Associative Memory Techniques for Quick Estimation of Network Performance Measures

Abstract

Many important decision-making processes in transportation planning and engineering involve repetitive computation of network performance measured by total network delay, throughput, network efficiency, and other measures. The computational complexity imposed by repetitive evaluation of these measures, especially under user equilibrium conditions, is a serious obstacle to timely decision making in network-related problems. This study applies associative memory techniques, which are conceptually and computationally simple, to quick estimation of these performance measures. The results of the numerical experiments are encouraging, and the relative error on average was found to be less than 2%. Furthermore, the applicability of this approximation method to bilevel network problems, a class of important but complex problems, is explored through a study of the network recovery problem, which seeks a quick and effective repair strategy for disturbed networks following natural or human-induced disasters.