SoylentNews

upstart writes for SoyCow4463:

A new genetic algorithm for traffic control optimization

Researchers at the University of Technology Sydney and DATA61 have recently developed a new method for optimizing the timing of signals in urban environments under severe traffic conditions. Their approach, presented in a paper pre-published on arXiv, entails the use of genetic algorithms (GAs), a popular computer science technique for solving optimization problems.

[...] "GAs are commonly used in optimization problems (e.g., finding the best phase duration that would minimise travel time in an intersection) by using bio-inspired functions such as individual mutation, crossover, and selection of best individuals to carry on the best genes of a population—in our case, best signal phases," [researcher Tuo] Mao said. "We thought that GAs would be a fantastic solution to solve this problem and decided to use them to generate the optimized traffic signal plans for the incident affected area."

The GA developed by Mao and his colleagues essentially explores all possible traffic signal control plans for a given intersection (e.g. the green time for "right turn" signals, "go straight' signals, etc.). Its key objective is to minimize the total travel time in an area affected by a road accident by identifying the best combination of signal phases across all intersections within that area.

"We first generate a large number of traffic control plans, including different phase durations evenly distributed in a large numerical space, which constitute the first generation of individuals from the entire population," Mao explained. "Then we apply selection, crossover and mutation in order to introduce more randomness in exploring the space of all possibilities, and select only the best candidates to carry on the optimization in a next generation."

Subsequently, the approach devised by Mao and his colleagues evolves the original population for a specific number of generations until the majority of individuals within that population are similar, and it has reached an optimal solution. The GA's final outcome is an optimized traffic signal control plan for all traffic lights in areas affected by road accidents.

[...] "Our method has three key advantages," Mao explained. Firstly, it considers non-recurrent traffic incidents, as we input the incident to the model actively after someone reported it, therefore the traffic signal control plan is aware of the incident and can respond faster. Secondly, it considers the rerouting behavior of drivers by applying a dynamic traffic assignment, which considers the road capacity drop caused by the traffic incidents. Finally, our method is efficient for exploring many possibilities of signal control plans."

The researchers evaluated their technique using a four-intersection network designed in AIMSUN, a renowned traffic modeling platform. They constructed three different scenarios in which the GA had to optimize traffic signal timings under both normal conditions and with severe traffic. In these tests, they observed that when traffic signal control plans can be adapted to a change of route by drivers after a traffic accident has occurred, congestion tends to dissipate faster.

Traffic signal control optimization under severe incident conditions using genetic algorithm (https://arxiv.org/abs/1906.05356)

Original Submission