With the rapid adoption of renewable energy and battery storage – including commercial systems, residential systems, electric vehicles (EVs), and EV charging stations – virtual power plants (VPPs) are going to be far more prominent and dynamic than in the recent past. The only way for grid operators to capitalize on the aggregate power of these sites and build a fast, flexible grid is with software and artificial intelligence (AI) that can rapidly adapt to the needs of the grid and its customers. 

The market milestones that helped evolve the adoption of VPPs

Utilities weren’t always comfortable with VPPs, even as a buzzword. In the early days of storage, energy experts shied away from calling them VPPs because ‘power plant’ insinuated certain capabilities for utilities that weren’t available yet. The scalability wasn’t yet there nor was the raft of ancillary services that a traditional power plant provides, such as the inertia that a spinning mass generator introduces to a power system. 

Then a handful of state mandates started shaping the market. California declared the first 1.3 MW of storage had to incorporate greenhouse gas emission reductions. Before that, Southern California established the local capacity requirements (LCR) program. Similarly, Hawaii Electric Company’s (HECO) rolled out the next iteration of its demand response (DR) program, the Grid Services Purchase Agreement (GSPA) framework. And this year, Virginia passed The Clean Economy Act, requiring the procurement of 600 MW of solar from customer-sited resources by 2030. Utilities have started needing to find ways to comply and fulfill state mandates. And now, they’re turning to energy services companies, like Stem, to adopt VPPs into their operations. 

Today, VPPs have become mainstream and a term with which utilities, in general, are comfortable thanks to customer-sited and distributed energy resources (DERs). VPPs are the best example of how DERs can be a meaningful part of a fast, flexible grid. 

How diversifying customer-sited energy resources make VPPs more intelligent

The difference between a VPP and a more traditional, point-to-point demand response (DR) program is key. A VPP is where the aggregator is responsible for delivering the total amount of energy or service to the grid. It’s the aggregator’s job to choose which of the resources is going to be activated and how much to ask from each. The offtaker or utility wants to know which resources gave how much, putting the onus on the aggregator to both deliver and report.

In comparison, the earlier stage of technology in this area was DR programs, which were not equivalent to VPPs because they are made of simpler agreements between end customers and utilities. When the DR signal is dispatched, the end user responds; it’s a one-to-one relationship between participating customers and the utility. VPPs, however, put a layer of abstraction between the utility and the end customers, deciphering which customer doesn’t have any energy to give from its on-site resources and which customers do when events are called. VPPs deliver the amount of resources to the utility that it needs, having solved for the complexity of all those disparate customer situations with an extra layer of intelligence. 

Notable trends for VPPs in 2021 and why true AI is key for today’s VPPs

A new trend we’re starting to see with VPPs is the mixing of different types of DERs. That’s where some of the recent VPP hype may still be just ahead of actual projects in operation. It’s the aggregation of resource types that haven’t been grouped together before – whether that’s building load controls plus batteries or building load plus batteries plus solar. For example, Hawaii has certain grid services that reward customers for adding load in the middle of the day: if there is too much solar energy coming onto the system on a Sunday at noon (typically a low load day), HECO will send out an ‘add load’ signal and the utility will pay customers to add load. A customer could respond to that signal either by turning up their air conditioning, by charging a battery, or by turning off their solar system. There are multiple ways to respond to that utility call. In that case, the VPP controller picks the way that creates the most economic benefit for all, taking into account the host customer, the utility, and possibly even a third party financier.

And that’s where AI and VPPs meet. There’s an edge-level decision that needs to be made quickly, simultaneous with a fleetwide decision on a scale of 1000s of end points. It’s a huge math problem that only AI can solve: to optimize the solution that has the highest total outcome for all parties.

What AI-driven VPPs can do to create a “fast, flexible” grid that regular VPPs cannot

There are regulatory hurdles and constructs to combine diverse asset classes – whether it’s residential vs. commercial & industrial or behind the meter (BTM) vs. front of meter (FTM) – and those hurdles are getting resolved in different regions on different timelines. There are a couple of utilities that allow residential and commercial resources to be aggregated together, but this is oddly against regulations in many places. The utilities want either residential or commercial aggregations, not mixed groups. Combining that with FTM resources, which have even higher regulatory boundaries, will take even longer to work through. Electrically, it doesn’t matter whether a resource sits in front of or behind a utility meter – but from a regulatory framework perspective, it matters a lot. That’s an area I think we could see significant changes with VPPs – not overnight but this decade for sure.

How AI capabilities widen the gap of effective VPPs 

It’s hard to imagine VPPs without some level of AI. What’s Stem’s definition of AI? There must be some level of optimization capability to call something a VPP in the first place, as opposed to a simple DR program. Otherwise, it’s no better than a pager signal. Once you’ve established that, you’re doing an intelligent dispatch and the AI component comes into play where the system has to look at how it performs with or without a human in the loop. It may look at different optimization algorithms and the results of what was originally run to do better next time. For Stem, that’s what Athena energy storage software does. The more complicated the network of DERs gets, the more value there is to absorb that learning and make smarter decisions with each new event.

The importance of efficient, transparent electricity markets so no one gets left behind

As we start talking about more complicated combinations of asset classes and combinations of things you can do with any given DER, that’s where AI really shines. The more complicated the combination, the greater the delta will be between the VPP’s dispatch strategy being static and inefficient versus the AI-driven VPP that is constantly improving. 

If you’re with the more run of the mill aggregators who don’t have AI-driven VPPs, then you’re leaving money on the table. Customers are the ones that have the most at stake in terms of potential financial gain from careful selection of participation in a VPP. The overall big picture transitioning our grid to be “fast, flexible” means we cannot afford the inefficiency to not use AI. Pick your target: utilities, regulators, incumbent fossil fuel providers. There’s efficiency to be gained at every level when adding AI to the mix of DERs. To reach our global decarbonization goals at the speed with which we need, AI is a must for VPPs.