Abstract:
Direct optimization pertains to the use of a single traffic model both for signal timing generation and for plan evaluation. Criteria for model selection include realistic traffic representation, adequate breadth to incorporate most urban traffic management features, and the ability to represent system variability. In the U.S., the CORSIM model is the closest model meeting these requirements.
A small urban traffic network, with nine signalized intersections in the city of Chicago, U.S.A was tested. Extensive field studies gathered all the inputs for the U.S. version of TRANSYT-7F (T7F) and CORSIM simulation model in the AM and PM peak hours. Field measurements under this base case show that the queue lengths produced by CORSIM are similar to the field values.
Traditional signal optimization was carried out using the T7F package. Twelve different signal strategies were tested in T7F and the one giving the best overall performance in CORSIM was selected. Each evaluation involved 100 confirmation runs in CORSIM. Two measures of effectiveness (MOE’s) were used: link delay and total network queue time. The mean, median and standard deviation for each MOE were produced for each set of the 100 confirmation runs.
Direct signal optimization was performed using a genetic algorithm (GA). GA is a guided random search that uses the concepts of natural section and evolution to evaluate and propose improved solutions by optimizing a given objective. Given adequate computing resources, GA converges to an optimal (not necessarily global) solution. A difficulty in CORSIM application is the inherent variability in system output, which slows down convergence.
The results indicate that overall network performance improves dramatically under the direct-optimization, GAdetermined settings. Both mean and median MOE values were substantially lower than T7F and the base case, as was the variability in system performance.
Keywords:
Signal, Optimization, Genetic Algorithm
tr109.pdf