Latest / The Edge Computing Podcast with Fexingo: Local Compute, CDNs, and Distributed Infrastructure / How Edge Computing Is Powering Real-Time Grid Frequency Regulation
Transcript
- Lucas: When you plug in a charger, flip a light switch, or start an induction cooktop, you're drawing power from a grid that must stay at exactly 60 hertz — sixty complete cycles per second. The moment that number drifts even half a percent, the grid operator has to act. Luna: And historically, that action has come from large power plants spinning turbines that can ramp up or down in a few seconds. But as we retire coal and gas, who's keeping the frequency steady? Lucas: That's the trillion-dollar question. And increasingly, the answer isn't a power plant at all — it's a small computer sitting at a substation, running edge inference models that sense a frequency dip and respond in under 20 milliseconds. Luna: Wait, so edge computing is literally balancing the grid right now? This isn't a lab demo? Lucas: It's in pilot. A mid-sized regional utility in the US — I won't name them because the pilot is still under NDA — deployed fifty edge nodes at key substations last year. Each node is essentially a ruggedized NVIDIA Jetson module connected to a phasor measurement unit, or PMU, that samples voltage and current sixty times per second. Luna: So it's a local compute unit that reads the PMU data, detects a frequency deviation, and then sends a command to something like a battery or a solar inverter to adjust output. Lucas: Exactly. The key is latency. Traditionally, a PMU sends data to a central SCADA system in the cloud, which analyses it and sends a command back. That round trip can take two to five seconds — acceptable for load forecasting, but way too slow for frequency regulation, especially as inverter-based resources replace spinning turbines. Luna: Because turbines have physical inertia. A spinning rotor doesn't stop instantly if there's a sudden load change. But solar panels and battery inverters have zero inertia — they respond electronically or not at all. Lucas: Right. And the more renewables we add, the less inertia the grid has. So the response has to come faster. The IEEE 1547-2018 standard for distributed energy resources actually requires inverters to respond to frequency deviations within a few hundred milliseconds. But the grid operator has to detect the deviation and send the signal first. Luna: Which is where the edge node fills the gap. The pilot utility's edge nodes are achieving a total detection to actuation time of eight to twelve milliseconds — well within the window to prevent under-frequency load shedding. Lucas: Let me put some numbers on that. The Eastern Interconnection, which covers most of the US east of the Rockies, has a frequency response requirement that any deviation greater than 0.05 hertz must be arrested within thirty seconds. Sounds generous, but with low inertia, the rate of change of frequency can be so steep that if you don't act in the first few cycles, you're in trouble. Luna: So shifting from cloud-based SCADA to edge-based control effectively buys the grid an extra couple of seconds of safety margin. Lucas: More than that — it enables a whole new category of fast frequency response that utilities couldn't reliably deploy before. The pilot is using the edge nodes to coordinate a fleet of about two hundred residential batteries enrolled in a virtual power plant. When a frequency dip is detected locally, the edge node sends a signal via LTE to a subset of those batteries to discharge for a few seconds. Luna: I've seen that concept before, but usually the coordination happens in the cloud. What's the advantage of doing it at the edge? Lucas: Resilience. If the cloud goes down — or if there's a communications outage — the edge node can operate autonomously based on its local PMU readings. The utility's central control room still has visibility, but the edge doesn't need permission to act. That's a big deal for grid reliability. Luna: And it's cheaper than the alternative. A new capacitor bank or synchronous condenser at a substation can cost half a million dollars. An edge node with a Jetson module and a PMU interface might run fifteen thousand. Lucas: The economics are compelling, but there are challenges. The edge nodes have to be hardened for substation environments — wide temperature ranges, high electromagnetic interference, limited physical security. And the machine learning models need to be trained on local grid behaviour to avoid false positives. Luna: False positives meaning the node detects a frequency dip that isn't real and commands batteries to discharge unnecessarily? Lucas: Exactly. In the pilot, they had an incident during a thunderstorm where a lightning strike near a substation caused a transient voltage spike that the PMU interpreted as a frequency excursion. The edge model hadn't been trained on lightning transients — it triggered a discharge command. The utility caught it in post-analysis, but it highlighted the need for more robust pre-processing. Luna: So they're now feeding the model historical PMU data that includes lightning events, switching transients, and even geomagnetically induced currents from solar storms. Lucas: Correct. After retraining, the false positive rate dropped from about one in twenty events to one in two hundred. The model now runs an initial filter that checks if the rate of change of frequency exceeds a threshold that's physically implausible for a genuine grid event — that catches most transients. Luna: It's fascinating how this intersects with the broader edge computing narrative. We usually talk about latency for self-driving cars or factory robots. But the grid might be the most latency-critical system of all — because failure means a blackout. Lucas: And the stakes are only rising. The North American Electric Reliability Corporation, NERC, recently published a report showing that the frequency response of the Eastern Interconnection has degraded by about fifteen percent over the past decade due to the loss of synchronous generation. They explicitly recommend faster-acting controls. Luna: So edge computing is moving from a nice to have to a must-have for grid operators. What's the next milestone for this pilot? Lucas: The utility plans to expand from fifty to two hundred nodes by the end of 2027, covering all major substations in its territory. They're also working on a peer to peer protocol so that if one node loses communication with the cloud, it can coordinate with neighbouring nodes over a local mesh network. Luna: That would make the grid more like the internet — distributed, self-healing, and less dependent on centralised control. Lucas: Exactly. And it's a reminder that edge computing isn't just about faster Netflix or lower-latency gaming. It's about critical infrastructure that we all rely on, every second of every day. Luna: You know, Lucas, conversations like this one — where we get into the nitty-gritty of how technology is quietly making essential systems work better — that's exactly why people tell us they appreciate this show. It's ad-free, independent, and focused on real substance. Lucas: And if today's deep dive into grid frequency regulation gave you something useful, or if you just enjoy having a tech podcast that doesn't waste your time with fluff, listener support is what keeps this going. You can find us at buy me a coffee dot com slash fexingo — all lowercase. Luna: No pressure, no premium tiers, no ads to skip. Just a way for folks who get value from the show to help cover costs. And it genuinely makes a difference. Lucas: Back to the grid — one more interesting detail. The utility's pilot also showed that edge nodes can provide what's called synthetic inertia. By commanding batteries to inject or absorb power in under 20 milliseconds, the edge node mimics the physical inertia of a spinning turbine. Luna: So you're essentially creating a software-defined inertia layer on top of the physical grid. Lucas: That's exactly how one of the utility engineers described it. And because the edge node can be reprogrammed remotely, that synthetic inertia can be tuned dynamically — more inertia during peak renewable generation, less when the grid is stable. That's something no physical turbine can do. Luna: It's a glimpse of where we're headed: a grid that's not just smart, but actively adaptive, with intelligence distributed all the way down to the substation level. Lucas: And we'll be watching how this pilot scales — and whether other utilities follow suit. For The Edge Computing Podcast, I'm Lucas. Luna: And I'm Luna. See you next time.