Abstract

Spatio-temporal problems arise in broad areas of environmental and transportation systems. These problems are challenging, because of both spatial and temporal neighborhood similarities and correlations. We consider traffic data, which is a complex example of spatio-temporal data. Traffic data is geo-referenced time series data, where fixed locations have observations for a period of time. Traffic data analysis and related machine learning tasks have an important role in intelligent transportation systems, such as designing navigation systems, traffic management, control systems and in the future will be essential for setting appropriate anticipatory tolls. Recent data collection methodologies dramatically increase the volume of available spatio-temporal data, which require scalable machine learning models. Moreover, deep learning models outperform traditional machine learning and statistical models due to their strong feature learning abilities in spatial and temporal domains. Increases in available data and recent advances in deep learning models in spatio-temporal domains are the main motivations of this dissertation.

We first study, non data-driven and optimization-based solutions for the network flow problem, which appears in a wide range of applications including transportation systems and electricity networks. In these applications, the underlying physical layer of the systems can generate a very large graph resulting in an optimization problem with a large decision variable space. We present a distributed solution for the network flow problem. The model uses cycle basis and an alternating direction method of multipliers (ADMM) method to find a lower computational time and number of communications, while obtaining a centrally optimal solution.

Second, we attempt to obtain spatio-temporal clusters in traffic data, which represent similar traffic data in terms of both spatial and temporal similarities. Clustering of traffic data are used to analyze traffic congestion propagation and detection. We obtain spatio-temporal clusters using a modification to Deep Embedded Clustering, which considers both spatial and temporal similarities in latent features. Also we define new evaluation metrics to evaluate spatio-temporal clusters of traffic flow data.

Third, when sensors collect spatio-temporal data in a large geographical area, the existence of missing data cannot be escaped, which negatively impacts of prediction models. Here, we investigate the problem of incorporating both spatial and temporal contexts in missing traffic data imputation using convolutional and recurrent neural networks. We propose a convolutional-recurrent autoencoder for missing data imputation, and illustrate the performance of autoencoders for missing data imputation in spatio-temporal data.

Finally, traffic flow prediction has an important role in diverse intelligent transportation systems and navigational systems. There is a large literature on this problem. However, the problem is challenging for high-dimensional traffic data. We explicitly design the neural network architecture for capturing various types of spatial and temporal patterns. We also define evaluation metrics for spatio-temporal forecasting problems to better evaluate generalization of the model over various spatial and temporal features.

Abstract

Much effort has been made in the past on the supply side to relieve road traffic congestion which undermines the mobility in urban networks and brings heavy social costs, but building additional roadway capacity is no longer considered a viable option. A better alternative is the efficient management of existing networks, for which we can envisage new possibilities that emerge in light of the recent increase in the use of private providers’ digital map and traffic information systems. These systems have evolved mostly without much public sector influence, but some paradigm shift is needed for thinking about the directions of future developments that will show societal benefits also open up private-sector opportunities. In this context, we develop a multi-agent advanced traffic management and information systems (ATMIS) framework with day-to-day dynamics where private agencies are included as traffic information service providers (ISPs) together with public agencies handling the traffic control and the users (drivers) as the decision-makers. One important paradigm shift is that the emergence of private ISPs makes it possible to obtain path-based data via retrieval of individual trajectory diaries and current position information from their subscribers. The availability of such path-based data can bring about the development of new path-based ATMIS algorithms. Such new algorithms can be capable of taking into account the routing effects of advanced traveler information systems (ATIS). Under the assumption that the traffic management center (TMC) has some (even approximate) knowledge of the ISPs’ optimal strategies, it is possible to design optimal route guidance and control strategies (ORGCS) that takes into account the anticipated ISP reactions in terms of route-level flows. In light of these issues, we develop a routing-based real-time cycle-free network-wide signal control scheme (R2CFNet) that uses path-based data. The scheme also allows the avoidance of day-to-day games between ISPs and signal control through the use of weights on the queue delays in the control objective function. The weights are essentially operator parameters designed to incorporate ORGCS and day-to-day behavior. The proposed control scheme, of course, responds to detected traffic (demand) rates on a real-time basis in response to the control delays on network routes. Another theoretical advance in the research is in the development of a modeling scheme that uses a new optimization algorithm for a convergent simulation-based dynamic traffic assignment (DTA) model. This model incorporates a Gradient Projection (GP) algorithm, as opposed to the traditionally-used Method of Successive Averages (MSA), and it displays significantly better convergence characteristics. A consistent day-to-day dynamic framework is also developed, incorporating an elaborate microscopic simulation model to capture traffic network performance, to study network dynamics under multiple private ISPs and the new signal control scheme. The results of parametric simulations have shown that the proposed framework is capable of effectively capturing the effects of the interplay of urban traffic route guidance, network control and user response. It is seen that an appropriate combination of ATIS market penetration rate and the special-purpose signal control settings could divert some portion of travel demand to different routes. This is achieved by constraining the signal settings to conform to certain longer-term strategies. The performance and efficiency of the components of the proposed framework such as the DTA model, the day-to-day dynamics model and the R2CFNet control scheme have been investigated through various numerical experiments that show promising results. Lastly, several future topics of relevance to the framework are discussed.

Abstract

Information and communication technologies (ICTs) are important research areas for scientists examining theories of communication, conflict resolution and collaborative decision-making, particularly because they offer impressive analytical capabilities and the capacity to integrate different modes of deliberation and forms of content. The exponential growth in the adoption and diffusion of these digital media currently has, and will likely continue to have, considerable social ecological implications in part because ICTs are increasingly positioned as places of convergence for politically contested information and knowledge. However, the nature of these implications, especially questions concerning how these technologies influence or mediate changes in policy and/or the policymaking process itself is unclear and controversial. Technological enthusiasts, for example, argue that ICTs have potential to upgrade democracy by improving the way we devise means to clarified ends whereas technological pessimists challenge that, far from ushering in a new age of democracy, new media technologies actually hinder coordinative action by reducing more personalized modes of communication. This research examined both face-to-face and online communication facilitated by three institutions in California–the South Coast Air Quality Management District (SCAQMD), the California Environmental Justice Action Committee (CEJAC) and Communities for a Better Environment (CBE)–as they sought to reach decisions concerning a series of environmental justice-related issues. Informed by a mixed methodological approach, this research characterizes the challenges and opportunities afforded by the traditional face-to-face (F2F) settings hosted by the three organizations (i.e., public hearings, public meetings and workshops, respectively) and communication within these settings differed from and integrated with EJ communication in the institutions’ corresponding new media or ICT-based environments (i.e., general content websites and interactive mapping applications). The research found that, while there were obvious limitations to F2F participation, the pragmatic modes of communication that took place in these three settings were not replicated in the online environment. Most troublesome was that ICT-based communication tended to be less trustworthy, interactive and coherent than corresponding communication in F2F settings. The dissertation concludes by putting forth an alternative ICT-based framework for just environmental governance that enables interdependent, multi-directional and adaptive forms of knowledge production and decision-making.

Abstract

Collecting rich individual trip data at an individual level has long been viewed as a hard task and has become a bottleneck in modeling and calibrating travel behavior models since traditional survey methods are both costly and time-consuming. New technologies make such data a possibility and thus there is a need for frameworks that model individual behavior in real-time using such data. Such modeling will find use in a variety of real-time network optimization and prediction schemes. This dissertation describes the details of plausible behavioral modeling of this kind, and develops new data structures that are needed both for handling the network combinatorics in the analysis and in the data storage. The work is presented in the context of a new technology we propose called the Persistent Traffic Cookie (PTC) system which uses the short range wireless connection between vehicles and road side controllers to store authenticated, time-stamped node sequences on an onboard database. The dissertation makes the premise that traditional travel behavior models, including those based on disaggregate decision paradigms were developed primarily for application in aggregate level prediction and are thus not very applicable for an individual’s route choice prediction in real-time. A scheme that does not require variation of explanatory variables across the choice sets or variation in the individual’s decisions for calibration may be essential. Thus the dissertation developed models based on observed frequencies of decisions. The research also stresses the importance of path and sub-path notions in route choice decisions and provides appropriate data structures that enable modeling with such notions. Two methods that directly query the collected sequence data using efficient data structures based on the suffix tree and the suffix array schemes and node/edge transition probability model, are proposed to predict individual travels from trip diary database. A day-to-day PTC simulation framework with behavior components is proposed to generate consistent PTC data and implemented in Paramics microscopic traffic simulator. Day-to-day PTC simulations are carried out for two Paramics networks, including the Irvine Triangle network, which is a well-calibrated real world network. Various scenarios are created to test the sensitivities of the proposed prediction methods. The simulation results shows that it seems the prediction methods are robust with regard to the underlying behavior models, traffic conditions and tracking periods.