AI control system, which has the capability to optimize battery operations for thousands of rate tariffs. Pason also deploys updates to the machine-learning model to explore other ways to best produce savings for this site given what’s been learned since the system was commissioned. This process is enabling continuous AI learning.

A primary application of C&I solar + storage is demand charge reduction. The most basic way to do this is to look at historical electrical consumption of the facility and configure the system with a fixed threshold. If site demand ever goes above X, the asset discharges. That requires no machine-learning sophistication.

But, what if you have a really hot month (an increasing likelihood in the age of global warming)? A fixed threshold approach might completely miss the peak demand charge.

“You might start to discharge the asset too soon because the threshold could be too low for this particular day,” Evans explains. “The HVAC system is going to run longer than usual because of the anomalous month, and because of how demand charges are billed, if you let one through the whole billing period is sunk from a cost avoidance standpoint.”

An adaptive system can assess such situations with more nuance.

“We see HVAC loads as a sweet spot for our system, which are highly correlated with weather,” Evans says. “We’ll be able to anticipate that increased need and be less aggressive in our demand charge reduction, counter intuitively, because raising the threshold makes sure we effectively shave the peak. If you discharge too soon, you miss part of the peak, and there’s no changing that now for the rest of the billing period.”

Consider the needs of others

Then there are the batteries themselves. Batteries need to be operated within specific use parameters to meet performance expectations and maintain their warranties, which is hugely important to buyers.

“We see a strong pull from the market for 10-year warranties for the batteries because customers want the modules warrantied in line with the system,” Evans says.

Demand for higher density battery modules at lower price points has driven rapid innovation and resulted in a market full of newer battery modules that don’t have enough testing behind them to know for certain how they will perform. To protect against this risk, battery warranties are getting more restrictive when specifying use conditions like state of charge, depth of discharge and temperature range.

It would be impossible for a non-adaptive control system to account for the unknowns of the battery’s performance while still adjusting the usage to not void the warranty.

“Our system has to keep all of that in mind while operating the asset and co-optimize all of these constraints,” Evans says.

Seeing the bigger picture

Intelligent energy storage control systems also mean more value stacking opportunities for system owners. Example: A system is designed with demand charge reduction as the primary application. Late in the monthly billing period, the system might reasonably surmise the highest peak for that month was already hit. Maybe instead of peak shaving that day, an intelligent system will perform PV self-consumption or TOU arbitrage instead.

Knowing a controller has the smarts to put every asset to optimal use, an asset owner might want to plan for it in system design by adding a little more storage than a system would need for just demand reduction.

Evans hints that Pason is preparing for that with a new application based on value stacking. The good news is, you don’t need to jump right in and buy those batteries now. The machine-learning of these systems will only improve over time as more data is compiled, and all of those smarts will be pushed back into the brains of your controller (software updates are included with the annual subscription fee).

Playing nice with others

Deploying all of this is becoming easier for novice, mid-market solar installers too. Consider Stem, another big name in machine-learning storage systems. Since 2012, Stem has captured data from its systems on a one-second basis to feed its predictive analytics, machine-learning and grid-edge computing AI. The company has largely functioned as its own project developer, becoming one of the top storage companies in California by combining its advanced AI Athena platform with the SGIP rebate to uncover big savings for customers.

Solar installers can now put all of this accumulated knowledge to use as Stem is now looking to partner with solar companies to deploy even more systems in the C&I mid-market through its Stem Partner Network.

“Through the Stem Partner Network, we deliver end-to-end partner support and services, such as training, project development advisory services, marketing and lead generation, deal support and access to a partner portal with educational resources,” Christy Martell says. “We worked with a few developers to help them understand how to design storage into their projects and ultimately bring more value to their end customers.”

Stem partners with the strongest names among solar providers across the United States to unlock new value from solar projects for their customers, backed by performance guarantees. To that end it has partnered with Energy Toolbase to provide developers an efficient path to design systems and map out projected savings.

Typically, Stem will work with the developer to suggest a combination of value streams, and this is especially valuable going forward as the market grows for complex grid services or utility programs such as demand response.

“We have seen an uptick in deal size going into 1 to 4 MW, and that’s continuing to grow,” says Than Tran, VP of global demand generation and marketing for Stem. “As we grow larger, we want to bring in other strategic providers to help us build a comprehensive solution to address all C&I customers.”

Pason Power recently partnered with Chint Power Systems (CPS), which integrates Pason Power’s software into its Energy Storage System as the exclusive platform for commercial and industrial (C&I) customers.

Smaller storage systems

Smaller storage systems leave less room for testing integrations in the field, which is why Pason Power partnered with CPS to pre-integrate their systems.

“CPS is bringing the entire solution to the commercial segment,” says Casey Miller, VP of products and business development for CPS America. “We are enabling solar installers to get in the storage game by offering their commercial building customers a turnkey storage solution with clear economic benefits and managed risk.”

This integrated energy storage solution fosters a simplified, single-source procurement process for customers rather than having to rely on multiple vendors. Hardware and software will arrive pre-built and pre-configured, making it easy for developers to install so customers can quickly begin seeing the benefits. Why this integration is important:

  1. The bankability of the firms involved is a huge factor in decision making when procuring storage components. Chint is a strong brand with a reputation for reliable products and excellent service, and Pason has been around for 40 years.
  2. The practical advantages. “By partnering with Chint, we can pre-integrate our control system into their product, do all of our testing in our facility and then sell a turnkey product that can be installed in half a day,” Evans says. This means decreased install costs, decreased commissioning costs, higher reliability and no misfires in the early billing periods that equate to missed savings opportunities.
  3. Because Pason’s modeling and energy management software were designed together, they use the same logic which ensures a truly integrated system. This enables users to pre-select the hardware which also improves accuracy when modeling the system and economics.

“Pason Power is doing the little things right to make the solution easy for customers,” Miller says “The net-net is our customers get the predicted economic benefits over the full life of the solution, not just year one or two.”

Leave a Reply

Your email address will not be published. Required fields are marked *

*

This site uses Akismet to reduce spam. Learn how your comment data is processed.