Traffic conditions in signalized junctions are highly dynamic and may be subject to abrupt changes due to unanticipated traffic incidents or network obstructions. These abrupt changing conditions are represented as different regimes or modes where each mode is represented by its own distinct model, forming a set of multiple models. At any instance in time, only one model of the set has the potential of representing the physical system dynamics at that time. However the dynamics may arbitrarily jump over to a different regime when an abnormal condition arises. Furthermore, it might be impossible to identify these modelsa priori. Hence, a multiple model approach is developed to self-detect these abrupt changes, identify which member of the set best represents the actual system and automatically self-configure and add a new model to the set when a previously unmodelled regime arises. This approach makes use of a real-time joint (dual) estimation algorithm to estimate traffic state variables such as queue lengths and traffic flow, as well as model parameters such as turning ratios, saturation flow values and noise covariance resulting from unmodelled dynamics and measurement errors. The proposed algorithm is validated through simulations on signalized 3-arm and 4-arm junctions with typical day-to-day traffic conditions including several network irregularities occuring at different times of the day such as arm closures as a result of traffic incidents. This work is aimed to form part of adaptive control loops for traffic light systems that are able to autonomously adjust to changing traffic conditions so as to ensure efficient vehicle flows.
展开▼
