SE PhD Final Defense of Jing Zhang

  • Starts: 11:00 am on Wednesday, September 6, 2017
  • Ends: 1:00 pm on Wednesday, September 6, 2017
TITLE: Detection and Optimization Problems with Applications in Smart Cities

ABSTRACT: We propose solutions to a selected set of detection and optimization problems, whose applications are focused on transportation systems. Our goal is to help build smarter and more efficient transportation systems, hence smarter cities.

We consider problems with dynamics evolving in two time-scales:

(i) In a fast time-scale, we consider the problem of detection, especially statistical anomaly detection in real-time. We develop an improved composite hypothesis test that can be efficiently applied as a Generic statistical Anomaly Detector (GAD) for any system whose behavior can be represented by a Markov process. In particular, under Markovian assumptions, for the widely used Hoeffding test we derive a novel threshold estimator as opposed to an existing alternative. We conduct extensive numerical simulations to show that, in a finite sample-size setting, GAD (with our threshold estimator) can typically control false alarms better than existing work. We then apply GAD to detecting non-typical traffic jams in the Boston road network using real traffic data reported by Waze.

(ii) In a slower time-scale, we investigate a host of optimization problems arising in transportation systems. In particular, for a general single-class road network, we investigate the user-optimum Traffic Assignment Problem (TAP), the Origin-Destination demand estimation problem, and the travel latency cost function estimation problem. For this TAP, we also analyze the sensitivity of the total user travel time with respect to road capacities and free-flow travel times, which would help prioritize road segments to reengineer. We formulate a system-optimum TAP to find socially optimal flows. We then investigate the network performance, in terms of the total latency, under a user-optimal routing policy versus a system-optimal one. The ratio of these two quantities is called the Price of Anarchy (PoA) and quantifies the efficiency loss of selfish actions compared to socially optimal ones. We propose data-driven strategies to evaluate and reduce PoA and conduct a case-study on the Eastern Massachusetts road network. Finally, we extend the cost function estimation formulation to multi-class cases and consider the joint problem of recovering the cost functions and adjusting the demand matrices.

COMMITTEE: Advisor: Ioannis Paschalidis, SE/ECE; Christos Cassandras, SE/ECE; Pirooz Vakili, SE/ME Manuel Egele, ECE; Chair: Alex Olshevsky, SE/ECE

110 Cummington Mall, Rm 205

Back to Calendar