Network-wide road travel time estimation with inconsistent sensor data

This paper proposes a new modeling approach for estimating road travel times in the territory-wide network. Owing to the rapid development of Intelligent Transportation Systems, travel times can be observed on links and/or paths in the road network by different sensor systems. The observed data from multiple sensor systems can provide more comprehensive and complete information on network-wide travel times than those from one sensor system. Thus, the use of the multiple sensor systems may have the potential for more accurate estimation of travel times in the territory-wide road network. However, the traditional travel time estimation methods typically consider the observed data from only one sensor system. The reason is that the observed data from different sensor systems may be inconsistent, which may bring difficulties in using these inconsistent input data for travel time estimation. This paper proposes a new two-stage model to overcome such data inconsistency difficulty for travel time estimation in the territory-wide road network with partial observed data from multiple sensor systems. In the first stage, the link travel time covariance matrix estimation model is proposed to estimate the covariance matrix of the link travel times for the whole network with the use of the second-order statistical property of the observed data from multiple sensor systems. Then, mean travel times on the links without sensors are updated according to the resultant estimated covariance matrix. In the second stage, a generalized least squares model is adopted to explicitly consider the data inconsistency issue for estimating the network-wide mean link and path travel times using the first-order statistical property of the observed data from multiple sensor systems. A heuristic solution algorithm is proposed for solving the two-stage model. A local road network in Hong Kong is used to demonstrate the applications of the proposed modeling approach.