Ask An Expert: Virtual Power Plants, Reliability, & AI
![virtual power plants](https://virtual-peaker.com/wp-content/uploads/2024/04/2024-VP-AaE_navy.png)
From our thoughts on the Department of Energy report to our upcoming launch, we’ve had virtual power plants on our minds often lately. To keep up with demand, the DOE reports that virtual power plant capacity must double from approximately 30-60 GW of currently deployed VPP technologies to an astonishing 80-160 GW by 2030. With questions about the efficacy of virtual power plants on our minds, we sat down with our Director of Emerging Technologies, Dr. Greg Ledva to learn more about the future potential of virtual power plants.
How is AI used in Virtual Power Plant Strategies?
Greg Ledva: Whew, a simple question with a complicated answer. We might have to start with what “AI” actually means, given how popular the term has gotten in the mainstream: AI is the ability to replace a human’s decision-making or action-making process with a machine. It could be something as simple as a sequence of “if this, do that” scenarios that are automated. It could also be a robot that does backflips. There are different levels of complexity in different applications of AI.
Given this definition, the ability to automatically change the settings on several distributed resources at a given time is technically AI. This might be a relatively simple peak reduction approach, meaning the decision/action is relatively simple — at this time, set devices to that setting. It replaces the action of a group of people changing the settings manually, and so falls into the AI classification.
As an industry, we’re moving towards more complex use cases where more information goes into the decision-making process for device settings. The decision-making process could draw on data from the power grid (for things like non-wires alternatives) and distribution grid optimization. It could use coordinated control across many device types, taking into account each device type’s ability to adjust its power consumption over a certain time frame. It could use knowledge of how device responses vary over time to coordinate things more effectively. All of this would rely on more sophisticated algorithms, which would push the approaches into more complex AI implementations.
At Virtual Peaker, we’ve been thinking a lot about better decision-making for device settings, and we’ve got some amazing functionality rolling out soon. It aims to make controlling an aggregation of devices more intuitive for program managers and system operators; it aims at handing all of the device-level settings in the background and letting users worry about the collection of devices as a whole. All of this is done using some advanced machine learning and optimization approaches, which puts it towards the more complex end of the spectrum for AI implementations.
How Reliable are Virtual Power Plants for Meeting Consistent Load Management Needs?
GL: As an engineer, we’re taught in school that most answers are “it depends,” and this falls squarely into that area. A battery might do exactly what you want it to do until its owner opts out of an event, in which case there isn’t a resource for the grid anymore. As you aggregate more devices of a single type and more device types, the ability to get a predictable response in total improves. This is because more of the “random” actions, like opt-outs, at each device get averaged away when looking at the total behavior of the set of devices. Overall, it depends on the use case, what devices you’re trying to use to execute a certain task for the larger power grid, and the number of devices. Program managers have gotten good at knowing what they can get out of the devices within their programs.
Alternatively, you could make things more reliable and consistent by using advanced algorithms and control techniques. For example, you can also make decisions using some of the uncertainty to help eliminate the impact, or using knowledge of how one device type behaves and coordinate it with another device type. For example, if you know on average a certain number of people will opt out of an event, you can build that into how to set device settings so that opt-outs don’t impact the overall performance of the group of devices. All of this would take pressure off of the program managers to learn how to control devices, and let the data on device behavior drive the decisions on how to control the devices within a group, making it more reliable.
Virtual Power Plants, Reliability, & AI Conclusion
Ultimately, virtual power plants represent an unquantifiable figure to utility operations, which require specific, reliable, and reproducible outcomes to meet public needs. Fortunately, the combination of AI, as well as forecasting and distributed energy resource management system (DERMS) technologies provides a possible path for utilities looking to operationalize distributed energy resources (DERs), while meeting mitigating high peak energy costs, deferring expensive infrastructure upgrades, and enhancing energy security, all while improving customer satisfaction. Want to know more? Read our blog on May 7, 2024, to learn more about the next generation of virtual power plants.