Determining Path Flows in Networks: Quantifying the Tradeoff between . . .

With the sensor technology in-place today, it is difficult to correctly predict path utilization coefficients (PUCs: the percentage of an origin-destination (OD) pair’s traffic flows traveling on a given path or route) because the problem is hugely underspecified, requiring inference to overcome the under-specificity. The shortcoming of this method, as discussed in detail, is that the inference rules applied may be wrong. In fact, little path flow research has addressed the issues of (a) reliability in the instrumentation, (b) reliability of other input data or models, (c) reliability and confidence level of the path flows solution, (d) the relationship between what is observed and what is inferred, and (e) verifying that defined path flows are those actually occurring in the network (and discerned), which are all directly relevant to the practical application of path flow estimation and to guiding instrumentation research to meet the needs of the evolving world of transportation. Therefore, this dissertation research addresses the key questions of (1) how well can path flow rates be determined from an instrumented network, and (2) how much inference is required to yield reasonable estimates of all path utilization coefficients in a network. It was discovered that observed link flow data are only slightly better than spurious input data for PUC prediction; automatic vehicl