About the Seminar
This Civil Systems Graduate Student Research Seminar intends to share progress made by graduate students who are conducting research within the umbrella of the Civil Systems group. It aims to address the emerging complexity of civilian infrastructure through the application of systemic tools. Such systems often have a "personality" that cannot be perfectly predicted from an analysis of their parts. This is often true for the large systems that form the infrastructure of our society, but it is also true of much smaller systems we design, build and use every day. Each week, speakers will discuss their research, and show how the tools they are applying could be beneficial to a broader audience. This seminar may also be especially useful to students of Civil Engineering who are interested in broadening their exposure to tools and methods taught outside the Civil curriculum.
If you have any questions, or are interested in giving a presentation please contact Branko Kerkez firstname.lastname@example.org.
Civil Systems Seminar: Usually on Wednesday's at 5:15PM (disregard Berkeley time) in 240 Sutardja Dai Hall (CITRIS Building). Please see specific speaker for room/time change.
11.09.11 Jack Reilly: Optimization of Highway Networks. 5:15PM 240 CITRIS.
Abstract: Mobile and location-based applications have enabled traffic engineers to give information and apply control to individual drivers as opposed to general policies. A unique problem emerges from the fact that control can only be applied to the fraction of drivers with the application enabled. Traditional literature on optimal routing strategies on highway networks typically assume full compliance of users on the network. While progress on game-theoretic models of behavior on partial-compliance systems has been made in the field of computer science, the models used for edge delays are vertical-queueing models, which do not capture some of the fundamental characteristics of traffic behavior. A rigorous analysis is done for partial-compliance optimization on simple networks in the framework of Stackelberg games using horizontal-queueing models. In order to consider more complex network configurations, we propose a steady-state horizontal-queueing model of network traffic flow based on the discretized LWR partial differential equation. We propose an framework for modelling mixed Eulerian-Lagrangian flow information problems on a partial-compliance network. In order to acknowledge the effect of leader-follow game-theoretic behavior inherent to partial-compliance networks, a unique optimization problem is formulated to create optimal routing assignments for compliant users.
Bio: I graduated from UCLA in 2009 with a civil engineering degree. I have worked on the iShake earthquake mobile sensing project, Clearsky vehicular emissions estimation project, and have recently been working on optimization and control of routing on highway networks. My interests include machine learning, network optimization, mobile computing. Outside of academia, I spend most of my time trying to convince others that Barenaked Ladies are the most innovative musicians of our era.
11.02.11 Branko Kerkez: Massive-scale wireless sensor networks for hydrologic monitoring. 5:15PM 240 CITRIS.
Abstract: Recent water shortages, particularly evident in the state of California, emphasize the need for a better hydrologic understanding, and improved water management techniques. The majority of the state's water originates in the Sierra Nevada as snow, melting throughout the year to provide various stakeholders with water. Current measurement techniques are unable to resolve variability of the snowpack at the basin scale, and snowmelt processes are not well captured by existing hydrologic models.
I'll be talking about a system-level solution to facilitate scientific understanding, and water management decisions in basins of the Sierra Nevada. The core of the proposal focuses on improving and expanding current sensing methods, while developing the tools necessary to analyze the resulting data. I'll talk about the viability of wireless sensor networks as a distributed real-time measurement platform. I'll also briefly cover some estimation techniques, and modeling methods that can be used in tandem with such networks.
Bio: Branko is a 5th year systems PhD student working with Porf. Glaser. His research interests include wireless sensor networks, and the applications of machine learning and control theory to the study of environmental phenomena.
Note: Talk moved to Friday 1pm Room 254 CITRIS
04.09.10 Timothy Hunter: Traffic Estimation and Machine Learning
Abstract: We consider the problem of estimating real-time traffic conditions in large
cities from sparse, noisy GPS probe vehicle data. We focus on estimating
historical trafﬁc patterns with some extensions to real-time trafﬁc
conditions. We assume that the data available for these estimation problems is
a small set of sparsely traced vehicle trajectories, which represents a small
fraction of the total vehicle flow through the network. We present an
expectation maximization algorithm that simultaneously learns the likely paths
taken by probe vehicles as well as the travel time distributions through the
network. A case study using data from San Francisco taxis is used to
illustrate the performance of the algorithm.
Bio:Tim is a first year PhD student in Computer Science working with Professor
02.24.10 Greg Mclaskey: Vibrating solids and sensors as linear systems
Abstract: A system is sometimes thought of as a black box defined by its inputs and outputs. This idea of a system can be used to characterize the behavior of things as simple as the bending of a beam to the complicated behavior of a building subjected to an earthquake, the response of a sensor to accelerations, or the amplification of ground motions in unconsolidated soils. In my research, I use the basic theoretical framework of a linear system to understand the behavior of nanoseismic sensors and the way that stress waves propagate in a solid body. This involves both the careful selection of experimental techniques and analytical and numerical solutions to the PDE which models wave propagation. The result is the absolute calibration of a sensor which records surface vibrations of a solid down to pm in amplitude in the frequency range of 20 kHz to over 1 MHz.
Bio: Greg is a fifth year PhD student working with Prof. Steven Glaser. His work involves the study of friction and fault rupture using an array of nanoseismic sensors attached to a laboratory sized fault.
03.05.10 Jerry Jariyasunant: Real-Time Data in Public Transportation
Abstract: Recent developments in transportation technology coupled
with the widespread usage of smartphones have lead to much attention
to new research in the distribution of information to help people
travel and make mode choice decisions. New data has come from many
sources: Governments have opened up previously protected public
transportation data, Crowd-sourced traffic, transit, and mapping data
has become more popular, and GPS devices have been put into buses,
trains, and taxis. As a result, new applications have been developed
to leverage the opening up of this wealth of data. The talk will be a
review of previous research done to evaluate the benefits of real-time
data in a mobile transit trip planning application as well as future
work discussing the use of smartphones and real-time data in mode
Bio: Jerry is a fourth year PhD student working with Prof. Raja Sengupta.
03.10.10 Sebastien Blandin: Traffic estimation using graphical models
Abstract: Traffic estimation using flow models is a common practice in the transportation community
and is one of the main components of the Mobile Millennium system. Since these types of
models require the knowledge of real time boundary conditions, they cannot be directly used for
traffic forecast. We propose a graphical model of traffic for variable horizon traffic forecast.
An inclusion-optimal structure of the graphical model
is learned in an offline phase from the output of a flow model estimation algorithm,
and used to forecast traffic
speeds at different horizons. Preliminary results and current findings are presented along with
a discussion of possible combinations of machine learning algorithms with classical models.
Bio: Sebastien is a 2nd year PhD student with Prof. Alex Bayen. His work involves traffic modeling,
estimation and control under uncertainty.
03.19.10 Dan Work: Real-time estimation of distributed parameters systems - application to traffic monitoring
Abstract: The coupling of the physical world with information technology promises to help meet increasing demands for efficient, sustainable, and secure management of our built infrastructure and natural environment. A mathematical abstraction of the physical environment can be achieved the form of distributed parameters systems described by partial differential equations. Yet, initial and boundary conditions, and other model parameters necessary for complete characterization of the model are often unknown, driving the need for distributed sensing of the physical environment. Because of the nonlinearities and distributed nature inherent to these processes, efficient estimation algorithms to reconcile modeling and measurement errors in real-time remains an open challenge for many applications.
This work investigates the problem of real-time estimation of distributed parameters systems in the context of monitoring traffic. The recent explosion of cell phones with Internet connectivity and GPS are rapidly increasing sensor coverage on roadways – with a catch. GPS velocity measurements, further degraded to preserve user anonymity, cannot be easily iterated into the well known density-based network of partial differential equations typically used to describe traffic. This challenge is circumvented by transforming the density-based traffic model into a new but equivalent evolution equation for velocity, which retains the nonlinearity and non- differentiability of traffic due to shocks. Because of this transformation, GPS velocity measurements become a direct observation of the state. The resulting state estimation problem is then solved in real-time for large road networks with an ensemble Kalman filtering algorithm. This approach has been implemented and forms the backbone of the Mobile Millennium traffic monitoring system at UC Berkeley, which has been deployed in Northern California for more than a year. Promising extensions to other domains will also be discussed.
Bio: Dan is a 4th year PhD student with Prof. Alex Bayen.
04.02.10 Branko Kerkez: Hybrid systems for environmental modeling and analysis
Abstract: Many real-life phenomena, particularly those witnessed in the natural environment, exhibit complicated nonlinear dynamics. This behavior makes modeling, analysis, and estimation of such systems a difficult task, both in terms of mathematical and computational complexity. Upon closer inspection, however, many such systems can be decoupled into a set of discrete states with distinct continuous dynamic. In such cases, the Hybrid Systems framework can be used to model the underlying phenomenon. This provides a comprehensive mathematical framework for analysis, while preserving physical intuition. A motivating example of a hybrid system relates to snow processes in the Sierra Nevada Mountains. It is estimated that snowmelt is the primary source of water for 60 million people in the western Unites States. Snowmelt is a nonlinear phenomenon, which requires the solution of nonlinear PDEs to be modeled properly. The nature of these equations prohibits closed form solutions, and ultimately requires solutions via finite element schemes. This often places strain on computational infrastructure and has the effect of completely removing intuition regarding the physical phenomenon.
This project is part of an initiative to monitor large-scale environmental processes in the Sierra Nevada using wireless sensing technology. This talk will give a brief introduction to hybrid systems, and will motivate their use with a series of real-world examples. Furthermore, it will be shown how hybrid systems can be employed to accurately model snowmelt process given the real-time input of wireless sensor networks.
Bio: Branko is a 3rd year systems PhD student working with Porf. Glaser. His research interests include wireless sensor networks, and the applications of machine learning and control theory to the study of environmental phenomena.