Fresh out of Princeton University with a degree in mechanical and aerospace engineering, Benjamin Tai joined the planning department at PSE&G in Newark in September 2012. The idea was to come up with ways to improve the entire energy system of the utility giant.
A week later, Superstorm Sandy hit.
For days, Tai worked out in the field, one of toppled trees, flooded streets and homes, twisted lines, and shell-shocked customers, assigned to hard-hit Hudson and Bergen counties.
In a company tag team approach called “look-ups,” he and a colleague interviewed customers distraught over their losses, magnified to a catastrophic level by the scale and length of power outages caused by the historic storm.
“That was my first learning experience at this company,” Tai said.
He would never forget how a sustained absence of power affects people. And from the cauldron of his encounter with disaster, he extracted insight. “Our job,” he said, “is to keep the lights on.”
With eyes on that goal, Tai, as principal engineer in PSE&G’s Strategic Reliability and Delivery Planning Department, has spent the intervening years developing a powerful computer model that is beginning to be used.
Company officials already are employing the model to take some of the guesswork out of predicting which pieces of equipment are most likely to fail. The idea is that by removing doomed devices, customers won’t suddenly find themselves out of power.
The approach is not as easy as it sounds. After all, if equipment is functioning well, no matter what the age, it doesn’t necessarily make sense to extract it. And no public utility can afford to send workers to remove functioning equipment simply because it is old.
“We need to be predictive,” Tai said. “How do we know that a given piece of equipment is going to fail soon? How can we know when we should be replacing things?”
Since Sandy, the utility, through its $1.22 billion Energy Strong program, has rebuilt much of its gas and electric infrastructure, to protect facilities from severe weather.
Among the slew of innovations are new microprocessor-powered relays that collect and send real-time data about thousands of machines and devices at hundreds of different power-generating substations – transformers, circuit breakers, disconnect switches, distribution switchgear – and beyond, including utility poles, and wires.
“The challenge,” Tai said, “is what do you do with the data? How do you analyze it? What does the data mean?”
That’s where Tai’s model comes in. Tai factors in key “inputs,” such as impact on customers if a piece of equipment fails, amount of time for repair, chance of failure, years in operation, condition, extent of use, and repair record.
As for the model itself, it’s “a very large Excel spreadsheet representing several hundred megabytes,” Tai said. “It’s a lot of data. We are looking at 30,000-plus pieces of equipment with everything that is known about the equipment’s age, condition, and so on. It’s just a very large computer model.”
Tai and others in his department also want to build in the knowledge and insights of the company’s many experts whose field experience over decades contributes to the analysis. “We’ve been trying to develop algorithms that reflect institutional knowledge,” Tai said. “An expert may have a gut feeling about a transformer, he may have a feeling that something is wrong. A lot of our experts have been around this equipment for 30 to 40 years. They know their stuff.”
In the model, Tai employs a Monte Carlo simulation, a mathematical numerical method that solves complex problems using random and probabilistic methods. The approach, usually employed when a problem is too complex and difficult to do by direct calculation, is widely used in chemistry, economics, physics, and many other disciplines.
The idea is to create actionable information employing the data, Tai said.
The model has already proven useful to the company. Based on model predictions, company officials in May retired the 91-year-old Plank Road substation in Newark, one of the utility’s oldest functioning centers.
The station originally was built to power a city streetcar line but also served an eastern portion of the Ironbound section in Newark, as well as parts of the downtown area. It has been replaced by the Kearny Substation in Kearny, which will now serve the 3,500 customers formerly served by Plank Road. This newer station employs sensor technology designed to improve reliability.
The power system maintained and operated by PSE&G is one of the oldest in the nation and includes equipment built and installed after the World Wars. It’s a point of pride to PSE&G engineers, Tai noted, that workers have assiduously maintained these machines over the years, making it possible for them to endure.
Yet, Tai said, the country’s electrical grid is inherently unstable. This means, he said, that any fluctuation in power can cascade, with dire effects. “A single failure can take out the Northeast,” he said.
This gives him impetus to refine the model and find other ways to “keep the lights on.”
“There has to be active work,” he said. “We can’t just walk away and wait until something breaks. We have to maintain the system.”
Artificial intelligence and “big data” analysis methods are beginning to be studied for use in gauging the efficacy of the utility’s overall system. And efforts to continue to harden the utility’s electric and gas systems from continuing severe weather are expected to continue through the company’s proposed Energy Strong II program.
As for Tai, his efforts to improve modelling will continue.
“We are pretty data focused,” he said.