RAILWAY AGE, DECEMBER 2021 ISSUE: The application of AI (Artificial Intelligence) and machine learning results in continuous, accurate reporting on railcar locations to improve shipping efficiency and reliability.
Rail offers many significant advantages for freight shippers as compared with trucking. For many shippers, the benefits of switching from highway to rail are significant:
• Safety: U.S. freight railroads have lower employee injury rates than most other major industries.
• Sustainability: On average, one ton of rail freight moves 480 miles on one gallon of fuel, and shippers who convert from truck to rail lower greenhouse gas emissions by 75%, according to Association of American Railroads statistics.
• Cost: The greater fuel and labor efficiency of rail often translates into lower transportation costs.
Historically, one challenge for rail shippers has been that continuous visibility and real-time shipment status information have not been as readily available as similar information provided by truckers.
The so-called “Amazon effect” has resulted in customers’ general expectation that the location of shipments should be available in real time, which is something truckers are able to provide by employing widely available and low-cost GPS tracking systems.
It’s not that simple for rail freight. While each Class I and many regional and short line railroads have very good systems for tracking and reporting shipments moving on their own tracks, providing high-quality visibility and estimated time of arrival for interline shipments that move on multiple railroads has traditionally been more challenging.
The scale of the challenge multiplies with the number of different carriers involved, as well as for shipments that begin and/or end with transport by water or highway. This is why Railinc has devoted considerable resources to the development of a new Advanced ETA to meet this critical industry need.
One of the reasons the railroad industry created Railinc was to serve as a neutral information provider of equipment condition and location throughout the North American rail network. Since its inception in 1999, Railinc has grown to become the industry’s largest and most accurate source for interline rail data. Recently, it launched its TransmetriQ unit, where transportation experts, UX designers, data scientists, and critical thinkers collaborate to tackle some of the industry’s most complex needs.
The development of Advanced ETA has been one of TransmetriQ’s first major projects.
The TransmetriQ team was tasked with improving ETAs that relied primarily on historical data to predict the performance of traffic moving across the rail network. When operations are relatively stable, this system works reasonably well. However, when changes occur in routing, interchange schedules, and other variables, the process has been less reliable.
The solution was not simple to design and implement, which is why TransmetriQ devoted an experienced and diverse team to work on it. That effort is proving to be effective. Initial results have already shown a marked improvement in arrival time predictions compared with the traditional ETA methods.
The Advanced ETA employs Artificial Intelligence (AI) and machine learning to create dynamic models based on thousands of origin-destination pairs.
With Advanced ETA, available historical data is just a starting point. Utilizing sequence modeling, the system itself learns with the passage of time how to identify the most important sequence elements and other real-world factors that will impact a shipment’s arrival time. This gives Advanced ETA the ability to predict and update expected arrival times in real time as shipments move over the rail network.
For example, if a train being tracked moves quickly through a certain location, this could be a signal that the train is on a high-priority schedule, and the system will take that into account in calculating how the train will perform for the balance of its trip.
This way, Advanced ETA can instantaneously update the arrival predictions as often as necessary to keep receivers apprised of the status of their inbound flow of commodities and products.
The first application of Advanced ETA has focused on high-volume intermodal lanes, which have generated vast data histories of operations that are less complex than for carload shipments.
A look behind the scenes at the development team members who have worked on the project since 2018 reveals how far they have come, and the promise Advanced ETA holds for the future.
A key member of the team has been Product Manager Danny Dever. Following a review of then-current ETA capabilities, Dever says “a goal was set to use machine learning to identify some of the errors that were inherent when historical data was the primary source of information.”
In cooperation with a consulting group, Dever and his team built a complex system that provided an ETA estimate from a model over a fixed route. This initial effort relied on projecting a train’s movement over the most probable route, based on historical data, to see how shipments would progress over that route at points along the way.
“Though it was pretty good within the limitations,” he says, “if you got the route wrong anywhere along the way, there was a high likelihood that predictions would not be accurate.”
Despite this, the team was gaining experience with machine learning, leading to the next and current development phase, which employed sequence modeling. These models can use data from various inputs including time-series data, text streams, emails, etc.
“Sequence modeling has greatly improved during the past two years,” Dever says. “It is now able to help us understand some of most complex real-world railroad operating scenarios.”
The models take information that has traditionally been manually analyzed and learn from it much faster.
The more advanced technologies let the machine learn the most important elements in a sequence. If a shipment moves on a route that was planned via Houston, for example, the model can pick up and analyze data that may indicate that it will actually travel via Austin and automatically update the ETA on that basis.
“Because we have direct access to all North American rail freight movements, we are able to evaluate current conditions on a given route,” Dever says. “This is another source of information that makes machine learning more effective.”
The initial marked improvement Advanced ETA has already achieved over previous ETA systems demonstrates the value of this approach. With continued experience, collection of more data, and ongoing learning, the system’s accuracy is expected to continue to improve.
Initial development has focused on the more stable and higher volume intermodal rail routes.
“Carload shipments pose additional challenges,” says Dever. “They are our next target.”
Advanced ETAs—coupled with other visibility and operating improvements that have been implemented—have North America’s railroads and their shippers poised to realize major benefits in supply chain efficiency and reliability.
About Railinc: Railinc is the railroad industry’s innovative and reliable resource for rail data, IT, and information services. The company deploys data that helps railroads, rail equipment owners, and other industry participants manage their businesses more effectively and efficiently. Railinc is the largest single source of real-time, accurate interline rail data for the North American railroad system. Located in Cary, N.C., Railinc is a wholly owned subsidiary of the Association of American Railroads. For more information, visit www.railinc.com.
About TransmetriQ: TransmetriQ is a group of Railinc transportation experts, product managers, UX designers, software developers, data scientists, and critical thinkers developing insights and solutions, which help customers build businesses that compete and win. Our teams work to improve our current products and develop the next generation of business-oriented transportation solutions. Visit us at www.transmetriq.com.