Indiana University-Purdue University Indianapolis (IUPUI) / Depth estimation is increasingly becoming more important in computer vision. The requirement for autonomous systems to gauge their surroundings is of the utmost importance in order to avoid obstacles, preventing damage to itself and/or other systems or people. Depth measuring/estimation systems that use multiple cameras from multiple views can be expensive and extremely complex. And as these autonomous systems decrease in size and available power, the supporting sensors required to estimate depth must also shrink in size and power consumption. This research will concentrate on a single passive method known as Depth from Defocus (DfD), which uses an in-focus and out-of-focus image to infer the depth of objects in a scene. The major contribution of this research is the introduction of a new Deep Learning (DL) architecture to process the the in-focus and out-of-focus images to produce a depth map for the scene improving both speed and performance over a range of lighting conditions. Compared to the previous state-of-the-art multi-label graph cuts algorithms applied to the synthetically blurred dataset the DfD-Net produced a 34.30% improvement in the average Normalized Root Mean Square Error (NRMSE). Similarly the DfD-Net architecture produced a 76.69% improvement in the average Normalized Mean Absolute Error (NMAE). Only the Structural Similarity Index (SSIM) had a small average decrease of 2.68% when compared to the graph cuts algorithm. This slight reduction in the SSIM value is a result of the SSIM metric penalizing images that appear to be noisy. In some instances the DfD-Net output is mottled, which is interpreted as noise by the SSIM metric. This research introduces two methods of deep learning architecture optimization. The first method employs the use of a variant of the Particle Swarm Optimization (PSO) algorithm to improve the performance of the DfD-Net architecture. The PSO algorithm was able to find a combination of the number of convolutional filters, the size of the filters, the activation layers used, the use of a batch normalization layer between filters and the size of the input image used during training to produce a network architecture that resulted in an average NRMSE that was approximately 6.25% better than the baseline DfD-Net average NRMSE. This optimized architecture also resulted in an average NMAE that was 5.25% better than the baseline DfD-Net average NMAE. Only the SSIM metric did not see a gain in performance, dropping by 0.26% when compared to the baseline DfD-Net average SSIM value. The second method illustrates the use of a Self Organizing Map clustering method to reduce the number convolutional filters in the DfD-Net to reduce the overall run time of the architecture while still retaining the network performance exhibited prior to the reduction. This method produces a reduced DfD-Net architecture that has a run time decrease of between 14.91% and 44.85% depending on the hardware architecture that is running the network. The final reduced DfD-Net resulted in a network architecture that had an overall decrease in the average NRMSE value of approximately 3.4% when compared to the baseline, unaltered DfD-Net, mean NRMSE value. The NMAE and the SSIM results for the reduced architecture were 0.65% and 0.13% below the baseline results respectively. This illustrates that reducing the network architecture complexity does not necessarily reduce the reduction in performance. Finally, this research introduced a new, real world dataset that was captured using a camera and a voltage controlled microfluidic lens to capture the visual data and a 2-D scanning LIDAR to capture the ground truth data. The visual data consists of images captured at seven different exposure times and 17 discrete voltage steps per exposure time. The objects in this dataset were divided into four repeating scene patterns in which the same surfaces were used. These scenes were located between 1.5 and 2.5 meters from the camera and LIDAR. This was done so any of the deep learning algorithms tested would see the same texture at multiple depths and multiple blurs. The DfD-Net architecture was employed in two separate tests using the real world dataset. The first test was the synthetic blurring of the real world dataset and assessing the performance of the DfD-Net trained on the Middlebury dataset. The results of the real world dataset for the scenes that were between 1.5 and 2.2 meters from the camera the DfD-Net trained on the Middlebury dataset produced an average NRMSE, NMAE and SSIM value that exceeded the test results of the DfD-Net tested on the Middlebury test set. The second test conducted was the training and testing solely on the real world dataset. Analysis of the camera and lens behavior led to an optimal lens voltage step configuration of 141 and 129. Using this configuration, training the DfD-Net resulted in an average NRMSE, NMAE and SSIM of 0.0660, 0.0517 and 0.8028 with a standard deviation of 0.0173, 0.0186 and 0.0641 respectively.

The feedforward neural network (NN) is a mathematical model capable of representing any non-linear relationship between input and output data. It has been succesfully applied to a wide variety of classification and function approximation problems. Various neural network training algorithms were developed, including the particle swarm optimiser (PSO), which was shown to outperform the standard back propagation training algorithm on a selection of problems. However, it was usually assumed that the environment in which a NN operates is static. Such an assumption is often not valid for real life problems, and the training algorithms have to be adapted accordingly. Various dynamic versions of the PSO have already been developed. This work investigates the applicability of dynamic PSO algorithms to NN training in dynamic environments, and compares the performance of dynamic PSO algorithms to the performance of back propagation. Three popular dynamic PSO variants are considered. The extent of adaptive properties of back propagation and dynamic PSO under different kinds of dynamic environments is determined. Dynamic PSO is shown to be a viable alternative to back propagation, especially under the environments exhibiting infrequent gradual changes. Copyright 2011, University of Pretoria. All rights reserved. The copyright in this work vests in the University of Pretoria. No part of this work may be reproduced or transmitted in any form or by any means, without the prior written permission of the University of Pretoria. Please cite as follows: Rakitianskaia, A 2011, Using particle swarm optimisation to train feedforward neural networks in dynamic environments, MSc dissertation, University of Pretoria, Pretoria, viewed yymmdd < http://upetd.up.ac.za/thesis/available/etd-02132012-233212 / > C12/4/406/gm / Dissertation (MSc)--University of Pretoria, 2011. / Computer Science / Unrestricted

In the literature, various Mobile Ad hoc NETwork (MANET) routing protocols proposed. Each performs the best under specific context conditions, for example under high mobility or less volatile topologies. In existing MANET, the degradation in the routing protocol performance is always associated with changes in the network context. To date, no MANET routing protocol is able to produce optimal performance under all possible conditions. The core aim of this thesis is to solve the routing problem in mobile Ad hoc networks by introducing an optimum system that is in charge of the selection of the running routing protocol at all times, the system proposed in this thesis aims to address the degradation mentioned above. This optimisation system is a novel approach that can cope with the network performance’s degradation problem by switching to other routing protocol. The optimisation system proposed for MANET in this thesis adaptively selects the best routing protocol using an Artificial Intelligence mechanism according to the network context. In this thesis, MANET modelling helps in understanding the network performance through different contexts, as well as the models’ support to the optimisation system. Therefore, one of the main contributions of this thesis is the utilisation and comparison of various modelling techniques to create representative MANET performance models. Moreover, the proposed system uses an optimisation method to select the optimal communication routing protocol for the network context. Therefore, to build the proposed system, different optimisation techniques were utilised and compared to identify the best optimisation technique for the MANET intelligent system, which is also an important contribution of this thesis. The parameters selected to describe the network context were the network size and average mobility. The proposed system then functions by varying the routing mechanism with the time to keep the network performance at the best level. The selected protocol has been shown to produce a combination of: higher throughput, lower delay, fewer retransmission attempts, less data drop, and lower load, and was thus chosen on this basis. Validation test results indicate that the identified protocol can achieve both a better network performance quality than other routing protocols and a minimum cost function of 4.4%. The Ad hoc On Demand Distance Vector (AODV) protocol comes in second with a cost minimisation function of 27.5%, and the Optimised Link State Routing (OLSR) algorithm comes in third with a cost minimisation function of 29.8%. Finally, The Dynamic Source Routing (DSR) algorithm comes in last with a cost minimisation function of 38.3%.

A Wireless Sensor Network (WSN) is a group of spatially dispersed sensors that monitor the physical conditions of the environment and collect data at a central location. Sensor deployment is one of the main design aspects of WSNs as this a ffects network coverage. In general, WSN deployment methods fall into two categories: planned deployment and random deployment. This thesis considers planned sensor deployment of a Mobile Wireless Sensor Network (MWSN), which is defined as selectively deciding the locations of the mobile sensors under the given constraints to optimize the coverage of the network. Metaheuristic algorithms are powerful tools for the modeling and optimization of problems. The Krill Herd Algorithm (KHA) is a new nature-inspired metaheuristic algorithm which can be used to solve the sensor deployment problem. A Multi-Agent System (MAS) is a system that contains multiple interacting agents. These agents are autonomous entities that interact with their environment and direct their activity towards achieving speci c goals. Agents can also learn or use their knowledge to accomplish a mission. Multi-agent systems can solve problems that are very difficult or even impossible for monolithic systems to solve. In this work, a modification of KHA is proposed which incorporates MAS to obtain a Multi-Agent Krill Herd Algorithm (MA-KHA). To test the performance of the proposed method, five benchmark global optimization problems are used. Numerical results are presented which show that MA-KHA performs better than the KHA by finding better solutions. The proposed MA-KHA is also employed to solve the sensor deployment problem. Simulation results are presented which indicate that the agent-agent interactions in MA-KHA improves the WSN coverage in comparison with Particle Swarm Optimization (PSO), the Firefly Algorithm (FA), and the KHA. / Graduate

This dissertation focuses on daily volt/var control in distribution networks with feeder capacitors, substation capacitors and transformers equipped with on-load tap changers. A hybrid approach is proposed to solve the daily volt/var control problem. To reduce the computational requirements of the problem, this approach combines two methods, namely heuristic and optimal scheduling for the substation and feeder sub-problems respectively. The feeder capacitor dispatch schedule is determined based on a heuristic reactive power setpoint method. At this stage the objective is to minimize the reactive power flow through the substation bus in every time-interval. And as such, mathematical modeling of the distribution network components is adapted to suit time-varying conditions. Furthermore, an optimization model to determine a proper dispatch schedule of the substation devices is formulated. The objective of this model is to minimize the daily total energy loss and voltage deviations. Additionally, the reference voltage of the substation secondary bus and the transformer tap position limits are modified to adapt to given load profiles. The optimization model is solved with a discrete particle swarm optimization algorithm, which incorporates Newton’s method to determine the power-flow solution. The proposed method is applied to a time-varying distribution system and evaluated under different operational scenarios. It is also compared to on-line volt/var control with various settings. Simulation results show that the proposed approach minimizes both the voltage deviations and the total energy loss, while on-line control prioritizes one objective over the other depending on the specified settings. / Dissertation (MEng)--University of Pretoria, 2015. / Electrical, Electronic and Computer Engineering / Unrestricted

The subjects of perfect information games, machine learning and computational intelligence combine in an experiment that investigates a method to build the skill of a game-playing agent from zero game knowledge. The skill of a playing agent is determined by two aspects, the first is the quantity and quality of the knowledge it uses and the second aspect is its search capacity. This thesis introduces a novel representation language that combines symbols and numeric elements to capture game knowledge. Insofar search is concerned; an extension to an existing knowledge-based search method is developed. Empirical tests show an improvement over alpha-beta, especially in learning conditions where the knowledge may be weak. Current machine learning techniques as applied to game agents is reviewed. From these techniques a learning framework is established. The data-mining algorithm, ID3, and the computational intelligence technique, Particle Swarm Optimisation (PSO), form the key learning components of this framework. The classification trees produced by ID3 are subjected to new post-pruning processes specifically defined for the mentioned representation language. Different combinations of these pruning processes are tested and a dominant combination is chosen for use in the learning framework. As an extension to PSO, tournaments are introduced as a relative fitness function. A variety of alternative tournament methods are described and some experiments are conducted to evaluate these. The final design decisions are incorporated into the learning frame-work configuration, and learning experiments are conducted on Checkers and some variations of Checkers. These experiments show that learning has occurred, but also highlights the need for further development and experimentation. Some ideas in this regard conclude the thesis. / Dissertation (MSc)--University of Pretoria, 2007. / Computer Science / MSc / Unrestricted

The object of this research is to optimize the overall power system performance using FACTS devices. Particularly, it is intended to improve the reliability, and the performance of the power system considering steady state operating condition as well as the system subjected to small and large disturbances. The methodology proposed to achieve this goal corresponds to an enhanced particle swarm optimizer (Enhanced-PSO) that is proven in this work to have several advantages, in terms of accuracy and computational effort, as compared with other existing methods. Once the performance of the Enhanced PSO is verified, a multi-stage PSO-based optimization framework is proposed for optimizing the power system reliability (N-1 contingency criterion). The algorithm finds optimal settings for present infrastructure (generator outputs, transformers tap ratios and capacitor banks settings) as well as optimal control references for distributed static series compensators (DSSC) and optimal locations, sizes and control settings for static compensator (STATCOM) units. Finally, a two-stage optimization algorithm is proposed to improve the power system performance in steady state conditions and when small and large perturbations are applied to the system. In this case, the algorithm provides optimal control references for DSSC modules, optimal location and sizes for capacitor banks, and optimal location, sizes and control parameters for STATCOM units (internal and external controllers), so that the loadability and the damping of the system are maximized at minimum cost. Simulation results throughout this research show a significant improvement of the power system reliability and performance after the system is optimized.

Submitted by Irene Nascimento (irene.kessia@ufpe.br) on 2017-01-30T18:17:26Z No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) DISSERTACAO_LEO_DEFESA - FINAL - CATALOGADA PDF.pdf: 4585329 bytes, checksum: 4b70c80127866cd2da97a6217bb6a34f (MD5) / Made available in DSpace on 2017-01-30T18:17:27Z (GMT). No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) DISSERTACAO_LEO_DEFESA - FINAL - CATALOGADA PDF.pdf: 4585329 bytes, checksum: 4b70c80127866cd2da97a6217bb6a34f (MD5) Previous issue date: 2016-08-16 / CNPQ / Um dos principais desafios no projeto de sensores baseados em Ressonância de Plasmons de Superfície — RPS — é maximizar sua sensibilidade. Neste trabalho é proposto o uso de dois algoritmos heurísticos, Monte Carlo e Enxame de Partículas, para otimização de sensores baseados em RPS em interfaces planares, i.e, nas configurações de Kretschmann e de Otto, sem o auxílio da aproximação lorentziana para a curva de ressonância. Devido à natureza probabilística dos algoritmos, consegue-se obter um método simples e robusto para atingir essa otimização. É feita uma comparação quanto à eficiência computacional dos algoritmos em relação ao método tradicional de otimização, ficando demonstrado que o método de Enxame de Partículas é o mais eficiente em relação às outras técnicas. Com o emprego desse método, a dependência espectral dos parâmetros ótimos é obtida para sensores utilizando vários metais nas configurações de Kretschmann e de Otto, tanto para aplicações em meios gasosos quanto em meios aquosos. Um aplicativo foi desenvolvido e sua funcionalidade demonstrada, que pode ser executado diretamente via web, com base na metodologia proposta, para otimização de sensores RPS em interfaces planares. / One of the main challenges in the design of surface plasmon resonance – SPR – sensor systems is to maximize their sensitivity. In this work one proposes the use of two heuristic algorithms, Monte Carlo and Particle Swarm, for optimization of SPR sensors in planar interfaces, i.e, in the Kretschmann and Otto configurations, without use of the Lorentzian approximation to the resonance curve. Because of the probabilistic nature of the algorithms, one manages to obtain a simple and robust method to achieve optimization. A comparison is made on the computational efficiency of the algorithm relative to the traditional method of optimization, showing that the particle swarm optimization method is more efficient compared to other techniques. By employing this method, the spectral dependence of optimum parameters is obtained for sensors using a wide range of metal films in the Kretschmann and Otto configurations, both for applications in gaseous an in aqueous media. An app was developed and its functionality can be demonstrated, by direct execution via web, based on the proposed methodology for optimization of SPR sensors on planar interfaces.

Context. Automatic placement of surveillance cameras&apos; 3D models in an arbitrary floor plan containing obstacles is a challenging task. The problem becomes more complex when different types of region of interest (RoI) and minimum resolution are considered. An automatic camera placement decision support system (ACP-DSS) integrated into a 3D CAD environment could assist the surveillance system designers with the process of finding good camera settings considering multiple constraints. Objectives. In this study we designed and implemented two subsystems: a camera toolset in SketchUp (CTSS) and a decision support system using an enhanced Particle Swarm Optimization (PSO) algorithm (HPSO-DSS). The objective for the proposed algorithm was to have a good computational performance in order to quickly generate a solution for the automatic camera placement (ACP) problem. The new algorithm benefited from different aspects of other heuristics such as hill-climbing and greedy algorithms as well as a number of new enhancements. Methods. Both CTSS and ACP-DSS were designed and constructed using the information technology (IT) research framework. A state-of-the-art evolutionary optimization method, Hybrid PSO (HPSO), implemented to solve the ACP problem, was the core of our decision support system. Results. The CTSS is evaluated by some of its potential users after employing it and later answering a conducted survey. The evaluation of CTSS confirmed an outstanding satisfactory level of the respondents. Various aspects of the HPSO algorithm were compared to two other algorithms (PSO and Genetic Algorithm), all implemented to solve our ACP problem. Conclusions. The HPSO algorithm provided an efficient mechanism to solve the ACP problem in a timely manner. The integration of ACP-DSS into CTSS might aid the surveillance designers to adequately and more easily plan and validate the design of their security systems. The quality of CTSS as well as the solutions offered by ACP-DSS were confirmed by a number of field experts. / Sarmad Rohani: 004670606805 Reza Shams: 0046704030897

Recently, continuous-valued optimization problems have received a great amount of focus, resulting in optimization algorithms which are very efficient within the continuous-valued space. Many optimization problems are, however, defined within the binary-valued problem space. These continuous-valued optimization algorithms can not operate directly on a binary-valued problem representation, without algorithm adaptations because the mathematics used within these algorithms generally fails within a binary problem space. Unfortunately, such adaptations may alter the behavior of the algorithm, potentially degrading the performance of the original continuous-valued optimization algorithm. Additionally, binary representations present complications with respect to increasing problem dimensionality, interdependencies between dimensions, and a loss of precision. This research investigates the possibility of applying continuous-valued optimization algorithms to solve binary-valued problems, without requiring algorithm adaptation. This is achieved through the application of a mapping technique, known as angle modulation. Angle modulation effectively addresses most of the problems associated with the use of a binary representation by abstracting a binary problem into a four-dimensional continuous-valued space, from which a binary solution is then obtained. The abstraction is obtained as a bit-generating function produced by a continuous-valued algorithm. A binary solution is then obtained by sampling the bit-generating function. This thesis proposes a number of population-based angle-modulated continuous-valued algorithms to solve binary-valued problems. These algorithms are then compared to binary algorithm counterparts, using a suite of benchmark functions. Empirical analysis will show that the angle-modulated continuous-valued algorithms are viable alternatives to binary optimization algorithms. Copyright 2012, University of Pretoria. All rights reserved. The copyright in this work vests in the University of Pretoria. No part of this work may be reproduced or transmitted in any form or by any means, without the prior written permission of the University of Pretoria. Please cite as follows: Pamparà, G 2012, Angle modulated population based algorithms to solve binary problems, MSc dissertation, University of Pretoria, Pretoria, viewed yymmdd < http://upetd.up.ac.za/thesis/available/etd-02242012-090312 / > C12/4/188/gm / Dissertation (MSc)--University of Pretoria, 2012. / Computer Science / unrestricted