Forecasting air pollution is an important activity for developing sustainable and smart cities. Generated by various sources, air pollutants distribute in the atmospheric environment due to the complex dispersion processes. The emerging sensor and data technologies have promoted the development of data-driven approaches to replace conventional physical models in urban air pollution forecasting. Nevertheless, it is still challenging to capture the intricate spatial and temporal patterns of air pollutant concentrations measured by heterogeneous sensors, especially for long-term prediction of the multi-variate time series data. This paper proposes a deep learning framework for longer-term forecast of air pollutants concentrations using air pollution sensing data, based on a conceptual framework of meta-graph deep learning. The key modules in the framework include meta-graph units and fusion layers, which are designed to learn temporal and spatial correlations respectively. A detailed case was formulated for forecasting air pollutants in Stockholm using air quality sensing data, meteorological data and so on. Experiments were conducted to evaluate the performance of the proposed modelling framework. The computational results show that it outperforms the baseline models and conventional deterministic dispersion models, demonstrating the potential of the framework to be deployed for the real air quality information systems in Stockholm.

Traffic state estimation is an important part of the traffic control process and aims to creates an accurate understanding of the current situation in traffic system. Bayesian Filtering is a statistical modelling framework that is useful in representing traffic state update as well as the relation between traffic state and detection data. This study develops a hybrid approach and uses non-parametric Gaussian Process (GP) to model the state-space transition of traffic system. Through representing the system models as either fully data-driven GP or as a hybrid model using a parametric mean function fusing the conventional principle of traffic flow with the data-driven approach, the requirement of an analytical model can be removed or relaxed. The computational results show that the proposed approach for lane based TSE can capture both short-term fluctuations and larger demand changes. In particular, the Bayesian nature of the GP models offer relative ease in quantifying the model uncertainties in combination with a conventional traffic flow model.

This paper proposes a parallel recommendation engine, PRECOM, for traffic control operations to mitigate congestion of road traffic in the metropolitan area. The recommendation engine can provide, in real-time, effective and optimal control plans to traffic engineers, who are responsible for manually calibrating traffic signal plans especially when a road network suffers from heavy congestion due to disruptive events. With the idea of incorporating expert knowledge in the operation loop, the PRECOM system is designed to include three conceptual components: an artificial system model, a computational experiment module, and a parallel execution module. Meanwhile, three essential algorithmic steps are implemented in the recommendation engine: a candidate generator based on a graph model, a spatiotemporal ranker, and a context-aware re-ranker. The PRECOM system has been deployed in the city of Hangzhou, China, through both offline and online evaluation. The experimental results are promising, and prove that the recommendation system can provide effective support to the current human-in-the-loop control scheme in the practice of traffic control, operations, and management. 

Agent-based approaches have gained popularity in engineering applications, but its potential for advanced traffic controls has not been sufficiently explored. This paper presents a multi-agent framework that models traffic control instruments and their interactions with road traffic. A constrained Markov decision process (CMDP) model is used to represent agent decision making in the context of multi-objective policy goals, where the policy goal with the highest priority becomes the single optimization objective and the other goals are transformed as constraints. A reinforcement learning-based computational framework is developed for control applications. To implement the multi-objective decision model, a threshold lexicographic ordering method is introduced and integrated with the learning-based algorithm. Moreover, a two-stage hybrid framework is established to improve the learning efficiency of the model. While the proposed approach is potentially applicable for different road traffic operations, this paper applies the framework for traffic signal control in a network of Stockholm based on traffic simulation. The computational results show that the proposed control approach can handle a complex case of multiple policy requirements. Meanwhile, the agent-based intelligent control has shown superior performance when compared to other optimized signal control methods.

An accurate and practical traffic-state estimation (TSE) method for signalized intersections plays an important role in real-time operations to facilitate efficient traffic management. This paper presents a generalized modeling framework for estimating traffic states at signalized intersections. The framework is non-parametric and data-driven, without any requirement on explicit modeling of traffic flow. The Bayesian filter (BF) approach is the core of the framework and introduces a recursive state estimation process. The required transition and measurement models of the BFs are trained using Gaussian process (GP) regression models with respect to a historical dataset. In addition to the detailed derivation of the integration of BFs and GP regression models, an algorithm based on the extended Kalman filter is presented for real-time traffic estimation. The effectiveness of the proposed framework is demonstrated through several numerical experiments using data generated in microscopic traffic simulations. Both fixed-location data (i.e., loop detector) and mobile data (i.e., connected vehicle) are examined with the framework. As a result, the method shows good performance under the different traffic conditions in the experiment. In particular, the approach is suitable for short-term estimation, a challenging task in traffic control and operations.

A left turn could be treated as a permitted or protected left turn. Good left turn treatment could improve the efficiency at intersections. This paper aims to solve the left turn phasing type problem with the goal of minimizing the total delay. The determination of left turn phasing types extends the lane-based signal optimization method by introducing the decision variables of left turn phasing type indicators. This problem is solved with a genetic algorithm with penalty functions. It is found that permitted left turns contribute to total delay reduction.

Predicting traffic speed is of importance in transportation management. Signalized road networks manifest highly dynamic speed patterns that are challenging to model and predict. We propose a hybrid deep-learning-based approach for link speed prediction, aiming at capturing heterogeneous spatiotemporal correlations between road intersections. After transforming original road networks and intersections into graphs, this approach leverages a layered graph convolution network structure to model traffic speed variations at both intersection and road network levels. The two levels are combined through a fully connected neural layer. Neural spatiotemporal attention mechanisms are applied to modulate the most relevant periodical traffic information during signal cycles. The proposed approach was evaluated using real-world speed data collected in Hangzhou City, China. Experiments demonstrate that the proposed approach can offer a scalable and effective solution for predicting short-term speed for signalized road networks.

Traffic signal control plays a crucial role in traffic management and operation practices. In the past decade, adaptive signal control systems, capable of adjusting control schemes in response to traffic patterns, have shown the abilities to improve traffic mobility. On the other hand, the negative impacts on environments by increased vehicles attract increased attentions from traffic stakeholders and the general public. Most of the prevalent adaptive signal control systems do not address energy and environmental issues. The present paper proposes an adaptive signal control system capable of taking multi-criteria strategies into account. A general multi-agent framework is introduced for modeling signal control operations. The behavior of each cognitive agent is modeled by a Constrained Markov Decision Process (CMDP). Reinforcement learning algorithms are applied to solve the MDP problem. As a result, the signal controller makes intelligent timing decisions according to a pre-defined policy goal. A case study is carried out for the stage-based control scheme to investigate the effectiveness of the adaptive signal control system from two perspectives, traffic mobility and energy efficiency. The control approach can be further applied to a large network in a decentralized manner. 

Nowadays, traffic congestion poses critical problems including the undermined mobility and sustainability efficiencies. Mitigating traffic congestions in urban areas is a crucial task for both research and in practice. With decades of experience in road traffic controls, there is still room for improving traffic control measures; especially with the emerging technologies, such as artificial intelligence (AI), the Internet of Things (IoT), and Big Data. The focus of this thesis lies in the development and implementation of enhanced traffic signal control systems, one of the most ubiquitous and challenging traffic control measures.

This thesis makes the following major contributions. Firstly, a simulation-based optimization framework is proposed, which is inherently general in which various signal control types, and different simulation models and optimization methods can be integrated. Requiring heavy computing resources is a common issue of simulation-based optimization approaches, which is addressed by an advanced genetic algorithm and parallel traffic simulation in this study.

The second contribution is an investigation of an intelligent local control system. The local signal control operation is formulated as a sequential decision-making process where each controller or control component is modeled as an intelligent agent. The agents make decisions based on traffic conditions and the deployed road infrastructure, as well as the implemented control scheme. A non-parametric state estimation method and an adaptive control scheme by reinforcement learning (RL) are introduced to facilitate such an intelligent system.

The local intelligence is expanded to an arterial road using a decentralized design, which is enabled by a hierarchical framework. Then, a network of signalized intersections is operated under the cooperation of agents at different levels of hierarchy. An agent at a lower level is instructed by the agent at the next higher level toward a common operational goal. Agents at the same level can communicate with their neighbors and perform collective behaviors.

Additionally, a multi-objective RL approach is in use to handle the potential conflict between agents at different hierarchical levels. Simulation experiments have been carried out, and the results verify the capabilities of the proposed methodologies in traffic signal control applications. Furthermore, this thesis demonstrates an opportunity to employ the systems in practice when the system is programmed on an intermediate hardware device. Such a device can receive streaming detection data from signal controller hardware or the simulation environment and override the controlled traffic lights in real time.

Increasing traffic congestion poses significant challenges for urban planning and management in metropolitan areas around the world. One way to tackle the problem is to resort to the emerging technologies in artificial intelligence. Traffic light control is one of the most traditional and important instruments for urban traffic management. The present study proposes a traffic light control system enabled by a hierarchical multi-agent modeling framework in a decentralized manner. In the framework, a traffic network is decomposed into regions represented by region agents. Each region consists of intersections, modeled by intersection agents who coordinate with neighboring intersection agents through communication. For each intersection, a collection of turning movement agents operate individually and implement optimal actions according to local control policies. By employing a reinforcement learning algorithm for each turning movement agent, the intersection controllers are enabled with the capability to make their timing decisions in a complex and dynamic environment. In addition, the traffic light control operates with an advanced phase composition process dynamically combining compatible turning movements. Moreover, the collective operations performed by the agents in a road network are further coordinated by varying priority settings for relevant turning movements. A case study was carried out by simulations to evaluate the performance of the proposed control system while comparing it with an optimized vehicle-actuated control system. The results show that the proposed traffic light system, after a collective machine learning process, not only improves the local signal operations at individual intersections but also enhances the traffic performance at the regional level through coordination of specific turning movements.