For example, some of the largest solar farms stretch over acres of cheap land often far away from where people live. If there is problem, technicians must get there, find out what’s wrong and order spare parts. Typically, they discover new issues while on-site. The farm can be offline for weeks before they fix it.
But what if operators could monitor their farms online in real time, spot problems before they strike and schedule repairs for evening hours, when the sky is dark and the plant doesn’t produce power? “It’s like having a crystal ball,” says Matt Perkins, chief digital officer of GE’s Power Conversion business. “If I can use machine learning and data science to help operators see the future performance of the 1 million parts of a solar farm, then I can have somebody out at the plant every Monday morning with a list of all the problems that will happen. He or she can do all the work by 1 p.m. on Monday and be done with the maintenance for the entire week or longer.”
GE’s vision is no flight of fancy. This fall, workers at a solar plant being built by Invenergy, North America’s largest independent, privately held renewable energy company, will take that crystal ball for a spin. GE will deploy at the 20-megawatt farm its solar asset performance management software, collect real-time performance data and send it to the cloud, where Predix, GE’s platform for the industrial internet, will process it. The results will give the plant’s managers a clearer understanding of the workings of their farm and a leg up on maximizing the total financial performance of the project.
GE also will use data from the farm and historical data from 70,000 other assets — pumps, motors and turbines connected to Predix — GE has installed around the world to build a digital twin of the solar farm, its virtual doppelganger. The twin will run simulations of real-life conditions and use machine-learning algorithms to pick the best operations mode and maximize its performance and optimize service. The twin also will create a “perfectly healthy” version of the farm and use it as a benchmark to spot anomalies that could signal a fault or inefficiency. The managers can use it to act early and prevent potential failure.
The solar farm has seven latest-generation GE inverters that transform the direct current pulled in by solar cells into the alternating current people use at home and at work. Each inverter comes with sensors that monitor 200 different pieces of data such as critical component temperatures and voltages, which they measure every 500 microseconds. “In the beginning, we collect as much data as possible, but we cut out pieces of it over time, as we see which elements are critical to the data science,” says Perkins. “It’s not about big data. It’s about intelligent data.”
Deployment is simple, Perkins says. GE will install a small, secure data-collection device that will draw information from the inverters and feed it into the cloud, where Predix will clean the data, analyze it and turn it into a useful set of advisories. “We see this like a morning newspaper,” says Perkins. “Plant managers can open it up in the morning and see the issues that they will need to deal with during the day. They can monitor existing problems and optimize the selection of staff members who are qualified to fix them.”
GE has already deployed modules of this technology at several other solar plants, with a total production of 2 gigawatts, Perkins says. It also has partnered with Invenergy in the past and is using similar technology on some of its gas and wind plants, including a 300 MW wind farm in Texas.
“From a plant perspective, we’re trying to take the bumps out of the long-term maintenance process for Invenergy,” Perkins says. “But from a larger perspective, we need to get solar to the point where it's a completely dispatchable, firm and predictable resource. This is the next logical step on that journey.”