Development of genetic algorithm-based signal optimization program for oversaturated intersections

In this dissertation, a genetic algorithm-based traffic signal optimization program especially designed for oversaturated signalized intersections is developed. The program consists of a genetic-algorithm (GA) optimizer and a mesoscopic traffic simulator. The GA optimizer is designed to search for a near-optimal signal timing plan on the basis of a fitness value obtained from the mesoscopic simulator.; The GA optimizer and mesoscopic simulator are separately tested. The limited exhaustive search, which considers all possible green splits, phase sequences, and offsets every 5 seconds for a fixed cycle length, shows that the GA optimizer provides an acceptable near-optimal solution. The mesoscopic simulator is tested by means of CORSIM simulation. The delays obtained from the mesoscopic simulator and CORSIM match well during undersaturated conditions; however, they show some discrepancies during oversaturated conditions. This result is because intersection blockage is modeled explicitly by CORSIM, but not by the mesoscopic simulator.; Two oversaturated closely-spaced intersections are considered as an initial testbed. Four models, (i) the GA-based program with queue blocking, (ii) the GA-based program without queue blocking, (iii) TRANSYT-7F version 7.2 (without queue blocking), and (iv) the newly released TRANSYT-7F version 8.1 (with queue blocking) are compared on the basis of CORSIM simulation runs. The queue blocking feature is an important enhancement to both programs. The statistical analysis shows that the GA-based program with queue blocking model produces the best signal timing plan in terms of average delay.; To examine the performance of the proposed program on a more general arterial system during oversaturated conditions, the GA-based program is also evaluated on an arterial system consisting of four intersections. The three GA-based optimization strategies being evaluated are throughput maximization, delay minimization, and modified delay minimization. The delay minimization strategy is recommended for real world implementation. The GA-based program is compared to the newly released TRANSYT-7F version 8.1. The results show that the GA-based program produces less queue time than does TRANSYT-7F version 8.1. A sensitivity analysis of results also indicates that the GA-based program outperforms TRANSYT-7F version 8.1 for oversaturated closely-spaced intersections in terms of queue time.