Keeping up with Climate Tech vol. 1

Year after year, electric utility companies across the country spend billions trying to minimize the impact of wildfires, floods, high winds, and other severe weather on their assets. Despite these investments, these companies’ assets are battered year after year by ever-intensifying climate events. What they need is effective, data-backed guidance on how much to invest and where, as well as how to best position those investments in front of regulators so they can provide the best experience to their customers.

This is where Rhizome Data, a startup founded by Mishal Thadani and Rahul Dubey, comes in.

Thadani was leading the regulatory and strategy team at Urbint, a utility infrastructure damage prevention company. It was here that he began noticing a trend in what utilities were looking for as they made their investments. 

“My job at Urbint was to get regulators comfortable with the models that utilities were using as a way to prioritize investments,” he said. “What really started to occur a few years ago was an interest from utilities, as extreme weather events have been getting worse, in better justifying and spending capital on the hardening and resilience of energy systems,” Thadani said in an interview with the Harvard Technology Review.

So, he decided to create a solution: Rhizome, an AI-powered platform that identifies the likelihood of various assets failing and disruptions of any kind that end up causing transformers to overload. 

“There is no platform that takes in as many factors to understand hazards and their impact on utility assets,” Thadani said. “Our goal is to help utilities show the benefit of their investments to customers at the end of the day in terms of fewer power outages, fewer restoration costs, lower operational costs in general, and of course, showing how it’ll alleviate the impact of extreme weather to communities.”

Rhizome’s risk calculation, very basically, is the likelihood of the failure times the consequence of that failure. To get the best estimate possible, Rhizome collects a significant amount of data from their utility customers, which include Seattle City and Light and Vermont Electric Power Company.

“So the first thing that we do with our utility customers is we ingest their data related to their assets, their history of outages, and the layout of their system,” Thadani said. “From a historical perspective, we can see what outages have occurred, and what the composites of those outages were. What were the extreme weather conditions during those outages, and was there a presence of vegetation or some sort of geographic characteristic that was more likely to drive that outage?”

Using that historical data, Rhizome trains a machine learning model to understand the likelihood of asset failure in the utility’s future. 

“When utilities do their own forecasts, they hit a 0.6 R-square value,” Thadani said. “That could use quite a bit of improvement, and it’s because those models are typically built with only a couple of variables in a linear regression. When we apply a lot of additional data, whether vegetation or data sets within the utility that are harder to reach for those internal teams, we actually have proven that we get to a 0.9 R-square. So it’s a very tangible improvement.”

Additionally, by getting the topology of a system with an impacted asset, Rhizome’s platform can determine which customers would lose power. This allows them to compute the economic loss of impacted customers and the restoration costs that go into repairing the system. 

With this calculation, the platform compares various investment scenarios (for example: should the company invest more heavily in vegetation management or undergrounding, or should they dedicate more to line rebuilds and reconductoring?). Using this analysis, Rhizome’s algorithm devises the most effective investment strategy possible. Then they file resilience plans to regulators with impact metrics and cost-benefit analyses.

To ensure that the investments the platform is recommending are up to the mark in our ever-changing world, high-resolution climate projections are used to forecast future events and asset failures.

“We take climate projection data to anticipate what extreme weather hazards are likely going to appear over time, and how those intensities are going to change,” Thadani said. “So an example of that is on the East Coast, what was once a one-in-ten-year storm in 2050 is likely going to be a one-in-two-year storm. We need to account for the return period of an event of that intensity.”

Thadani said there have been several decisions that Rhizome helps utilities make that he hadn’t anticipated from the start, including that some utilities would broadly change their design standards based on the results of Rhizome’s models. 

“An example is that average summer temperatures in the Pacific Northwest are going up a few degrees. That few degrees might actually cross the temperature that the utility’s transformers are rated for. So over time, you’ll see faster degradation on your transformers and increase transformer failures. So the utility we worked with actually considered, is it better to proactively replace some transformers with newer transformers, or, as assets fail, do we replace each failed transformer with a new, higher rated transformer? We learned over time that we can help support those decisions too, and that’s why we have such a broad range of investments we consider.”

Thadani said he hopes to leverage new data collection technologies to make Rhizome an even more robust platform.

“Gridware is a great example. They place sensors on the grid to identify load fluctuations, outages or potential flashes and arcs in real time,” he said. “So they give the district operators a whole lot of visibility while also giving them a record of historical load and voltage fluctuations. We can leverage that data to better train our models.”

Rhizome is also looking for better ways to capture data, with companies like Noteworthy, Overstory, and AIDash offering exciting avenues such as through satellites or cameras on utility vehicles.