Evaluation and improvement of traffic signal settings by simulation.

Several methods of setting traffic signals are studied by means of traffic simulation. A test problem, consisting of a six-signal network in Boston, is used to compare three methods' /1/ maximal bandwidth progression, /2/ simswitch /in which the center of red is simultaneous along all parallel streets/, and /3/ random search /choosing the best of a number of randomly selected settings/. performance is measured as a weighted combination of average delay per vehicle and average number of stops per vehicle. Two systematic procedures for improving a given setting are explored. Both involve one-variable-at-a-time search in the neighbourhood of the original setting. In the first, the absolute offset of an individual signal is the changed variable, in the second, the relative offset between a pair of adjacent signals is changed. The conclusions are reached that /1/ operationally significant differences exist among settings that a priori might be expected to be good ones, /2/ the criterion adopted for evaluating performance substantially affects which setting will be chosen, /3/ many traffic situations do not conform to the simplest models and it is difficult to predict good settings without detailed examination of traffic movement as through simulation, /4/ an effective way to obtain good settings through simulation may be to test out several settings considered in advance to be good and then improve the best one or two by systematic search, /5/ performance is flow-dependent, i.e, a setting good at one level may be poor at another even though the patterns of flow are similar, and /6/ settings with progressive timing seem more likely to be degraded in performance by turning vehicles than are systems with simultaneous switching. /author/.