| Summary: | The need to measure urban link travel time (ULTT) is becoming increasingly important for the purposes both of network management and traveller information provision. This thesis develops a methodology by which data from single inductive loop detectors (ILDs) can be used to derive estimates of ULTT. Research is then undertaken to assess how effectively travel time data from this ULTT model can be used to model travel time variability (TTV). This research first looks at the issue of data quality. The Overtaking Rule method is developed to clean travel time data derived from ANPR cameras in London. A way of quantifying the effectiveness of ILD data cleaning treatments is proposed and the Daily Statistics Algorithm is identified as the best treatment. There are various limitations in existing ULTT models. To address these limitations, an alternative k nearest neighbors (k-NN) based method is proposed for use as a ULTT model. The key design parameters are identified and a 5-step process to optimise this model proposed. The k-NN method is found to outperform all other ULTT models on two links in central London. It is also found to be robust to changes in network characteristics and does not necessarily have to be optimised for each link. Data from the k-NN ULTT model allows various components of TTV to be modelled. In particular, it allows the true patterns of day-to-day variation to be identified. It also correctly models the distribution of travel time in approximately 50% of 15-minute time periods. As an application, the k-NN method could be used to aggregate GPS travel time records from different times but the same underlying travel time distribution together to enable more accurate estimates of ULTT.
|
|---|
