Identification of normal traffic patterns from large datasets.

摘要

Knowledge of "normal traffic patterns" is essential for a number of transportation applications, such as signal system retiming and performance measurement. The simple historic average - the average of all the traffic data in a dataset, by the time of day - has traditionally been used to derive these traffic patterns. However, this method is biased by the presence of traffic abnormalities (such as crashes, and inclement weather). To avoid this bias, experts currently inspect the data visually. After the identification and elimination of the traffic abnormalities, the underlying "normal traffic pattern" is identified. Three main challenges of this approach are: (1) the bias introduced due to subjectivity, (2) the additional time required to analyze the data manually, and (3) the increasing sizes of the available traffic data sets (both spatially and temporally).;To address the above challenges in exploiting the traffic data archives, new data analysis tools are essential. In this research study, a new method, the Quantum-Frequency algorithm, was developed. Three other algorithms - K-Means Clustering, Wavelet-based Clustering and Median - were identified as promising algorithms, and developed further.;A methodology was developed to evaluate these promising algorithms, along with the traditional historic average algorithm. When applied to several real-world datasets from across Virginia, the K-means clustering and wavelet-based clustering failed to converge; and the historic average was significantly biased. The Quantum-Frequency Algorithm and the Median both converged and were accurate, when compared to expert analyses. Based on these findings, a final practical methodology for identifying normal traffic patterns is developed and demonstrated with two further datasets from Virginia and California.;Key contributions of this study include (1) a detailed definition for "normal traffic pattern," (2) development, sensitivity analysis and application of the Quantum-Frequency Algorithm, (3) development and application of the evaluation methodology, (4) first documented quantification of the bias in the widely-used historic average algorithm, and (5) development and demonstration of a proposed final methodology for identifying normal traffic patterns.