Editor’s Note: This is the first in a four-part series examining the growing role of machine learning and artificial intelligence in the power sector. Tomorrow, we look at how regional grid operators are using AI to optimize operations.
The future of the electric grid is undoubtedly cleaner and more efficient and distributed, with hefty doses of technology and machine learning helping to operate it all. But if you’re expecting a system dramatically transformed, experts say you’ll be left waiting.
Artificial intelligence and machine learning are already helping utilities run their networks more efficiently, extending the life of equipment and helping to dispatch energy into markets more efficiently. But rather than flashy solutions, AI tends to appear in updated computing algorithms and faster solutions to complex data problems.
“I get really frustrated with the way people talk about artificial intelligence,” GlobalData analyst Stuart Ravens told Utility Dive. “There is an awful lot of talk about AI that is, frankly, BS. People hear the term ‘artificial intelligence’ and they think it means HAL from 2001,” in Stanley Kubrick’s classic science fiction film.
100% renewable energy? I’m sorry Dave, I’m afraid I can’t do that yet.
What AI does is bring together various branches of research with advanced analytics, and Ravens says many have potential utility and grid uses. “There are massive opportunities,” he said, even if they may fall short of people’s most-futuristic expectations.
There are billions of internet-connected devices right now. As more interact with the power grid, experts say utilities will need smarter solutions. Machine learning has a place in an array of utility use cases, from electric vehicle management to market dispatch.
Often times AI is about improving tools that already exist. Álinson Xavier is a researcher at Argonne National Laboratory focused on how electricity markets can more efficiently dispatch generation. His work looks at a common calculation called “security-constrained unit commitment problems.”
The calculation helps grid operators set dispatch schedules and is important because of the short clearing windows enforced in most electricity markets. And while the work is already done multiple times daily, Xavier said using machine learning to solve it faster can be of significant value.
“AI’s ability to interpret data and anticipate events “will help transform electric providers’ ability to produce and distribute power.”
Co-President, SAP Industries
“Improvements in computational performance” would allow markets to improve their efficiency through more accurate modeling, enhanced operation of combined cycle units, higher resolution for production cost curves, sub-hourly dispatch, and longer planning horizons, he concluded in a February paper.
“In our research, we are not trying to replace the tools operators are using with something done completely through AI,” Xavier told Utility Dive. “We are trying to use machine learning to enhance certain tools.”
But while improvements to behind-the-scenes calculations and mechanisms may be subtle shifts, electricity customers will see the results. AI’s ability to interpret data and anticipate events “will help transform electric providers’ ability to produce and distribute power,” Peter Maier, co-president of SAP Industries, a leader in enterprise application software, told Utility Dive.
So how are utilities harnessing the power of machine learning to better run their operations? Experts say the biggest opportunities lie in areas with great uncertainty.
Optimizing the delivery of cleaner energy will ultimately require aligning dynamic pricing with energy consumption, according to Michael Lee, CEO of Evolve Energy. And as the number of connected devices grows, that task will lean heavily on machine learning to utilize as much green energy as possible by shifting consumption to times when there are abundant renewables available on the grid.
“Giving people the opportunity to procure electricity at the wholesale power price amplifies the value of the electrify-everything trend.”
CEO, Evolve Energy
Evolve is a retail provider that offers residential customers in Texas access to real-time wholesale electricity market prices, and claims savings up to 50% through an app and system that automates a customer’s energy management. While the company’s algorithms right now only work to control AC and HVAC systems through smart thermostats, Lee says the plan is to add pool pumps and water heaters and eventually electric vehicles.
“Giving people the opportunity to procure electricity at the wholesale power price amplifies the value of the electrify-everything trend,” said Lee. And he says the app could also be used in regulated markets, helping utilities to manage their loads.
“Currently the tools available to energy companies are very blunt force” said Lee. “Demand response is a great tool to get megawatts off the grid but we’ve seen some challenges” in customer response and satisfaction.
Lee says Evolve’s AI-driven energy management platform examines three factors — the cost of the electricity, comfort of the customer, and the carbon impact of the energy — in adjusting how a home uses energy. “By looking at forward curves we get a good idea of what will happen,” said Lee. “We’re optimizing for a multi-variable equation.”
Similarly, SAP’s Maier points to a Netherlands energy provider, Vandenbron, which has developed a peer-to-peer network connecting consumers with renewable energy.
“Energy trading depends on swift and efficient connections between buyers and sellers,” Maier said. “As supply and demand levels continuously fluctuate, artificial intelligence can be used to more quickly match producers with consumers.”
While utilities are already turning to AI for energy dispatch, observers say other use cases are more nascent but rapidly growing in complexity.
Electric vehicle management
Utilities have so far not experienced many issues in integrating a modest number of electric vehicles (EV) onto their distribution systems. But as adoption grows and more vehicles are charging simultaneously, Ravens says the complexity will become an issue that machine learning can address.
Before Ravens was at GlobalData he worked at Navigant and last year authored research that concluded annual, worldwide electricity demand from EVs could exceed 400 TWh by 2035. That makes it the largest opportunity for new load growth in decades. But “the vehicles will also pose significant problems to network utilities, particularly in areas where grids are already constrained,” the report warned.
A full integration of EVs is complex because the vehicles can act as load, storage and supply. “They also move around, so utilities will never have full visibility of their location,” Navigant’s report noted.
“From planning where you install EV charging infrastructure to real-time balancing to cope with the extra load, there’s a whole raft” of opportunities for utilities to utilize AI.
Ravens says the sector must adopt a flexible approach to connected infrastructure, allowing vehicles to help balance the distribution system — soaking up renewable energy and acting as a grid resource. Executing that will require AI that can predict everything from the weather’s impact on renewable generation to traffic patterns.
“From planning where you install EV charging infrastructure to real-time balancing to cope with the extra load, there’s a whole raft” of opportunities for utilities to utilize AI, Ravens said. “You’re also dealing with assets that are mobile. And doing all of that in real time.”
The system “will have to learn from itself, to manage something as complex as that.”
Artificial intelligence has been used for years to help maintain utility transmission and distribution networks, with computer algorithms comparing expected with actual system performance to predict possible system failures. These “digital twins” are at the heart of New York Power Authority’s approach, for instance, which utilizes tens of thousands of sensors across generation and transmission assets.
“Preventive maintenance is a top priority across industry sectors,” Maier said. “In the energy sector, a digital twin can represent power plant equipment’s actual performance, leading to better simulation of future performance.”
Sensors can track a generator’s peak loads, outputs, temperature levels and network connections, said Maier, allowing for better oversight that not only helps eliminate equipment downtime, but “will automate the monitoring process, helping plants run more efficiently.”
“There are huge cost savings that can be had just through basic monitoring,” said Ravens. Something as simple as knowing when lubricant in a wind turbine is running low, can mean the difference between basic maintenance and expensive repair.
And from simple monitoring, utility systems move on to “predictive” and then “prescriptive” maintenance recommendations, said Ravens. “Not only only does the system tell you an asset could fail, it tells you what you should do to alleviate those problems. There are some really smart things going on with maintenance.”
But as utilities face the impacts of climate change, AI for system maintenance is taking on new forms.
New threats, new use cases
Consider Pacific Gas & Electric, which has struggled to protect its system from wildfires in high wind and dry conditions. The utility has thousands of miles of system to monitor and maintain, and has turned to AI to improve vegetation management.
The utility is now utilizing machine learning software to identify dead or dying trees that could pose wildfire hazards. “We’re also leveraging computer vision to help detect anomalies on a line from remote sensing sources,” PG&E spokesman Paul Doherty said.
The utility has developed new inspection tools and methods it says can quickly identify issues and enable proactive asset and system maintenance.
“This in turn reduces the risk of asset failure and potential impacts on our customers,” PG&E said in a recent smart grid report to state regulators. The utility says it is leveraging new technologies like Light Detection and Ranging and remote sensing technologies such as drones, to identify risks, including encroachment clearance and vegetation health.
“Combined with machine learning software, these data are being evaluated to identify dead or dying trees that could pose wildfire hazards or contribute to a wires-down situation,” PG&E said.
“Utilities need to grow profits from somewhere, and if they can’t grow in terms of revenue growth they have to get more efficient. There’s an incredible pressure on making the utility industry more efficient.”
Using computer analytics to examine both photographs and video of utility assets is a growing trend, said Ravens, and use cases extend beyond vegetation. “Rust is the number one threat to utilities,” he said. “There are now analytic algorithms out there better than humans at identifying rust.”
Data analytics have been used in the electric industry for years, said Ravens, and “even machine learning isn’t that new a thing.” But he also says use of the technology is now “ramping up” as utilities capture more operational data, identify new use cases and ultimately see the potential for profit.
Looking ahead, the use of machine learning to help utilities balance their distribution systems and deliver electricity more efficiently may also be a key to increasing bottom lines.
“Utilities need to grow profits from somewhere, and if they can’t grow in terms of revenue growth they have to get more efficient. There’s an incredible pressure on making the utility industry more efficient,” said Ravens.
Credit: Google News