Faster Than Real Time
By Theresa Pepin
Approximately one hour after a 9.0 magnitude earthquake occurred off the coast of Japan in 2011, a powerful tsunami hit the country’s northeast coastline. Surging water obliterated entire cities in just minutes, leaving emergency personnel very little time to figure out how to effectively evacuate the area while trying to save the lives of residents trapped under rubble and debris.
This is the type of nightmare scenario that has inspired Lee Han, UT professor of civil and environmental engineering, to devise a revolutionary traffic planning system to deal with such emergencies.
Together, Han and his phalanx of students are building the foundation for a faster-than-real-time traffic simulation program that operates more than 1,000 times quicker than anything currently available.
At UT’s Transportation Systems Laboratory (TSL), Han and his team continually test, compile data, and develop multiple models in preparation for everything from everyday traffic congestion to extreme emergency management.
In describing his work, Han mentions the terms “microscopic” and “large scale” at the same time. While constructing large-scale models capable of projecting a range of predictive scenarios for decision making in time to avert citywide or regional catastrophe, traffic engineers must also ground those systems in the behaviors and experiential learning of individual drivers; the “microscopic” evidence from where, as Han says, “the rubber meets the road.”
For example, realistic field situations where vehicles, road configurations, traffic laws, driving customs and cultures, lighting, and weather conditions are vastly different must be considered. In order to extend and generalize the research, Han’s students have traveled to many different countries to collect driving behavior data.
In traffic simulation, it is also critical to overcome the sequential nature of traditional models–a major constraint holding back the potential gains of applying complex parallel computational sciences. At the microscopic level, sequential in the context of driving refers to the idiosyncratic behaviors and reaction times of drivers as they follow behind other vehicles, change lanes on the interstate at high speed, look for gaps to turn, navigate stop-and-go city driving, and idle in frustrating traffic jams.
The benefits of a successful, large-scale traffic simulation system are many: smoother driving on shorter commutes; increased safety and fuel savings; emission reductions; and heightened security in the event of terrorist attacks, mass evacuations, and major accidents.
However, this complex technique requires massive amounts of real-time data of all different types to ensure the simulations in the lab are closely mimicking and timely relevant to the real world. The system must be dynamic enough to keep up with the endless stream of data from a growing number of highway and on-board wireless sensors.
The computational tasks of handling both the data and calculations are clearly daunting; however, they have been addressed by a Joint Directed Research and Development grant that allows Han to take advantage of the petascale supercomputing facilities at the nearby Oak Ridge National Laboratory.
“The shared ambition of tackling a very challenging sequential problem with parallel means has been a driving force behind the collaboration,” Han says. He has begun to ponder the complex algorithms required for multiple models that can be crunched simultaneously by ORNL’s Titan supercomputer, capable of processing 20 petaflops, or 20 trillion calculations, per second.
Several of those models are based upon the structure of the honeycomb, an organizing principle for developing new algorithms that captures the critical importance of both the boundaries–highway edges, lane lines, and vehicular bodies known as “force fields”–and the “holes”–openings in traffic flow–in the space on the road.
“Anyone who has traveled widely can appreciate the importance of models and calculations that exploit the shifting force fields and holes in traffic, because for drivers in many cultures, lane lines are merely a suggestion,” Han says.
As Han’s traffic flow optimization and simulation models continue to undergo validation and refinement, the promise of a safer world without traffic jams crawls closer to reality. And while no one can predict the future, the ability to make predictions and test them ahead of time in a virtual environment will make it easier to keep up with the present–especially when disaster strikes or chaos threatens.