Stochastic filtering for real-time traffic state estimation using multiple data sources.

Abstract

The rapid population growth and urbanization in many parts of the world have resulted in an increase in traffic demand over the last few decades. However, expanding and developing transportation infrastructure is insufficient and sometimes impractical to meet the increasing need. Instead, the existing road networks should be used more efficiently in conjunction with a shift to more sustainable means of transport. In order to accomplish this, accurate information about the current and future traffic state is essential as it is a crucial input to traffic management systems and can help travellers make more informed decisions. The traffic state information is usually obtained from a process known as traffic state estimation (TSE) using data collected from traffic sensors. In addition to fixed loop detectors, modern sensors like GPS, Bluetooth, and mobile phones are being used to collect data. To take advantage of these data types, existing TSE methods need to be adapted or new methods developed. Of particular interest in this thesis is incorporating multiple data sources efficiently to improve the accuracy of traffic estimation for optimal utilization of transportation infrastructure in general.

This thesis’s main objective is to develop novel stochastic filtering frameworks for real-time traffic state estimation in existing road networks, mainly focusing on urban roads using multiple data sources. Several methods are proposed to handle some challenging problems in TSE and data fusion, such as different spatial and temporal resolutions among data sources and missing data. At first, an incremental Unscented Kalman Filter (UKF) for effective multi-source state estimation on motorways is developed. Next, we design a novel estimation framework for urban arterials that accounts for the stochasticity of the boundary flows and traffic parameters in a traffic model. In terms of the traffic model, we introduce a new stochastic Switching Mode Model (SMM) applied to arterial roads with both signalized and unsignalized intersections. This model does not consider stochasticity in the current mode of operation, which substantially reduces its computational complexity because there is only one mode at each time step. Furthermore, we propose three different filtering algorithms for multi-source traffic estimation, including the incremental stochastic Kalman Filter (SKF), the incremental stochastic Unscented Kalman Filter (SUKF), and a hybrid approach. Finally, we improve the urban estimation framework by taking mode stochasticity into account using the interacting multiple model (IMM) estimator. However, the use of the IMM in the existing SMM cannot utilize many types of data, which leads to the emergence of a new mode system for the SMM and the proposal of IMM fusion algorithms. Based on these methods, we present a hybrid IMM and link-independent IMM methods that improve estimation accuracy and reduce the computational complexity of the frameworks.