<p>Online decision making with switching costs has received considerable attention in many practical problems that face uncertainty in the inputs and key problem parameters. Because of the switching costs that penalize the change of decisions, making good online decisions under such uncertainty is known to be extremely challenging. This thesis aims at providing new online algorithms with strong performance guarantees to address this challenge.</p> <p><br></p> <p>In part 1 and part 2 of this thesis, motivated by Network Functions Virtualization and smart grid, we study competitive online convex optimization with switching costs. Specifically, in part 1, we focus on the setting with an uncertainty set (one type of prediction) and hard infeasibility constraints. We develop new online algorithms that can attain optimized competitive ratios, while ensuring feasibility at all times. Moreover, we design a robustification procedure that helps these algorithms obtain good average-case performance simultaneously. In part 2, we focus on the setting with look-ahead (another type of prediction). We provide the first algorithm that attains a competitive ratio that not only decreases to 1 as the look-ahead window size increases, but also remains upper-bounded for any ratio between the switching-cost coefficient and service-cost coefficient.</p> <p><br></p> <p>In part 3 of this thesis, motivated by edge computing with artificial intelligence, we study bandit learning with switching costs where, in addition to bandit feedback, full feedback can be requested at a cost. We show that, when only 1 arm can be chosen at a time, adding costly full-feedback is not helpful in fundamentally reducing the Θ(<em>T</em>2/3) regret over a time-horizon <em>T</em>. In contrast, when 2 (or more) arms can be chosen at a time, we provide a new online learning algorithm that achieves a significantly smaller regret equal to <em>O</em>(√<em>T</em>), without even using full feedback. To the best of our knowledge, this type of sharp transition from choosing 1 arm to choosing 2 (or more) arms has never been reported in the literature.</p>

A thesis which models the DNP3 and IEC 61850 protocol in OPNET / Supervisory Control and Data Acquisition (SCADA) is a critical part of monitoring and controlling of the electrical substation. The aim of this dissertation is to investigate the performance of the Distributed Network Protocol Version 3.3 (DNP3) protocol and to compare its performance to that of International Electro-technical Commission (IEC) 61850 protocol in an electrical substation communication network environment. Building an electrical substation control room and installing the network equipment was going to be expensive and take a lot of time. The better option was to build a model of the electrical substation communication network and run simulations. Riverbend modeller academic edition known as Optimized Network Engineering Tool (OPNET) was chosen as a software package to model substation communication network, DNP3 protocol and IEC 61850 Protocol stack. Modelling the IEC 61850 protocol stack on OPNET involved building the used Open System Interconnection (OSI) layers of the IEC 61850 protocol stack onto the application definitions of OPNET. The Transmission Control Protocol/Internet Protocol (TCP/IP) configuration settings of DNP3 protocol were also modelled on the OPNET application definitions. The aim is to compare the two protocols and determine which protocol is the best performing one in terms of throughput, data delay and latency. The substation communication model consists of 10 ethernet nodes which simulate protection Intelligent Electronic Devices (IEDs), 13 ethernet switches, a server which simulates the substation Remote Terminal Unit (RTU) and the DNP3 Protocol over TCP/IP simulated on the model. DNP3 is a protocol that can be used in a power utility computer network to provide communication service for the grid components. DNP3 protocol is currently used at Eskom as the communication protocol because it is widely used by equipment vendors in the energy sector. DNP3 protocol will be modelled before being compared to the new recent robust protocol IEC 61850 in the same model and determine which protocol is the best for Eskom on the network of the power grid. The network load and packet delay parameters were sampled when 10%, 50%, 90% and 100% of devices are online. The IEC 61850 protocol model has three scenarios and they are normal operation of a Substation, maintenance in a Substation and Buszone operation at a Substation. In these scenarios packet end to end delay of Generic Object Oriented Substation Event (GOOSE), vi © University of South Africa 2020 Generic Substation Status Event (GSSE), Sampled Values (SV) and Manufacturing Messaging Specification (MMS) messages are monitored. The throughput from the IED under maintenance and the throughput at the Substation RTU end is monitored in the model. Analysis of the results of the DNP3 protocol simulation showed that with an increase in number of nodes there was an increase in packet delay as well as the network load. The load on the network should be taken into consideration when designing a substation communication network that requires a quick response such as a smart gird. GOOSE, GSSE, SV results on the IEC 61850 model met all the requirements of the IEC 61850 standard and the MMS did not meet all the requirements of the IEC standard. The design of the substation communication network using IEC 61850 will assist when trying to predict the behavior of the network with regards to this specific protocol during maintenance and when there are faults in the communication network or IED’s. After the simulation of the DNP3 protocol and the IEC 61850 the throughput of DNP3 protocol was determined to be in the range (20 – 450) kbps and the throughput of IEC61850 protocol was determined to be in the range (1.6 – 16) Mbps. / College of Engineering, Science and Technology / M. Tech. (Electrical Engineering)

<p>The swift emergence and wide-ranging utilization of machine learning (ML) across various industries, including healthcare, transportation, and robotics, have underscored the escalating need for efficient, scalable, and privacy-preserving solutions. Recognizing this, we present an integrated examination of three novel frameworks, each addressing different aspects of distributed learning and privacy issues: Two Timescale Hybrid Federated Learning (TT-HF), Delay-Aware Federated Learning (DFL), and Differential Privacy Hierarchical Federated Learning (DP-HFL). TT-HF introduces a semi-decentralized architecture that combines device-to-server and device-to-device (D2D) communications. Devices execute multiple stochastic gradient descent iterations on their datasets and sporadically synchronize model parameters via D2D communications. A unique adaptive control algorithm optimizes step size, D2D communication rounds, and global aggregation period to minimize network resource utilization and achieve a sublinear convergence rate. TT-HF outperforms conventional FL approaches in terms of model accuracy, energy consumption, and resilience against outages. DFL focuses on enhancing distributed ML training efficiency by accounting for communication delays between edge and cloud. It also uses multiple stochastic gradient descent iterations and periodically consolidates model parameters via edge servers. The adaptive control algorithm for DFL mitigates energy consumption and edge-to-cloud latency, resulting in faster global model convergence, reduced resource consumption, and robustness against delays. Lastly, DP-HFL is introduced to combat privacy vulnerabilities in FL. Merging the benefits of FL and Hierarchical Differential Privacy (HDP), DP-HFL significantly reduces the need for differential privacy noise while maintaining model performance, exhibiting an optimal privacy-performance trade-off. Theoretical analysis under both convex and nonconvex loss functions confirms DP-HFL’s effectiveness regarding convergence speed, privacy performance trade-off, and potential performance enhancement with appropriate network configuration. In sum, the study thoroughly explores TT-HF, DFL, and DP-HFL, and their unique solutions to distributed learning challenges such as efficiency, latency, and privacy concerns. These advanced FL frameworks have considerable potential to further enable effective, efficient, and secure distributed learning.</p>

<p>    </p> <p>This thesis studies networked Gaussian linear control systems with random delays. Networked control systems is a popular topic these years because of their versatile applications in daily life, such as smart grid and unmanned vehicles. With the development of these systems, researchers have explored this area in two directions. The first one is to derive the inherent rate-cost relationship in the systems, that is the minimal transmission rate needed to achieve an arbitrarily given stability requirement. The other one is to design achievability schemes, which aim at using as less as transmission rate to achieve an arbitrarily given stability requirement. In this thesis, we explore both directions. We assume the sensor-to-controller channels experience independently and identically distributed random delays of bounded support. Our work separates into two parts. In the first part, we consider networked systems with only one sensor. We focus on deriving a lower bound, R_{LB}(D), of the rate-cost tradeoff with the cost function to be E{| <strong>x^</strong>T<strong>x </strong>|} ≤ D, where <strong>x </strong>refers to the state to be controlled. We also propose an achievability scheme as an upper bound, R_{UB}(D), of the optimal rate-cost tradeoff. The scheme uses lattice quantization, entropy encoder, and certainty-equivalence controller. It achieves a good performance that roughly requires 2 bits per time slot more than R_{LB}(D) to achieve the same stability level. We also generalize the cost function to be of both the state and the control actions. For the joint state-and-control cost, we propose the minimal cost a system can achieve. The second part focuses on to the covariance-based fusion scheme design for systems with multiple > 1 sensors. We notice that in the multi-sensor scenario, the outdated arrivals at the controller, which many existing fusion schemes often discard, carry additional information. Therefore, we design an implementable fusion scheme (CQE) which is the MMSE estimator using both the freshest and outdated information at the controller. Our experiment demonstrates that CQE out-performances the MMSE estimator using the freshest information (LQE) exclusively by achieving a 15% smaller average L2 norm using the same transmission rate. As a benchmark, we also derive the minimal achievable L2 norm, Dmin, for the multi-sensor systems. The simulation shows that CQE approaches Dmin significantly better than LQE. </p>