A probabilistic multipath traffic assignment model which obviates path enumeration

A probabilistic assignment model that attempts to circumvent path enumeration, in such a way that the resulting effect is identical to what would have been obtained if each path had been assigned trips separately under certain choice probability assumptions.

Abstract

FOLLOWING A BRIEF DISCUSSION OF SHORTCOMINGS OF ALL-OR- NOTHING ASSIGNMENT AND OF PRIOR EFFORTS IN MULTIPATH ASSIGNMENT, TWO ALGORITHMS ARE PRESENTED, BOTH OF WHICH YIELD RESULTS WITH THE ABOVE PROPERTIES. THE ALGORITHMS DIFFER IN THEIR DEFINITION OF A REASONABLE PATH AND IN THE NUMBER OF TIMES EACH IS EXECUTED TO ASSIGN ALL TRIPS FROM A GIVEN ORIGIN NODE. THE FIRST USES MORE RESTRICTIVE DEFINITION THAN THE SECOND AND REQUIRES A SINGLE EXECUTION FOR EACH DISTINCT ORIGIN/DESTINATION NODE PAIR. THE SECOND, A MINOR VARIATION OF THE FIRST, USES A RELAXED DEFINITION OF A REASONABLE PATH, WHICH ALLOWS THE SIMULTANEOUS ASSIGNMENT OF ALL TRIPS ORIGINATING AT A GIVEN NODE IN A SINGLE EXECUTION. COMPARED WITH THE FIRST ALGORITHM, THE LATTER PARALLEL MULTIPATH ALGORITHM IS MUCH MORE EFFICIENT COMPUTATIONALLY, BUT IS LESS DISCRIMINATING IN ITS SELECTION OF PROBABLE PATHS. (A).