Page 9 - CEGE Magazine Spring 2021

Fig. 2. Khani’s algorithm makes use of social contact network of passengers on a bus; colors show different communities of passengers riding the bus (left), and (on the right) weighs relative risk and waiting time as a function of bus capacity.
activates the phase with the maximum pressure (the most cars).
The max-pressure control has several properties that recommend it. First, the system allows maximum throughput even when demand is random. Drivers see shorter queues at intersections during the most congested AM and
PM peak hours. Second, the control is amenable to decentralized computation, meaning that the optimal signal timing at each intersection depends only on the upstream and downstream queue lengths for that intersection, which makes computation of queue-length optimization easy in real-time. Third, the system is adaptive: rather than using average demand, it uses detectors to obtain queue lengths and adapts to actual demand.
Levin’s research group has partnered with Hennepin County to upgrade their adaptive traffic signal systems. Levin’s group aims to revise the model and the signal control to make the system more practically applicable. They have already addressed a significant challenge.
The max-pressure phase activation can seem quite random to drivers, who could experience arbitrarily long wait times, especially on low-demand roads. Levin’s group modified Varaiya’s max-pressure control to follow a signal cycle, that is, phases are selected
in a predefined order. The max- pressure control adaptively determines the duration of each phase and of the overall cycle based on observed queue lengths. This modification required extensive revision of the mathemati- cal model and its stability proof, with additional simplification work on the max-pressure control to retain its com- putational efficiency.
Levin is investigating the benefits of using max-pressure control on Hennepin County roads, constructing a detailed simulation of seven intersections, in- cluding traffic counts every 15 minutes. Previous results from a simulation based in Michigan showed surprisingly large reductions in delay. If the simulation pre- dicts similar results for Minnesota roads, Levin will conduct a pilot study on actual roads so that drivers can start to reap the benefits of this new control method.
Khani’s team developed an algorithm built on statistical and machine learning methods. It uses MetroTransit’s Automat- ic Passenger Count (APC) data. The new tool quantifies the relative risk to riders in each segment of a transit route and sug- gests a limited rider capacity that would ensure social distance and minimize rela- tive risk. Data visualization on the busiest bus routes suggested that allowing only 10-15 passengers on buses at one
time leads to a great reduction in virus transmission risk and avoids significant passenger wait time (Fig. 2).
Through his research, Khani strives to develop new knowledge and computer tools to solve emerging transportation problems for smart, connected, and resilient communities.
AVs AND INTERSECTIONS
MICHAEL LEVIN
has been working on a new traffic signal timing meth- od that is compu- tationally easy to implement, yet still
has favorable impacts across an entire traffic system. Levin’s research focuses on modeling connected autonomous vehicles (CAVs) and intelligent transpor- tation systems to predict and optimize how these future technologies will affect travel demand and traffic flow. He uses traffic flow, transportation network analysis.
An intersection can be a major bottleneck on an arterial road. Methods for solving this problem have been studied for decades, but research- ers are still making new discoveries.
In 1990, researchers Tassiulas and Ephremides wrote a germinal paper
on routing policies in communication networks. They thought of data being transmitted through the internet as many individual packets being moved from website servers through many routers and finally to a user’s computer. With billions of people browsing the web simultaneously, the data network could easily become overwhelmed with demand. Tassiulas and Ephremides created a stochastic process model for the movement of packets and proved that their routing policy would serve
all demand whenever possible; they even accounted for randomness in the demand patterns.
In 2013, another researcher, Varaiya, extended these ideas to traffic signals. Varaiya’s model moves vehicles through a traffic network where the control is
the traffic signal phase activation. Like Tassiulas and Ephremides, Varaiya’s signal timing policy is mathematically proven to serve all demand whenever possible. His signal timing is often called max-pressure, which describes its mathematical form. Basically, the pres- sure for a signal phase activation is its queue length, the number of cars wait- ing at a signal. The signal control seeks to move vehicles from longer queues
in the system to shorter queues. Every 10-15 seconds, the signal control policy
University of Minnesota College of Science and Engineering | DEPARTMENT OF CIVIL, ENVIRONMENTAL, AND GEO- ENGINEERING 9