Mark Chung’s unexpectedly high $ 500 monthly electric bill zapped his curiosity.
So when his Silicon Valley utility couldn’t explain it — despite “smart meters” installed throughout its system — he took matters into his own hands.
The Stanford-trained electrical engineer hacked some inexpensive meters from his local hardware store to be wi-fi enabled, and then built an electrical map of his home.
They showed that a small failure in his pool pump was causing a massive current overload, which couldn’t have been detected with traditional tools. More importantly, he learned how hard it is to get information from buildings, which typically lack any kind of computerized management.
Thus was born the idea for Verdigris, a startup that wants to help conserve energy in buildings using GPU-powered artificial intelligence.
And it’s a large problem.
Buildings gobble up about 70 percent of the world’s electricity — and waste 60 percent of it. That’s $ 100 billion wasted on electricity each year — and a chance to cut an estimated 15 percent of the world’s greenhouse gas emissions.
To address the issue, Verdigris clamps proprietary wireless sensors onto electrical mains, panels and circuits. While most buildings get occasional walk-through energy usage audits, Verdigris’ digital system uploads electricity consumption data to the cloud 24/7.
From there, it can sell the raw data to building managers or apply its own AI algorithms and provide insights it gleans. It can even integrate the data with building management systems to automate electricity usage controls.
By forecasting problems and identifying areas for optimization and automation, Chung, who serves as Verdigris’s CEO, says the firm can help facilities get more done. Hotel managers, for example, could detect and fix building issues before guests even notice them.
All this wouldn’t be possible without the plummeting costs of bandwidth, sensors, data collection, processing and storage that have fueled AI’s growth.
“AI is extending into every facet of our lives: how we travel, how we produce food, how we work, how we live,” Chung said. “Smart buildings are one of the most valuable and largest opportunities for this trend.”
It didn’t take long for Chung to realize this insight could help other homeowners as well as large buildings. (In fact, NVIDIA is exploring how it can use Verdigris’ technology to reduce electrical consumption on its Silicon Valley campus.)
Verdigris’ trains its models on NVIDIA GeForce GTX 1070 GPUs. Chung estimates this helps Verdigris train models 20 times as fast as on CPUs. He expects the role of GPU acceleration will grow as the company moves into circuit classification and other problems that will benefit from convolutional neural networks.
From Smart Buildings to Smart Cities
In the meantime, the company is applying its technology to reduce carbon footprints by increasing energy efficiency. Eventually, Chung said he’d like Verdigris to expand beyond smart building optimization and into enabling smart cities.
“If there is a large disaster, you could responsively adjust the city to shut down everything except central services and redirect energy to best support people,” Chung said.
The work won’t be done, Chung said, “until we have an automated planet.”