As electric utilities continue to modernize and evolve along with technology, an increasing focus is on how the electric grid can use automation and artificial intelligence within operations, particularly to help with the growing complexity within the distribution system. The two concepts are often intertwined, and certainly can be, but there is an important distinction.
Artificial intelligence, or AI, is about training machines to aid in tasks such as learning and decision making — often at a scale and pace beyond what is capable for humans.
Automation is about making systems and machines be able to operate or perform certain functions without human intervention.
For utilities, AI might be used to cull through data from equipment sensors and pull out patterns or trends for utility workers to take note of in regard to needed maintenance. Automation, on the other hand, would be setting up devices on the system such as reclosers or switches that self-trigger when certain criteria are met. Both rely on human input and design to get established and therefore are subject to flaws.
A number of technologies allowing for automation have been available to utilities for decades, including automated meters, reclosers, and switches. However, this availability has not always translated to adoption — whether because of cost or other resources needed to deploy the technologies. In the American Public Power Association’s 2020 Distribution System Reliability and Operations Survey Report, only about 30% of respondents reported having automated switching in place. And while the Energy Information Administration reports that more than 70% of all electric customers in the United States are now served by advanced metering infrastructure, just over half of public power customers are connected to smart meters.
Moving to Automation
The electric department for the city of Leesburg, Florida, began its foray into automation a decade ago as part of a larger transition to smart grid technologies. While working on various smart grid upgrades, the team singled out the need to upgrade the system’s aging feeder breakers and thought they’d give automation a try.
That decision led to the installation of automated reclosers on a section of four feeders that formed a ring bus around one of the commercial areas in the city. The city engaged ABB to provide the reclosers and other equipment needed to establish the automation, including customizing a fault detection, isolation and restoration, or FDIR, logic.
Steven Davis, Leesburg’s electric service planner supervisor, noted how setting up the system required close coordination between line crews and ABB to figure out the system, which he said was a first for both ABB and the utility. There had been some skepticism on the team about implementing automation, given that earlier iterations of automated technology were viewed as not very reliable. For the rollout, a lot of testing was done at night, said Davis, so that crews could see exactly how the system worked — and that it worked — and to run through what they would need to do in terms of safety protocols.
Since implementing the system, Davis said there have been a few times when the automation kicked in, and that it helped improve recovery time for the sections that weren’t affected. He also mentioned how switching can be achieved more quickly and easily, and how it requires fewer field personnel. “Everything had already been swapped out and switched out before the crews could even make it on site,” he said. “So, all they had to worry about was that one section that was still out.”
Since outages are rare events, the day-to-day benefit of the automation comes from the data that can be collected more easily from the system.
“Our operations folks really like having the automated reclosers at midpoints and other locations, so that, No. 1, we can look at the loads,” said Davis, mentioning that the loops include some large subdivisions in the utility’s territory. “Having that data coming in, we use it quite a bit in trying to balance the loads on the feeders. Being able to look at some real-time measurements instead of having to send crews out and get those measurements.”
The two initial loops where the automated reclosers were added constitute about 20% of Leesburg’s load. Davis said the utility has continued to add more automatic reclosers to midpoints at each of the five substations on Leesburg’s system and for about half of its feeders. The reclosers can be controlled from the operations center, which helps personnel to better monitor load and allows for real-time communication with the reclosers and to see faults as they happen. In addition, having the detailed data on the load patterns has been helpful in planning more accurately for new construction in the area.
With the communication technology that supports the feeders now outdated and no longer supported by ABB, the crews that manage the substations are reviewing other options for being able to keep the automatic restoration as part of the system design. The utility is looking at enhancing the communications on the system, by adding either fiber cable or enhanced cellular networks, and upgrading the radio system.
Focusing on High-use Areas
When Lakeland Electric, the third-largest public power utility in Florida, looked to begin implementing a distribution automation scheme on its system, it decided to focus on the Publix Manufacturing Center, a high-use area within its system that was already served by its own substation.
The distribution automation system uses a real-time automation controller and automated reclosers that are connected via a high-speed fiber optic system. Lakeland Electric personnel worked with Schweitzer Engineering Laboratories, which developed the equipment, on the system specifications and programming required to get the automation correct.
Before implementing the automation scheme, Lakeland experienced fault conditions that resulted in power outages across multiple zones, which affected multiple manufacturing centers across its territory. Now, should a fault occur within the manufacturing complex, the system can quickly isolate only the faulted section and maintain power to the remainder of the complex. This setup results in minimum power disruption to the other centers and to the system overall.
The system is designed so that distribution relays are programmed to coordinate timed overcurrent elements with downstream recloser controllers. The distribution relay trip logic contains phase and ground-time overcurrent, instantaneous overcurrent relay elements, and trip commands received from the automation controller. The real-time automation controller relies on several programmed equations to determine the appropriate relay response, which depends on the status of the tagging relay and reclosing shot bit.
Scott Fowler, manager of substation operations at Lakeland Electric, noted that implementation of the distribution automation system has allowed Lakeland to improve its reliability and overall availability to the manufacturing center and allows for rapid fault isolation if and when faults occur.
Fowler said that the system equipment, which has been in place since 2011, is now somewhat obsolete, so staff will soon need to reprogram and replace the system using Schweitzer Lab’s current “standard libraries.”
Artificial Intelligence Gets Real
While more nascent than automation technologies, utilities have already begun to explore and deploy AI for a variety of uses. Applications within more traditional utility operations range from helping to do speedy and deep analysis of drone imagery to conduct more thorough inspections of assets and predict imminent failure of equipment, to helping system operators better manage and balance assets from a growing array of distributed energy resources and renewable sources. This can help predict future output and forecast demand more accurately based on a wide set of factors.
AI can also potentially help save lives, not just time and effort. The Electric Power Research Institute recently studied whether tools such as natural language processing and machine learning could conduct meaningful analysis of utility incident and injury reports to find any common precursors to events that cause more serious injury, leading to more lost workdays and to fatality. EPRI researchers compiled data from eight utilities with their Occupational Health and Safety Database, providing the AI with 100,000 records spanning 26 years of incidents. The analysis identified the top serious incident and fatality precursor conditions, which included tree trimming and falling from heights, so that utilities could cross-compare this data with safety programming and protocols to bring down risk of injury among workers.
Outside of direct grid operations, AI could also help utilities to better understand and break down customer usage trends and needs to identify who would benefit from specific programs and incentives, or analyze customer service calls and correspondence to alert utilities to patterns that show customers might be dissatisfied. The former would require utilities to install smart meters that can identify usage at the appliance level.
Given the upfront investment needed to deploy technologies that support AI, and the reality that AI offers a better return when it can pull from a larger data set, smaller utilities might be dissuaded from exploring any machine learning applications. When weighing the tradeoff in cost versus savings to make it worthwhile, utilities might want to explore taking smaller steps to test the concept, so that they don’t find themselves completely behind the curve or left behind compared to their peers when any such practices become the norm.
Beyond simply keeping up with the industry, weighing the cost of deploying AI-driven technology requires a long-term assessment on how such a system might impact anything from rate design to infrastructure investments to future load.
Perhaps one of the stronger arguments for utilities investing in systems to crunch customer data from AMI revolves around transportation electrification. This is because the overwhelming majority of electric vehicle charging occurs at drivers’ homes. Utilities are not only in the best position to capture and make sense of this charging behavior data, but are also in the position to potentially suffer most from the inability to predict or influence load fluctuations from EV charging. This usage will be highly variable based on the specific community’s characteristics — including average commutes and typical housing types. Even if a utility has only a handful of EVs within its community, getting a foundation for understanding the local patterns will help utilities to be better assured and prepared if and when adoption expands.
The potential for return on investment is more difficult to calculate when it comes to AI applications in customer service, but there is value. Utility leaders, particularly in public power, are likely keenly aware that customers can make decisions about how much power they use — and how trusting they are of adopting additional electric end-uses — based on their experiences with the utility. The utility holds sway over how much a customer sees value in acquiring energy efficient appliances, handles home heating and cooling, and in acquiring any assets for backup power. So, as homeowners start to make long-term decisions that affect the electrification of their homes (or transport), which will ultimately influence the utility’s future load growth, getting trust now will be paramount. Further, a PricewaterhouseCoopers analysis found that customer satisfaction is necessary in getting rate increases approved.
The nature of a changing industry is that the historical data we have might not be as helpful in identifying patterns for the future. When it comes to AI and automation, the important focus is for continual learning — whether that’s the people or the machines.
