Dissertations and Theses - Electrical and Computer Engineering
http://hdl.handle.net/2142/8888
Dissertations and Theses in Electrical and Computer EngineeringTue, 20 Mar 2018 07:49:53 GMT2018-03-20T07:49:53ZModeling and management of dynamic loads in power systems
http://hdl.handle.net/2142/99097
Modeling and management of dynamic loads in power systems
Zhang, Kaiqing
Recent advances in power systems have led to the proliferation of dynamic, diverse, and even flexible loads in the system operations. An accurate as well as identifiable model that is able to characterize the dynamics of loads is of paramount importance for various power system operational tasks. Towards the goal of advanced load modeling, we are particularly interested in modeling this type of dynamic load, a diverse category of loads that pose different challenges in different contexts of power system operations. In this thesis, improved dynamic load modeling approaches are developed and analyzed for two critical operational tasks in power systems: transient stability analysis and demand side management.
As regards transient stability analysis, one newly proposed load model structure, the WECC composite load model (CMPLDW), is investigated for its complexity with an large number of parameters to identify. We verify the underlying parameter redundancy stemming from the insensitivity and interdependency of these parameters. A general framework is then put forward to effectively visualize the redundancy and exhibit the identifiability issues of this load model. Furthermore, an improved parameter estimation scheme is developed by regularizing the nonlinear least squares error objectives in the measurement-based modeling approach. The effectiveness of the proposed dependency analysis and parameter estimation scheme is validated using both synthetic and real measurement data.
In demand side management, one appealing objective of load modeling is to explore its spatio-temporal variability and flexibility for socially economic benefits. To this end, the demand of loads can be managed by pricing signals, i.e., the loads are modeled as price-responsive ones. In particular, here we consider one primary type of dynamic load, the charging load of electric vehicles (EV) en route. To comply with the time-varying property of EV travel demand, we integrate the characterization of EV traffic flow into the modeling of charging loads. Therein the power and transportation networks are coupled to jointly maximize the total social welfare of both systems. Additionally, to achieve the maximum social welfare, an optimal pricing scheme that preserves the privacy of the two infrastructure networks is developed. Through extensive numerical tests, the proposed pricing is shown to outperform other pricing schemes that fail to consider either the interaction of the two networks or the time-varying property of EV travel demand.
Load modeling; Model identification; Dynamic load; Clustering; Nonlinear least-squares; Electric vehicle charging; Demand response; Dual-decomposition
Fri, 14 Jul 2017 00:00:00 GMThttp://hdl.handle.net/2142/990972017-07-14T00:00:00ZZhang, KaiqingDefending distributed systems against adversarial attacks: consensus, consensus-based learning, and statistical learning
http://hdl.handle.net/2142/98441
Defending distributed systems against adversarial attacks: consensus, consensus-based learning, and statistical learning
Su, Lili
A distributed system consists of networked components that interact with each other in order to achieve a common goal. Given the ubiquity of distributed systems and their vulnerability to adversarial attacks, it is crucial to design systems that are provably secured. In this dissertation, we propose and explore the problems of performing consensus, consensus-based learning, and statistical learning in the presence of malicious components.
(1) Consensus: In this dissertation, we explore the influence of communication range on the computability of reaching iterative approximate consensus. Particularly, we characterize the tight topological condition on the networks for consensus to be achievable in the presence of Byzantine components. Our results bridge the gap of previous work.
(2) Consensus-Based Learning: We propose, to the best of our knowledge, consensus-based Byzantine-tolerant learning problems: Consensus-Based Multi-Agent Optimization and Consensus-Based Distributed Hypothesis Testing. For the former, we characterize the performance degradation, and design efficient algorithms that can achieve the optimal fault-tolerance performance. For the latter, we propose, as far as we know, the first learning algorithm under which the good agents can collaboratively identify the underlying truth.
(3) Statistical Learning: Finally, we explore distributed statistical learning, where the distributed system is captured by the server-client model. We develop a distributed machine learning algorithm that is able to (1) tolerate Byzantine failures, (2) accurately learn a highly complex model with low local data volume, and (3) converge exponentially fast using logarithmic communication rounds.
Fault-tolerance; Adversaries; Security; Consensus; Optimization; Hypothesis testing; Statistical learning
Wed, 12 Jul 2017 00:00:00 GMThttp://hdl.handle.net/2142/984412017-07-12T00:00:00ZSu, LiliA series-stacked architecture with isolated server-to-bus converters for high-efficiency data center power delivery
http://hdl.handle.net/2142/98440
A series-stacked architecture with isolated server-to-bus converters for high-efficiency data center power delivery
Zhang, Yizhe
Series-stacked server power delivery architectures have been proposed recently that can achieve much higher energy efficiency than conventional power delivery architectures. When servers are connected in series, differential power processing (DPP) converters can be used to regulate the server voltages when the servers are consuming different amounts of current. Server-to-bus DPP architecture has unique advantages among several other DPP architectures such as being able to achieve the minimum power processed in the DPP converters, and having a higher reliability than other DPP architectures. This work presents the development of a server-to-bus DPP architecture for server power delivery. The hardware prototype is built with four 4-to-1 isolated DPP converters with GaN switches. Four 12V 120W Dell servers are used in the bench test to validate the operation of server-to-bus DPP. 98.99% efficiency is achieved while the servers are running a real-life data center computational load.
Power electronics; Data center; Power delivery architecture; Series-stacking; Differential power processing (DPP); Direct current to direct current (DC-DC) conversion
Fri, 21 Jul 2017 00:00:00 GMThttp://hdl.handle.net/2142/984402017-07-21T00:00:00ZZhang, YizheAutomated testing and machine-learning-based modeling of air discharge ESD
http://hdl.handle.net/2142/98434
Automated testing and machine-learning-based modeling of air discharge ESD
Sagan, Sam
An IEC 16000-4-2 compliant, high-accuracy air-discharge automation system is used to study the properties of air discharge electrostatic discharge (ESD). This work corroborates conclusions of previous works and presents new insights into the effects of approach speed on ESD. A methodology for machine-learning-based ESD modeling is presented. Models are validated with a high degree of accuracy against measurement data.
Air discharge; Electrostatic discharge (ESD); Machine learning
Thu, 20 Jul 2017 00:00:00 GMThttp://hdl.handle.net/2142/984342017-07-20T00:00:00ZSagan, SamA novel weighted rank aggregation algorithm with applications in gene prioritization
http://hdl.handle.net/2142/98424
A novel weighted rank aggregation algorithm with applications in gene prioritization
Raisali, Fardad
We propose a new family of algorithms for bounding/approximating the optimal solution of rank aggregation problems based on weighted Kendall distances. The algorithms represent linear programming relaxations of integer programs that involve variables reflecting partial orders of three or more candidates. Our simulation results indicate that the linear programs give near-optimal performance for a number of important voting parameters, and outperform methods based on PageRank and Weighted Bipartite Matching. Finally, we illustrate the performance of the aggregation method on a set of test genes pertaining to the Bardet-Biedl syndrome, schizophrenia, and HIV and show that the combinatorial method matches or outperforms state-of-the art algorithms such as ToppGene.
Weighted Kendall; Rank aggregation; Linear programming
Wed, 19 Jul 2017 00:00:00 GMThttp://hdl.handle.net/2142/984242017-07-19T00:00:00ZRaisali, FardadDesign of fully digital inductors for low bandwidth filter applications
http://hdl.handle.net/2142/98402
Design of fully digital inductors for low bandwidth filter applications
Salz, Braedon Lenox
We are currently living in the age of intelligent machines, where we are interested in acquiring data and making decisions all from some sort of embedded environment. Of particular value are personal health metrics, such as the analysis of heart rate, muscle action potentials, and brain waves. Collecting this data requires new advances in the circuitry behind much of classical filter design. In this thesis, we present a digital inductor based on time-domain signal processing. This approach uses the phase-domain theory that is well-known and understood in the fields of clocking and serial links and applies it to analog circuit design. By using a ring oscillator to integrate the input voltage and a switched transconductor to inject current into the input node, the proposed time-domain gyrator achieves inductive input impedance without using either large resistors or capacitors. Realizing the gyrator in this manner makes it significantly more amenable for technology scaling. Fabricated in 65 nm CMOS process, the inductor operates from a 0.7 V supply voltage and consumes 528 μW. Measurement results show inductance values in the range of 150 μH to 1.5 mH can be achieved.
Inductor; Phase-domain; Time-domain signal processing
Mon, 17 Jul 2017 00:00:00 GMThttp://hdl.handle.net/2142/984022017-07-17T00:00:00ZSalz, Braedon LenoxAlgorithms for interactive, distributed and networked systems
http://hdl.handle.net/2142/98367
Algorithms for interactive, distributed and networked systems
Bojja Venkatakrishnan, Shaileshh
In recent years, massive growth in internet usage has spurred the emergence of complex large-scale networking systems to serve growing user bases, bandwidth and computation requirements. For example, data center facilities -- workhorses of today's internet -- have evolved to house upward of several hundreds of thousands of servers; content distribution networks with high capacity and wide coverage have emerged as a de facto content dissemination modality, and peer-to-peer applications with hundreds of thousands of users are increasingly becoming popular. At these scales, it becomes critical to operate at high efficiencies as the price of idling resources can be significant. In particular, the interaction between agents (servers, peers etc.) is a defining factor of efficiency in these systems -- applications are often communication intensive, whereas agents share links of only limited bandwidth. This necessitates the use of principled algorithms, as efficient communication to a large extent depends on the interaction protocols.
We study data center networks and peer-to-peer networks as canonical examples of modern-day large-scale networking systems. Server-to-server interaction is an integral part of the data center's operation. The latency of these interactions is often a significant bottleneck toward overall job completion times. We study complementary approaches toward reducing this latency: (i) design of computation algorithms that minimize interaction and (ii) optimal scheduling algorithms to maximally utilize the network fabric. We also consider peer-to-peer networks as an emerging mode of content distribution and sharing. Unlike data centers, these networks are flexible in their network structure and also scale well, but require decentralized algorithms for control. Of central importance here is the design of a network topology that enables efficient peer interactions for optimal application performance. We propose novel topology designs for two popular applications: (i) multimedia streaming and (ii) anonymity in Bitcoin's peer-to-peer network.
Network algorithms; Interactive communication; Communication complexity; Protocol compression; Scheduling; Circuit switch; Data center networks; Submodularity; Peer-to-peer; Streaming; Topology; Bitcoin; Anonymity; Distributed algorithms; Cryptocurrency
Wed, 12 Jul 2017 00:00:00 GMThttp://hdl.handle.net/2142/983672017-07-12T00:00:00ZBojja Venkatakrishnan, ShaileshhGeometry of compositionality
http://hdl.handle.net/2142/98362
Geometry of compositionality
Gong, Hongyu
Word embedding is a popular representation of words in vector space, and its geometry reveals the lexical semantics. This thesis further explores the interesting geometric properties of word embedding, and looks into its interaction with the context representation. We propose an innovative method to detect whether a given word or phrase is used literally in a specific context. This work focuses on three specific applications in natural language processing: idiomaticity, sarcasm and metaphor detection. Extensive experiments have shown that this embedding-based method achieves good performance in multiple languages.
Compositionality detection
Tue, 11 Jul 2017 00:00:00 GMThttp://hdl.handle.net/2142/983622017-07-11T00:00:00ZGong, HongyuSwitched and hybrid systems with inputs: small-gain theorems, control with limited information, and topological entropy
http://hdl.handle.net/2142/98353
Switched and hybrid systems with inputs: small-gain theorems, control with limited information, and topological entropy
Yang, Guosong
In this thesis, we study stability and stabilization of switched and hybrid systems with inputs. We consider primarily two topics in this area: small gain theorems for interconnected switched and hybrid systems, and control of switched linear systems with limited information.
First, we study input-to-state practical stability (ISpS) of interconnections of two switched nonlinear subsystems with independent switchings and possibly non-ISpS modes. Provided that for each subsystem, the switching is slow in the sense of an average dwell-time (ADT), and the total active time of non-ISpS modes is short in proportion, Lyapunov-based small-gain theorems are established via hybrid system techniques. By augmenting each subsystem with a hybrid auxiliary timer that models the constraints on switching, we enable a construction of hybrid ISpS-Lyapunov functions, and consequently, a convenient formulation of a small-gain condition for ISpS of the interconnection. Based on our small-gain theorem, we demonstrate the stabilization of interconnected switched control-affine systems using gain-assignment techniques.
Second, we investigate input-to-state stability (ISS) of networks composed of n ≥ 2 hybrid subsystems with possibly non-ISS dynamics. Lyapunov-based small-gain theorems are established based on the notion of candidate ISS-Lyapunov functions, which unifies and extends several previous results for interconnected hybrid and impulsive systems. In order to apply our small-gain theorem to different combinations of non-ISS dynamics, we adopt the method of modifying candidate exponential ISS-Lyapunov functions using ADT and reverse ADT timers. The effect of such modifications on the Lyapunov feedback gains between two interconnected hybrid systems is discussed in detail through a case-by-case study.
Third, we consider the problem of stabilizing a switched linear system with a completely unknown disturbance using sampled and quantized state feedback. The switching is assumed to be slow enough in the sense of combined dwell-time and average dwell-time, each individual mode is assumed to be stabilizable, and the data rate is assumed to be large enough but finite. By extending the approach of reachable-set approximation and propagation from an earlier result on the disturbance-free case, we develop a communication and control strategy that achieves a variant of input-to-state stability with exponential decay. An estimate of the disturbance bound is introduced to compensate for the unknown disturbance, and a novel algorithm is designed to adjust the estimate and recover the state when it escapes the range of quantization.
Last, motivated by the connection between the minimum data rate needed to stabilize a linear time-invariant system and its topological entropy, we examine a notion of topological entropy for switched systems with a known switching signal. This notion is formulated in terms of the number of initial points such that the corresponding trajectories approximate all trajectories within a certain error, and can be equivalently defined using the number of initial points that are separable up to a certain precision. We first calculate the topological entropy of a switched scalar system based on the active rates of its modes. This approach is then generalized to nonscalar switched linear systems with certain Lie structures to establish entropy bounds in terms of the active rate and eigenvalues of each mode.
Switched systems; Hybrid systems; Input-to-state stability; Lyapunov methods; Small-gain theorems; Sampling; Quantization; Data-rate constraints; Topological entropy
Fri, 14 Jul 2017 00:00:00 GMThttp://hdl.handle.net/2142/983532017-07-14T00:00:00ZYang, GuosongStudy of waves observed in the equatorial ionospheric valley region using Jicamarca ISR and VIPIR ionosonde
http://hdl.handle.net/2142/98349
Study of waves observed in the equatorial ionospheric valley region using Jicamarca ISR and VIPIR ionosonde
Reyes, Pablo Martin
Incoherent scatter (IS) radar and ionosonde (VIPIR, vertical incidence pulsed ionospheric radar) data were taken concurrently at Jicamarca during campaigns of January, April, June, and July 2015, January 2016, and most recently April 2017 to bring more insight into the state and dynamics of the ionospheric E-F valley region and the 150-km radar echoes detected from this region. To better understand the rich and dynamic vertical structure of 150-km echoes observed at the Jicamarca Radio Observatory (JRO) and other equatorial stations and to contribute to the understanding of the physics of this region, we used JRO ISR and VIPIR ionosonde techniques to perform high spatial and temporal resolution measurements.
We found correlations between VHF backscatter radar measurements and fluctuations detected with the VIPIR ionosonde, which is an indication of gravity waves playing a role in modulating the space-time structure of the 150-km echoes. Fluctuations with periods from 5 to 15 minutes are observed in VIPIR ionograms as well as in the layers found in the 50 MHz radar range-time-intensity (RTI) plots. The quiet-time stratified electron density contours are being rippled by waves propagating through the ionosphere. Evidence for this is the fluctuation of virtual reflection heights and angle of arrival (AOA) of the ionosonde echoes. The AOA is provided by interferometry, which indicates that the echo is not always coming from overhead. Scatter plots of the AOA in the receiving antenna’s orthogonal baselines give us the propagation direction. Plots of virtual height and AOA obtained using VIPIR data show phase fronts propagating downwards, which is characteristic of internal gravity waves (IGW). Other characteristics of IGW are present in the oscillations of virtual height: their frequencies are just below the Brunt-Väisälä frequency, their amplitudes increase with altitude, and shorter vertical wavelengths seen in lower altitudes are heavily damped in higher altitudes. The observed IGW exhibit fluctuations similar to those seen in the thin “forbidden” or “quite” zone of the 150-km echo undulations, which indicates some IGW-driven modulations of the 150-km echo as has been suggested previously [e.g. Kudeki and Fawcett, 1993; Chau and Kudeki, 2013]. Phase profiles of cross- correlation pair of antennas in the IS Faraday rotation experiment exhibit a smooth progression with altitude. That means that there are no sharp density gradients that could be a source of plasma instabilities. Still, density variations across the magnetized plasma in the region can be key to explaining the enhanced echoes observed via the electrodynamics that they can drive.
We also found that there exist sub-minute quasi-periodic (SMQP) fluctuations when zooming into high time resolution RTI plots. This is a new observation that has not been reported in the literature to date. A method was designed in order to validate the existence of SMQP fluctuations. The method consisted of identifying episodes of sub-minute fluctuations in a non-exhaustive search of high resolution RTI plots using a web-based interactive tool designed for zooming in and marking the episodes where the sub-minute period fluctuations were found. We found a wide range of sub-minute periods, with a predominance between 15 and 20 seconds. This was a first step towards reporting SMQP; a more exhaustive method to search for these fluctuations is being produced.
This multi-instrument approach helps us to characterize the daytime electron density fluctuations in the equatorial valley region, and aims to contribute to the goal of understanding better the fundamental physics of the region.
Equatorial ionosphere; Ionospheric irregularities; 150-km echoes; Internal gravity waves; Jicamarca Radio Observatory incoherent scatter radar (JRO ISR); Vertical incidence pulsed ionospheric radar (VIPIR) ionosonde
Thu, 13 Jul 2017 00:00:00 GMThttp://hdl.handle.net/2142/983492017-07-13T00:00:00ZReyes, Pablo Martin