Latest / Internet of Things with Fexingo: Connected Devices, Sensors, and Industrial IoT / How IoT Sensors Are Preventing Wildfires in the Power Grid
Transcript
- Lucas: California's wildfire season is already shaping up to be severe this year, and one of the biggest ignition sources is something you might not expect: the power grid itself. Downed or sagging power lines touching dry vegetation have caused some of the most destructive fires in the state's history. Luna: We're talking the Camp Fire, the Tubbs Fire — both traced back to utility equipment. So what's actually being done about it? Lucas: The short answer is: a massive deployment of IoT sensors. Pacific Gas & Electric — PG&E — has now installed over 80,000 grid-mounted sensors across 25,000 miles of transmission and distribution lines. These aren't just weather stations; they're purpose-built for fire risk. Luna: Eighty thousand is a lot. What are they actually measuring? Lucas: Three main things. First, accelerometers to detect line sway — if a conductor starts oscillating in high winds, that's a potential contact with a tree. Second, temperature probes to track conductor temperature, which tells you if the line is overheating and sagging. Third, humidity and ambient temperature sensors to assess local fire danger right at the pole. Luna: So each sensor node is like a mini weather station strapped to a power pole. Lucas: Exactly. And the data feeds into a machine learning model that PG&E calls its 'Wildfire Safety Operations Center.' The model processes readings every few seconds and can predict a fault — or a fire risk — hours before it would be visible to a human crew. Luna: How accurate is the model? I've seen some utilities claim 90% plus, but those numbers can be misleading. Lucas: You're right to be skeptical. PG&E published data from last season showing that their model correctly identified high-risk conditions 87 percent of the time within a 500-foot radius. But the false positive rate was around 12 percent — meaning crews got dispatched to a lot of locations where nothing actually happened. That's expensive. Luna: Still, 87 percent is pretty good if it prevents even one major fire. What's the cost per sensor? Lucas: Roughly $2,000 per node, including installation and the first year of cellular data transmission. For 80,000 nodes, that's $160 million in hardware alone. But PG&E has already committed over $5 billion to wildfire mitigation since 2020, so the sensor network is a relatively small slice. Luna: And the data — all those sensors transmitting every few seconds — that must be a huge volume. How do they handle it? Lucas: They use edge computing at the pole level. Each sensor node has a small processor that runs a pre-filtering algorithm. It only sends an alert if the readings exceed a certain threshold — like wind speed above 30 miles per hour or conductor temperature above 75 degrees Celsius. That cuts the data load by about 90 percent. Luna: Smart. Otherwise you're drowning in noise. So the utility isn't just collecting data; they're doing real-time decision-making at the edge. Lucas: Right. And that's the key shift. For decades, utilities relied on manual patrols and helicopter flyovers. A crew might inspect a given line once a year. Now they have continuous monitoring. And it's not just PG&E — Southern California Edison and San Diego Gas & Electric have similar programs. The entire industry is spending billions. Luna: Before we go deeper, I want to mention something. Conversations like this — where we actually dig into the engineering and economics of a solution — that's what keeps this show interesting for me. And if you find it useful too, listener support is what keeps it going without ads. You can help at buy me a coffee dot com slash fexingo. No pressure, just if the show adds value for you. Lucas: Yeah, it's a small way to keep the episodes coming. And speaking of value — let me add one more detail about the sensor hardware itself. The nodes are solar-powered with a battery backup that lasts up to 72 hours without sun. That's critical in remote mountain terrain where you can't run power. Luna: So they're self-sustaining. What about communication — cellular coverage isn't great in the Sierra Nevada, for instance. Lucas: Good question. PG&E uses a mix of cellular and satellite backhaul. In areas with no cell signal, they've deployed low-earth-orbit satellite modems — think Starlink-like terminals — to relay the data. That adds cost but ensures coverage. Luna: Let's talk about the regulatory side. California's Public Utilities Commission now requires utilities to submit wildfire mitigation plans with specific sensor deployment targets. Is that driving adoption? Lucas: Absolutely. In 2023, the CPUC mandated that all major investor-owned utilities in high-fire-threat districts must have sensors on every transmission line by 2027. That's a hard deadline. And it's forcing utilities to move fast — sometimes faster than the technology is ready. Luna: What do you mean by 'faster than the technology is ready'? Lucas: Well, early sensor deployments had reliability issues. Some nodes failed in extreme heat — the electronics inside got cooked. Others had battery drain problems during cloudy weeks. PG&E's first-generation sensors had a 15 percent annual failure rate. They've since switched to industrial-grade components rated for -40 to 85 degrees Celsius, and the failure rate dropped to under 3 percent. Luna: So the tech is maturing fast. What about the data analytics side — is the machine learning model getting better each season? Lucas: Yes, and that's the real moat. PG&E trained their model on historical ignition data — over 1,200 fire events from 2014 to 2020. Each event was correlated with sensor readings, weather data, and vegetation maps. The model learns patterns like 'if wind speed exceeds 40 mph and conductor temperature drops below 10°C, the risk of galloping lines increases by a factor of six.' That kind of granularity. Luna: Galloping lines — that's when ice builds up on conductors and causes them to oscillate, right? I've seen videos, it's wild. Lucas: Exactly. And in California, it's rare but dangerous because the oscillations can snap a conductor or bring two phases together, creating a spark. The model can now predict galloping conditions about four hours in advance, giving operators time to de-energize the line if needed. Luna: De-energizing lines is a big deal — it means cutting power to thousands of customers. How do they balance safety against reliability? Lucas: That's the trillion-dollar question. PG&E faced massive backlash after the 2019 public safety power shutoffs that left millions without power. So now they use a more targeted approach — instead of shutting down entire circuits, they use sectionalizing switches to isolate just the risky segment. Sensors tell them exactly where the risk is. Luna: So the sensors enable precision. Instead of a sledgehammer, you've got a scalpel. Lucas: Right. And that's where the economics get interesting. A targeted shutoff might affect 500 customers instead of 50,000. The cost savings in avoided economic disruption are enormous. A 2024 study from UC Berkeley estimated that every dollar spent on grid sensors saves about seven dollars in avoided wildfire damages and power outage costs. Luna: Seven-to-one ROI. That's compelling. But is it applicable outside California? I mean, other states have wildfire risks too — Colorado, Oregon, even parts of Texas. Lucas: Absolutely. Xcel Energy in Colorado is rolling out a similar system this year. And in Australia, where bushfires are a perennial threat, Energy Queensland has started a pilot with 5,000 sensors across the most fire-prone regions. The technology is basically portable — the specific risk models need retraining for local vegetation and weather, but the hardware is standard. Luna: One question I keep coming back to: how do they secure these sensors? If you've got 80,000 internet-connected devices in the field, that's a huge attack surface. Lucas: Security is actually a major focus. PG&E requires all sensor nodes to have hardware-based encryption — each device has a unique certificate that authenticates to the network. They also use a private LTE network for data transmission, not the public internet. And the edge processor only runs signed firmware — any tampering triggers an automatic shutdown and alert. Luna: So it's a closed-loop system. That makes sense for critical infrastructure. Lucas: Exactly. And the regulators are paying attention — the North American Electric Reliability Corporation recently proposed new cybersecurity standards specifically for grid-edge IoT devices. So the industry is moving toward a common framework. Luna: Let's zoom out for a second. We've talked about sensors, models, ROI, security — but what's the one thing that could derail this whole approach? Lucas: Probably supply chain. The sensors rely on specialized microcontrollers and MEMS accelerometers that are largely manufactured in Asia. During the chip shortage of 2021-2023, lead times for some components stretched to 52 weeks. PG&E had to delay parts of their deployment by nearly a year. If geopolitical tensions disrupt that supply chain again, the whole wildfire sensor rollout could stall. Luna: So we're trading one vulnerability — aging infrastructure — for another — supply chain dependence. But at least the sensors are giving us data to make better decisions. Lucas: That's the bottom line. We can't eliminate wildfire risk entirely, but we can move from reacting to fires after they start to preventing the ignition in the first place. And IoT sensors, combined with machine learning, are the best tool we have for that job right now.