Dissertations and Theses - Electrical and Computer Engineering
http://hdl.handle.net/2142/8888
Dissertations and Theses in Electrical and Computer EngineeringThu, 26 Nov 2015 17:51:04 GMT2015-11-26T17:51:04ZAdaptive sparse representations and their applications
http://hdl.handle.net/2142/88322
Adaptive sparse representations and their applications
Ravishankar, Saiprasad
The sparsity of signals and images in a certain transform domain or dictionary has been exploited in many applications in signal processing, image processing, and medical imaging. Analytical sparsifying transforms such as Wavelets and DCT have been widely used in compression standards. Recently, the data-driven learning of synthesis sparsifying dictionaries has become popular especially in applications such as denoising, inpainting, and compressed sensing. While there has been extensive research on learning synthesis dictionaries and some recent work on learning analysis dictionaries, the idea of learning sparsifying transforms has received no attention. In the first part of this thesis, we study the sparsifying transform model and its relationship to prior linear sparse models. Then, we propose novel problem formulations for learning square sparsifying transforms from data. The proposed algorithms for transform learning alternate between a sparse coding step and a transform update step, and are highly efficient.
Specifically, as opposed to sparse coding in the synthesis or noisy analysis models which is NP-hard, the sparse coding step in transform learning can be performed exactly and cheaply by zeroing out all but a certain number of nonzero transform coefficients of largest magnitude.
The transform update step is performed using iterative conjugate gradients.
The proposed algorithms give rise to well-conditioned square sparsifying transforms in practice. We show the superiority of our approach over analytical sparsifying transforms such as the DCT for signal and image representation. We also show promising performance in signal denoising using the learned sparsifying transforms. The proposed approach is much faster than previous approaches involving learned synthesis, or analysis dictionaries.
Next, we explore a specific structure for learned sparsifying transforms, that enables efficient implementations.
Following up on the idea of learning square sparsifying transforms, we propose novel problem formulations for learning doubly sparse transforms for signals or image patches. These transforms are a product of a fixed, fast analytic transform such as the DCT, and an adaptive matrix constrained to be sparse. Such transforms can be learned, stored, and implemented efficiently. We show the superior promise of our learned doubly sparse transforms as compared to analytical sparsifying transforms such as the DCT or Wavelets for image representation.
Adapted doubly sparse transforms also generalize better than the ‘unstructured’ (or non-sparse) transform.
We show promising performance and speedups in image denoising using the learned doubly sparse transforms compared to approaches involving learned synthesis dictionaries such as the K-SVD algorithm.
In the third part of this thesis, we further develop the alternating algorithms for learning unstructured (non-sparse) well-conditioned, or orthonormal square sparsifying transforms.
While, in the first part of the thesis, we provided an iterative method involving conjugate gradients for the transform update step, in this part, we instead derive efficient and analytical closed-form solutions for transform update. Importantly, we establish that the proposed algorithms are globally convergent to the set of local minimizers of the non-convex transform learning problems. In practice, our algorithms are shown to be insensitive to initialization.
In the next part of the thesis, we focus on compressed sensing (CS), which exploits the sparsity of images or image patches in a transform domain or synthesis dictionary to reconstruct images from highly undersampled or compressive measurements. Specifically, we focus on the subject of blind compressed sensing, where the underlying sparsifying transform is unknown a priori, and propose a framework to simultaneously reconstruct the underlying image(s)/volume(s) as well as the square sparsifying transform from highly undersampled measurements. The proposed block coordinate descent type algorithms involve highly efficient closed-form optimal updates. Importantly, we prove that although the proposed blind compressed sensing formulations are highly nonconvex, our algorithms converge to the set of critical points of the objectives defining the formulations.
We illustrate the usefulness of the proposed framework for magnetic resonance image (MRI) reconstruction from highly undersampled k-space measurements. As compared to previous state-of-the-art methods involving the synthesis model, our approach is 10x faster for reconstructing 2D MR images, while also providing promising reconstruction quality. The proposed transform-based blind compressed sensing has the potential to revolutionize medical imaging technologies by highly accelerating both the imaging and image reconstruction processes.
In the fifth part of this thesis, we study the design of sampling schemes for compressed sensing MRI.
The (pseudo) random sampling schemes used most often for CS may have good theoretical asymptotic properties; however, with limited data they may be far from optimal. Therefore, we propose a novel framework for improved adaptive sampling schemes for highly undersampled CS MRI.
While the proposed framework is general, we apply it with some recent MRI reconstruction algorithms.
Numerical experiments demonstrate that our adaptive sampling scheme can provide significant improvements in image reconstruction quality for MRI compared to non-adapted methods.
In the next part of the thesis, we develop a methodology for online learning of square sparsifying transforms. Such online learning is particularly useful when dealing with big data, and for signal processing applications such as real-time sparse representation and denoising. The proposed transform learning algorithms are shown to have a much lower computational cost than online synthesis dictionary learning. In practice, the sequential learning of a sparsifying transform typically converges much faster than batch mode transform learning. Preliminary experiments show the usefulness of the proposed schemes for sparse representation (compression), and denoising. We also prove that although the associated optimization problems are non-convex, our online transform learning algorithms are guaranteed to converge to the set of stationary points of the learning problem. The guarantee relies on few (easy to verify) assumptions.
In the seventh part of this thesis, we propose a novel convex formulation for doubly sparse square transform learning. The proposed formulation has similarities to traditional least squares optimization with $\ell_1$ regularization. Our convex learning algorithm is a modification of FISTA, and is guaranteed to converge to a global optimum, and moreover converges quickly. We also study two non-convex variants of the proposed convex formulation, and provide local convergence proof for the algorithm for one of them. These proposed non-convex variants use the $\ell_0$ ``norm" for measuring the sparsity of the transform and/or sparse code.
We show the superior promise of our learned transforms here as compared to analytical sparsifying transforms such as the DCT for image representation. In these examples, the performance is sometimes comparable to the previously proposed non-convex (non guaranteed) doubly sparse transform learning schemes.
While we studied the learning of square transforms in the initial parts of the thesis, in the eighth part of the thesis, we instead briefly study the learning of tall or overcomplete sparsifying transforms from data. We propose various penalties that control the sparsifying ability, condition number, and incoherence of the learned transforms. Our alternating algorithm for overcomplete transform learning converges empirically, and significantly improves the quality of the learned transform over the iterations. We present examples demonstrating the promising performance of adaptive overcomplete transforms over adaptive overcomplete synthesis dictionaries learned using the popular K-SVD algorithm, in the application of image denoising. The overcomplete transforms also denoise better than adaptive square transforms.
In the final part of the thesis, we explore the idea of learning efficient structured overcomplete sparsifying transforms. Since natural images typically contain diverse textures that cannot be sparsified well by a single transform, we therefore propose a union of sparsifying transforms model. Sparse coding in this model reduces to a form of transform-domain clustering. This makes the model appealing for classification tasks. The proposed model is also equivalent to a structured overcomplete sparsifying transform model with block cosparsity, dubbed OCTOBOS. The alternating algorithm introduced for learning such transforms involves simple closed-form solutions. A theoretical analysis provides a convergence guarantee for this algorithm. It is shown to be globally convergent to the set of partial minimizers of the non-convex OCTOBOS (or, union of transforms) learning problem. We also show that under certain conditions, the algorithm converges to the set of stationary points of the overall objective. When applied to images, the algorithm learns a collection of well-conditioned square transforms, and a good clustering of patches or textures. The resulting sparse representations for the images are much better than those obtained with a single learned transform, or with analytical transforms. We show the promising performance of the proposed approach in image denoising, which compares quite favorably with approaches involving a single learned square transform or an overcomplete synthesis dictionary, or Gaussian mixture models. The proposed denoising method is also faster than the synthesis dictionary based approach.
Inverse problems; Computer vision; Classification; Structured overcomplete transform learning; Union of transforms; Overcomplete transform learning; Structured transforms; Convex formulation; Real-time applications; Big data; Online learning; Adaptive sampling; Image reconstruction; Block Coordinate descent; Blind compressed sensing; Convergence guarantees; Efficient updates; Closed-form solutions; Machine learning; Nonconvex optimization; Alternating minimization; Doubly sparse transform learning; Square transform learning; Adaptive sparse models; Denoising; dictionary learning; Magnetic resonance imaging; Compressed sensing; Sparse representations; Sparsifying transform learning
Fri, 05 Dec 2014 00:00:00 GMThttp://hdl.handle.net/2142/883222014-12-05T00:00:00ZRavishankar, SaiprasadA characteristic mode perturbation approach for antenna loading design
http://hdl.handle.net/2142/88289
A characteristic mode perturbation approach for antenna loading design
Young, Matthew William
Reactive loading is commonly applied to antennas to modify input impedance and radiation pattern properties. However, reactive loading design approaches based on experience, intuition, and modeling are challenged as the demand grows for antennas with increased functionality and performance. New systematic design methods are needed that can manage complicated performance tradeoffs while providing physical insight into the fundamental antenna operation. Characteristic mode theory has shown promise for systematic antenna design, yet significant limitations still exist that restrict its usefulness. The transformations of characteristic modes with respect to frequency or reactive loading are currently understood only qualitatively, and insight into their behavior must be developed through experience. In this thesis, a novel characteristic mode perturbation approach is developed that provides a quantitative description of how mode eigenvalues and eigencurrents transform under reactive loading and frequency variation. Analytical equations are derived using a novel application of eigenvalue perturbation theory to the characteristic mode problem. The equations characterize the effect of impedance loading on the characteristic mode eigenvalues and eigencurrents and reveal the explicit factors governing the mode transformations.
Insight from the perturbation equations suggests a new characteristic mode design paradigm in which loading is used to control the eigencurrent contributions between modes. The new eigencurrent contribution perspective can be used to understand and solve loading problems that traditional characteristic mode theory cannot. The approach is used to design the reactive loading of an Archimedean spiral antenna to produce beam tilt or an endfire radiation pattern while maintaining desirable VSWR properties. Finally, the perturbation approach is used to analyze antenna loss and is applied to the analysis and improvement of antenna radiation efficiency.
Antenna design; Characteristic mode theory; Perturbation theory; Spiral antenna; Reactive loading; Radiation efficiency; Compensation theorem
Thu, 16 Jul 2015 00:00:00 GMThttp://hdl.handle.net/2142/882892015-07-16T00:00:00ZYoung, Matthew WilliamHelium resonance fluorescence LiDAR
http://hdl.handle.net/2142/88241
Helium resonance fluorescence LiDAR
Mangognia, Anthony Dominic
Recent advancements in LiDAR for atmospheric applications have allowed measurements to reach higher altitudes and have increased temporal resolu- tion. Increased output power, larger apertures, and higher efficiency detec- tors have made this possible. The helium resonance fluorescence LiDAR has the capability to probe the metastable helium content in the thermosphere and exosphere (within 250 km-750 km), where helium (23S) is most abundant. Strategies have been employed to increase the output power of the He Li- DAR transmitter using fiber amplifier technology, increase the light gathering power of the receiver, and utilize detectors with higher quantum efficiencies at 1083 nm. A 45 W He resonance fluorescence LiDAR transmitter has been designed and fabricated, and is being tested in Urbana, IL, with plans for deployment at an astronomical observatory in the near future. The He reso- nance fluorescence LiDAR has the potential to further our understanding of upper atmosphere dynamics. It will provide insight into metastable helium, its temperature in the upper atmosphere, and atmospheric densities, which affect satellite drag, and possibly pave the way for new applications, such as guide star lasers. The technology may be applied from ground based, as well as satellite based, platforms for global measurement applications. This dissertation discusses the planned approach to detect the first LiDAR gener- ated resonantly fluoresced scattered He photon, details of the He resonance fluorescence LiDAR transmitter, and the simulations for the expected signal return.
Light Radar (LiDAR); Resonance Fluorescence; Helium
Tue, 12 May 2015 00:00:00 GMThttp://hdl.handle.net/2142/882412015-05-12T00:00:00ZMangognia, Anthony DominicAsymmetric interleaving in low-voltage CMOS power management with multiple supply rails
http://hdl.handle.net/2142/88219
Asymmetric interleaving in low-voltage CMOS power management with multiple supply rails
Ho, Aaron Daniel
Recent years have seen the proliferation of electronic devices that require multi-phase power converters to provide heterogeneous power rails to different systems. Typical systems will utilize symmetric interleaving as a method of reducing the input current ripple for the power converter. Asymmetric interleaving is a method of control that allows for a further reduction, and in some cases complete cancellation, of this input current ripple. This work looks at some of the challenges for a practical implementation using digital control, and provides results to quantify this improvement. This work demonstrates a control algorithm implementation capable of achieving nearly 3x reduction in the input current ripple via the asymmetric interleaving method.
complementary metal–oxide–semiconductor (CMOS) integrated circuits; digital control; low-power electronics; power convertors; asymmetric interleaving; digital control; heterogeneous power rails; low voltage complementary metal–oxide–semiconductor (CMOS) power management; multiphase (complementary metal–oxide–semiconductor) CMOS power management IC system; multiphase power converters; multiple supply rails; reduced input current ripple; size 180 nm; Hardware; Mathematical model; Prototypes; Table lookup; Time-domain analysis
Tue, 21 Jul 2015 00:00:00 GMThttp://hdl.handle.net/2142/882192015-07-21T00:00:00ZHo, Aaron DanielGraphene nano-ribbon and transition metal dichalcogenide field-effect transistor modeling and circuit simulation
http://hdl.handle.net/2142/88212
Graphene nano-ribbon and transition metal dichalcogenide field-effect transistor modeling and circuit simulation
Chen, Ying-Yu
This dissertation presents a modeling and simulation study of graphene nano-ribbon and transition metal dichalcogenide field-effect transistors. Through compact modeling, SPICE implementation of the transistors is realized, and circuit-level simulation is enabled. Extensive simulation studies are performed to evaluate the performance of these two emerging devices.
graphene; transition metal dichalcogenide; transistor; flexible transistor; modeling; simulation
Fri, 17 Jul 2015 00:00:00 GMThttp://hdl.handle.net/2142/882122015-07-17T00:00:00ZChen, Ying-YuDevelopment of novel series and parallel sensing system based on nanostructured surface enhanced Raman scattering substrate for biomedical application
http://hdl.handle.net/2142/88211
Development of novel series and parallel sensing system based on nanostructured surface enhanced Raman scattering substrate for biomedical application
Chang, Te-Wei
With the advance of nanofabrication, the capability of nanoscale metallic structure fabrication opens a whole new study in nanoplasmonics, which is defined as the investigation of photon-electron interaction in the vicinity of nanoscale metallic structures. The strong oscillation of free electrons at the interface between metal and surrounding dielectric material caused by propagating surface plasmon resonance (SPR) or localized surface plasmon resonance (LSPR) enables a variety of new applications in different areas, especially biological sensing techniques.
One of the promising biological sensing applications by surface resonance polariton is surface enhanced Raman spectroscopy (SERS), which significantly reinforces the feeble signal of traditional Raman scattering by at least 104 times. It enables highly sensitive and precise molecule identification with the assistance of a SERS substrate. Until now, the design of new SERS substrate fabrication process is still thriving since no dominant design has emerged yet. The ideal process should be able to achieve both a high sensitivity and low cost device in a simple and reliable way. In this thesis two promising approaches for fabricating nanostructured SERS substrate are proposed: thermal dewetting technique and nanoimprint replica technique. These two techniques are demonstrated to show the capability of fabricating high performance SERS substrate in a reliable and cost efficient fashion. In addition, these two techniques have their own unique characteristics and can be integrated with other sensing techniques to build a serial or parallel sensing system. The breakthrough of a combination system with different sensing techniques overcomes the inherent limitations of SERS detection and leverages it to a whole new level of systematic sensing.
The development of a sensing platform based on thermal dewetting technique is covered as the first half of this thesis. The process optimization, selection of substrate material, and improved deposition technique are discussed in detail. Interesting phenomena have been found including the influence of Raman enhancement on substrate material selection and hot-spot rich bimetallic nanostructures by physical vapor deposition on metallic seed array, which are barely discussed in past literature but significantly affect the performance of SERS substrate. The optimized bimetallic backplane assisted resonating nanoantenna (BARNA) SERS substrate is demonstrated with the enhancement factor (EF) of 5.8 × 108 with 4.7 % relative standard deviation. By serial combination with optical focusing from nanojet effect, the nanojet and surface enhanced Raman scattering (NASERS) are proved to provide more than three orders of enhancement and enable us to perform stable, nearly single molecule detection.
The second part of this thesis includes the development of a parallel dual functional nano Lycurgus cup array (nanoLCA) plasmonic device fabricated by nanoimprint replica technique. The unique configuration of the periodic nanoscale cup-shaped substrate enables a novel hybrid resonance coupling between SPR from extraordinary (EOT) and LSPR from dense sidewall metal nanoparticles with only single deposition process. The sub-50nm dense sidewall metal nanoparticles lead to high SERS performance in solution based detection, by which most biological and chemical analyses are typically performed. The SERS EF was calculated as 2.8 × 107 in a solution based environment with 10.2 % RSD, which is so far the highest reported SERS enhancement achieved with similar periodic EOT devices. In addition, plasmonic colorimetric sensing can be achieved in the very same device and the sensitivity was calculated as 796 nm/RIU with the FOM of 12.7. It creates a unique complementary sensing platform with both rapid on-site colorimetric screening and follow-up precise Raman analysis for point of care and resource limited environment applications. The implementations of bifunctional sensing on opto-microfluidic and smartphone platforms are proposed and examined here as well.
Plasmonics; Nanofabrication; Surface Enhanced Raman Spectroscopy (SERS); Surface Plasmon Resonance (SPR); Localized Surface Plasmon Resonance (LSPR); Colorimetric Sensing; Bifunctional Deivce; extraordinary transmission (EOT); nano Lycurgus cup array (nanoLCA)
Fri, 17 Jul 2015 00:00:00 GMThttp://hdl.handle.net/2142/882112015-07-17T00:00:00ZChang, Te-WeiHigh performance and error resilient probabilistic inference system for machine learning
http://hdl.handle.net/2142/88204
High performance and error resilient probabilistic inference system for machine learning
Choi, Jungwook
Many real-world machine learning applications can be considered as inferring the best label assignment of maximum a posteriori probability (MAP) problems. Since these MAP problems are NP-hard in general, they are often dealt with using approximate inference algorithms on Markov random field (MRF) such as belief propagation (BP). However, this approximate inference is still computationally demanding, and thus custom hardware accelerators have been attractive for high performance and energy efficiency.
There are various custom hardware implementations that employ BP to achieve reasonable performance for the real-world applications such as stereo matching. Due to lack of convergence guarantees, however, BP often fails to provide the right answer, thus degrading performance of the hardware. Therefore, we consider sequential tree-reweighted message passing (TRW-S), which avoids many of these convergence problems with BP via sequential execution of its computations but challenges parallel implementation for high throughput. In this work, therefore, we propose a novel streaming hardware architecture that parallelizes the sequential computations of TRW-S. Experimental results on stereo matching benchmarks show promising performance of our hardware implementation compared to the software implementation as well as other BP-based custom hardware or GPU implementations.
From this result, we further demonstrate video-rate speed and high quality stereo matching using a hybrid CPU+FPGA platform. We propose three frame-level optimization techniques to fully exploit computational resources of a hybrid CPU+FPGA platform and achieve significant speed-up. We first propose a message reuse scheme which is guided by simple scene change detection. This scheme allows a current inference to be made based on a determination of whether the current result is expected to be similar to the inference result of the previous frame. We also consider frame level parallelization to process multiple frames in parallel using multiple FPGAs available in the platform. This parallelized hardware procedure is further pipelined with data management in CPU to overlap the execution time of the two and thereby reduce the entire processing time of the stereo video sequence. From experimental results with the real-world stereo video sequences, we see video-rate speed of our stereo matching system for QVGA stereo videos.
Next, we consider error resilience of the message passing hardware for energy efficient hardware implementation. Modern nanoscale CMOS process technologies suffer in reliability caused by process, temperature and voltage variations. Conventional approaches to deal with such unreliability (e.g., design for the worst-case scenario) are complex and inefficient in terms of hardware resources and energy consumption. As machine learning applications are inherently probabilistic and robust to errors, statistical error compensation (SEC) techniques can play a significant role in achieving robust and energy-efficient implementation. SEC embraces the statistical nature of errors and utilizes statistical and probabilistic techniques to build robust systems. Energy-efficiency is obtained by trading off the enhanced robustness with energy.
In this work, we analyze the error resilience of our message passing inference hardware subject to the hardware errors (e.g. errors caused by timing violation in circuits) and explore application of a popular SEC technique, algorithmic noise tolerance (ANT), to this hardware. Analysis and simulations show that the TRW-S message passing hardware is tolerant to small magnitude arithmetic errors, but large magnitude errors cause significantly inaccurate inference results which need to be corrected using SEC. Experimental results show that the proposed ANT-based hardware can tolerate an error rate of 21.3%, with performance degradation of only 3.5 % with an energy savings of 39.7 %, compared to an error-free hardware.
Lastly, we extend our TRW-S hardware toward a general purpose machine learning framework. We propose advanced streaming architecture with flexible choice of MRF setting to achieve 10-40x speedup across a variety of computer vision applications. Furthermore, we provide better theoretical understanding of error resiliency of TRW-S, and of the implication of ANT for TRW-S, under more general MRF setting, along with strong empirical support.
Belief propagation (BP); field programmable gate array (FPGA); hybrid-core computing platform; Markov random field (MRF); sequential tree-reweighted message passing (TRW-S); stereo matching; Error resilience; message passing; algorithmic noise tolerance
Fri, 17 Jul 2015 00:00:00 GMThttp://hdl.handle.net/2142/882042015-07-17T00:00:00ZChoi, JungwookDevelopment of antenna-coupled metal-insulator-metal diodes for infrared detection
http://hdl.handle.net/2142/88202
Development of antenna-coupled metal-insulator-metal diodes for infrared detection
Winoto, Ardy
The long-wavelength infrared (LWIR) band between 7 and 14 µm is significant for thermal imaging purposes because it coincides with the blackbody radiation peak of the human body and a low absorption window in the earth’s atmosphere. Also, at 10 µm wavelength, the earth’s terrestrial radiation is maximized and provides a good opportunity for untapped energy harvesting. These two factors combine to make a very strong need for a low-cost detector that operates at room temperature in this band. Antenna-coupled metal-insulator-metal (ACMIM) diodes are a leading candidate due to their high speed and potential manufacturability on Si substrates. In this work, Ni-NiO-Ni ACMIM diodes are fabricated using a two-step lithography process and an oxygen plasma oxidation. The device is characterized to both have current in the nA range and responsivities nearing 1000pA/W.
infrared (IR) detector; metal-insulator-metal (MIM) diode
Fri, 17 Jul 2015 00:00:00 GMThttp://hdl.handle.net/2142/882022015-07-17T00:00:00ZWinoto, ArdyPower of d choices for large-scale bin packing: a loss model
http://hdl.handle.net/2142/88185
Power of d choices for large-scale bin packing: a loss model
Dong, Xiaobo
A system with N parallel servers is considered in our thesis. Each server
consists of B units of a resource and jobs arrive at this system according
to a Poisson process. Each job stays in the system for an exponentially
distributed amount of time. Moreover, each job may request different units
of the resource from the system. Our goal is to understand how to route
arriving jobs to the servers to minimize the probability that an arriving job
does not find the required amount of resource at the server, i.e., the goal is
to minimize blocking probability. Our motivation arises from the design of
cloud computing systems in which the jobs are virtual machines (VMs) that
request resources such as memory from a large pool of servers. In our thesis,
we consider power-of-d-choices routing, where a job is routed to the server
with the largest amount of available resources among d 2 randomly chosen
servers. We consider a fluid model that corresponds to the limit as N goes to
infinity, and use numerical methods to approximate the blocking probability.
Moreover, we also show the simulation for the system.
Resource Allocation; Markov Process; Cloud Computing; Queueing
Tue, 14 Jul 2015 00:00:00 GMThttp://hdl.handle.net/2142/881852015-07-14T00:00:00ZDong, XiaoboTechniques for determining hidden properties of large-scale power systems
http://hdl.handle.net/2142/88179
Techniques for determining hidden properties of large-scale power systems
Mohapatra, Saurav
The contributions in this dissertation are towards augmenting and enhancing the knowledge in power system equivalent modeling, and dynamic mode estimation. Work related to these respective topics is presented herein in two parts -- (i) Network Based Methods, and (ii) Measurement Based Methods.
The first part focuses on the problem of creating limit preserving equivalents (LPEs). There is a push to develop LPEs for power system interconnections to be used in markets and reliability studies. The equivalents that exist for these interconnections do not capture thermal limits of equivalent lines, which results in their transmission limits being significantly different from the original interconnection limits. Assigning non-infinite and non-zero limits to equivalent lines is the niche of this work. This is done by considering an unloaded network, which is operating point independent. A solution method is developed and discussed, which is capable of assigning lower, best and upper estimates for equivalent line limits, and is proposed for use towards developing LPEs.
In the second part, a relatively new method for simultaneous modal analysis of multiple time-series signals is presented. Here, Dynamic Mode Decomposition (DMD) is successfully applied towards transmission-level power system measurements in an implementation that is able to run in real-time. Since power systems are considered as non-linear and time-varying, on-line modal identification is capable of monitoring the evolution of large-scale power system dynamics by providing a breakdown of the constituent oscillation frequencies and damping ratios, and their respective amplitudes. The outputs provided by DMD can enable on-line spatio-temporal analyses, improve situational awareness, and could even contribute towards control strategies. This work presents the theory of DMD, followed by results and visualization. It shows that using frequency and voltage data together helps with precision, while maintaining fast calculation speeds. The key advantage of this implementation is its relatively fast computation; for example, it is able to process each time-window, consisting of 3392 signals with 211 time points, in 0.185 s. Modal content alarm processing, and efficient wide-area modal visualization are two proposed on-line applications.
The desire to reduce model-dependency has driven measurement-based modal identification methods, as an alternative to analyzing linearized system models. Using this relatively fast DMD algorithm, this work also presents an interactive modal-identification tool for spatio-temporal analysis of measurement data. The tool can automatically scan through measurements, and display the values of oscillation frequency and damping ratio, as well as reconstruct signals. The use of this tool, its options, and visualization capabilities are illustrated using simulated measurements from an interconnected power grid.
DMD being a data-driven modeling technique is able to handle large data sets and has shown fast computation times. The by-products of DMD provide an understanding of the wide-area spatio-temporal structures in power systems. Studies based on a large-scale model of an interconnected power grid are presented, along with visualizations that elucidate the spatial structure of wide-area dynamics, and their dependency on operating points.
Attribute preserving equivalents; Equivalent line limits; Transmission limits; Power transfer distribution factors; Total transfer capability; Quadratic program; Power system dynamics; On-line modal identification; Alarm processing; Situational awareness; Visualization; Dynamic Mode Decomposition; Linearized Dynamic Analysis; Measurement-based modeling; Data-driven methods
Tue, 14 Jul 2015 00:00:00 GMThttp://hdl.handle.net/2142/881792015-07-14T00:00:00ZMohapatra, Saurav