Towards autonomic control of urban traffic junctions

Abstract

As traffic demands are increasing and reaching critical levels worldwide, advanced traffic signal management is becoming an important requirement. The evolution and generation of traffic signal control concepts integrated with advances in control, communications and computational technologies provide intelligent control of traffic lights that adapt themselves to meet the time-varying traffic demands or to changing road conditions. Despite these recent advances, current systems can become suboptimal because the controller parameters of such systems are not tuned to changing traffic behaviour. Hence such systems can fail when networks are subject to major unanticipated irregularities, such as roadworks, accidents and extreme weather conditions. An adaptive system which can self-tune and adjust the controller parameters to adapt to changing traffic conditions is required. The need to design selfmanaging systems, which self-handle the complexity and uncertainties and thus reduce human intervention to a minimum is of utmost importance. Autonomic systems can self-handle these complexities by modelling the network behaviour and adapting to the changes as required. Hence this work is directed towards the development of autonomic systems for urban signalized junctions. The aims of this research are: i) to model in real-time the traffic dynamics within signalized junctions, with little prior knowledge of the underlying traffic parameters. In literature, the model parameters are typically assumed known apriori from past traffic measurements or by applying nonlinear recursive estimation algorithms. However, difficulties with nonlinear estimation algorithms such as divergence issues, motivated the development of joint state and parameter estimation algorithms from a different perspective, by developing novel variations on the expectation-maximization algorithm. A quasi real-time joint estimation of model states, parameters and noise is first proposed, whereby the standard expectation-maximization algorithm is modified to carry out estimation in quasi real-time, making use of a short uniform window of sensor measurements of fixed time length, which looks back in time. Full real-time algorithms are developed to reflect changing traffic conditions making use of Robbins-Monrostochastic approximation posed first as a single variant estimation algorithm, followed by a multivariate estimation algorithm. These proposed algorithms require partial derivative of the likelihood function with respect to each parameter to be worked out analytically. In practice, this approach is impractical for larger and complex junctions due to the complexity of the derivatives involved in deriving such equations. Hence a derivative-free approach is proposed, allowing for easier generalisation to other junctions and scalability to more complex junctions; ii) develop real-time multiple model adaptive estimation methods to estimate states and parameters of signalized traffic junctions suitable for structurally diverse dynamic regimes arising abruptly from for example, road works. Several proposed adaptive systems in literature, relied on traffic surveillance technologies to warn the commuters of any detected irregularities and in most cases rely on human experience to evaluate the impact on the network performance and to provide route diversion recommendations that might not guarantee the optimal use of the available network capacity. A less infrastructure demanding multiple model adaptive estimation method is proposed to estimate states and parameters of signalized traffic junctions subject to time-varying parameters, jump dynamics and unpredictable disturbances; iii) the development of an autonomic control scheme. Although many so-called “adaptive” systems were proposed in literature, the controller parameters of these systems are not autonomously tuned to changing traffic behaviour so that the controller is able to adapt itself to changing traffic conditions and maintain optimal levels of performance. Hence, a truly adaptive system which can self-tune and adjust the controller parameters to adapt to changing traffic conditions is developed in this work. This is based upon model predictive control using linear and quadratic programming optimization techniques; iv) to integrate the latter two novelties (multiple model adaptive estimation and autonomic control) to obtain a multiple model adaptive control scheme for jump structural changes in junction dynamics. This scheme is validated by simulating typical signalized 3-arm and 4-arm junctions with arms and lane closure, with the ability to autonomously adjust to changing traffic conditions.