摘要：AbstractTraffic flow estimates play a key role for strategic and operational planning of transport networks. Although the amplitude and peak times in flows change from location to location, some consistent patterns emerge across a region. Clustering solutions appear as a powerful tool to reveal hidden trends that can easily be applied on historical traffic data to estimate traffic flows. However, these historical data traditionally are collected by detectors on only a limited number of road sections. This communication presents a methodology for estimating traffic flows using road features as clustering variables, so that it can be applied to any road section. In particular, a factor related to the attractiveness of road sections, in terms of characteristics of nearby areas, will be used to cluster road sections, deriving typical flow profiles of the resulting groups. To obtain these typical profiles, data collected by permanent detectors on a broad geographic distribution of sites across the Spanish road network have been studied. Then the flow prediction procedure for a given location is based on obtaining its attractiveness factor, finding its best match, and associating the typical flow pattern of such a group (weighted by a correction factor) to the location. The results show that the methodology make good use of historical data and, in most cases, the times of the main peaks are approximately determined. Although the prediction accuracy in the amplitude of the curves varies somewhat from location to location, the accuracy is acceptable for roads classified into groups with better similarity measurements. The applicability of the procedure to any road location makes this alternative attractive for practical applications when no detector data is available, besides no previous traffic information at the desired location is required to obtain its flow profile.
