M.Ing. (Electrical and Electronic Engineering Science) / Scientific workflows (SWFs) and artificial neural networks (ANNs) have attracted the attention of researchers in many fields and have been used to solve a variety of problems. Examples of these are (a) the use of scientific workflows for the sensor web in the hydrology domain and (b), the use of ANNs for the prediction of a number of water resource variables such as rainfall, flow, water level and various other water quality variables. ANNs have proved to be a powerful tool for prediction when compared with statistical methods. The aims of this research are to develop ANNs that act as predictive models for water resources and to deploy these models as predictive tools in a scientific workflow environment. While there are guidelines in the literature relating to the factors affecting network performance, there is no standard approach that is universally accepted for determining the optimum architecture of a neural network for a given problem. The parameters of a neural network and for the learning algorithm have a major effect on the performance of the neural network. We consider various recurrent and feed-forward neural network architectures for predicting changes in the water levels of dams. We explore various' hidden layer dimensions in learning the characteristics of the training data using the back propagation learning algorithm. Trained networks are deployed as predictive model in a scientific workflows environment called VisTrails. ': We review and discuss the use of SWFs and ANNs in the hydrology domain with emphasis on the development of neural network architecture that will give the best predictions for water resources. A number of architectures are employed to examine the best accurate predictive network for historical rainfall data. The findings of training experiments are promising in terms of the use of ANNs as a water resources predictive tool. Experimental results showed how the architecture of a neural network impacts on its predictive performance. This study shows that the number of hidden nodes is important factor for the improvement of the quality of the predictions.

M.Ing. / Columbus Stainless, a subsidiary of Acerinox, manufactures stainless steel in their plant located in Middelburg, South Africa. During the hot rolling operation the steel is rolled on a 4-high finishing mill where strip movement perpendicular to the rolling direction occurs. This movement is undesirable because it causes inferior product quality and may also lead to downtime if the strip moves past the edge of the rolls. In the past the operator made adjustments to the relative alignment of the rolls in the mill in an attempt to limit the sideways movement of the strip. In order to improve product quality and production throughput, the manual action of adjusting the parallelism of the rolls was replaced with an automatic steering control system. Analysis of the process revealed that several variables have an impact on the way the strip reacts to changes in the alignment of rolls in the mill. An adaptive fuzzy logic control system was designed and implemented in the real time control system of the mill. During commissioning the system did not have an adverse effect on production and all initial project criteria were met, as was stipulated in Section 1.4 of this document. The control system improved the strip movement by an average of 11% on various products rolled. Based on production data, the system potentially prevented two coils from leaving the rolls during the month long evaluation period and saved 40 minutes of production time. If the savings in material losses and the potential gain in production time are added the possible anticipated monetary saving is estimated to be about 24 million Rand a year.

D.Ing. / This thesis deals with an investigation into facial recognition and some variables that influence the performance of such a system. Firstly there is an investigation into the influence of image variability on the overall recognition performance of a system and secondly the performance and subsequent suitability of a neural network based system is tested. Both tests are carried out on two distinctly different databases, one more variable than the other. The results indicate that the greater the amount of variability the more negatively affected is the performance rating of a specific facial recognition system. The results further indicate the success with the implementation of a neural network system over a more conventional statistical system.

A thesis submitted to the School of Economic and Business Sciences, Faculty of Commerce, Law and Management, University of the Witwatersrand in fulfilment of the requirements for the degree of Doctor of Philosophy (Ph/D). Johannesburg, South Africa June 2016 / In recent years, the debate on market efficiency has shifted to providing alternate forms of the hypothesis, some of which are testable and can be proven false. This thesis examines one such alternative, the Adaptive Market Hypothesis (AMH), with a focus on providing a framework for testing the dynamic (cyclical) notion of market efficiency using South African equity data (44 shares and six indices) over the period 1997 to 2014. By application of this framework, stylised facts emerged. First, the examination of market efficiency is dependent on the frequency of data. If one were to only use a single frequency of data, one might obtain conflicting conclusions. Second, by binning data into smaller sub-samples, one can obtain a pattern of whether the equity market is efficient or not. In other words, one might get a conclusion of, say, randomess, over the entire sample period of daily data, but there may be pockets of non-randomness with the daily data. Third, by running a variety of tests, one provides robustness to the results. This is a somewhat debateable issue as one could either run a variety of tests (each being an improvement over the other) or argue the theoretical merits of each test befoe selecting the more appropriate one. Fourth, analysis according to industries also adds to the result of efficiency, if markets have high concentration sectors (such as the JSE), one might be tempted to conclude that the entire JSE exhibits, say, randomness, where it could be driven by the resources sector as opposed to any other sector. Last, the use of neural networks as approximators is of benefit when examining data with less than ideal sample sizes. Examining five frequencies of data, 86% of the shares and indices exhibited a random walk under daily data, 78% under weekly data, 56% under monthly data, 22% under quarterly data and 24% under semi-annual data. The results over the entire sample period and non-overlapping sub-samples showed that this model's accuracy varied over time. Coupled with the results of the trading strategies, one can conclude that the nature of market efficiency in South Africa can be seen as time dependent, in line with the implication of the AMH. / MT2017

Orientador: José Paulo Fernandes Garcia / Banca: Marcelo Carvalho Minhoto Teixeira / Banca: Falcondes José Mendes de Seixas / Banca: Cristiano Quevedo Andrea / Banca: Ivan Nunes da Silva / Resumo: Neste trabalho investiga-se a capacidade de uma rede neural artificial, com ajustes de pesos em tempo real, executar o controle de sistemas por meio de uma estrutura de rastreamento de sinais em três contextos: inicialmente em uma série de sistemas lineares e não-lineares; em um segundo momento, a rede neural é utilizada no controle de sistemas sujeitos a incertezas paramétricas; e por fim, no controle de máquinas elétricas que podem ou não estar sujeitas a variações paramétricas, incertezas e perturbações lineares e não-lineares. Na primeira aplicação da rede neural artificial verifica-se o desempenho de rastreamento de sistemas lineares de 1 a , 2 a e 3 a ordens controlados em malha fechada por meio de simulações computacionais. Nesse sentido, calcula-se um índice de desempenho utilizando a integral do valor absoluto do erro (IAE - Integral of the Absolute Magnitude of the Error). Esse índice indica a proximidade da saída real do sistema com relação ao sinal de referência. Essa estrutura rastreadora de sinais poderá funcionar conjuntamente com controlador clássico Proporcional, Integral e Derivativo (PID). Testes são realizados utilizando controladores com estrutura variável e modos desli- zantes, os quais, são estratégias robustas às incertezas paramétricas do tipo casadas, todavia não apresentam o mesmo comportamento no que tange às incertezas do tipo não casadas. Dentro desse contexto, apresenta-se uma estratégia de controle utilizando redes neurais artificiais em conjunto com modos deslizantes para reduzir as influências de quaisquer tipos de incertezas e perturbações. A eficácia da estrutura de controle proposta é verificada por meio de simulações computacionais considerando um modelo de eixo lateral de um avião L-1011 em condições de voo. O controle de máquinas elétricas é realizado inicialmente em motor de corrente contínua e posteriormente em motor ... / Abstract: This thesis investigates the ability of an Artificial Neural Network (ANN), with real-time adjustable weights, to execute the control systems through a tracking structure for signals in three applications: in a series of linear and non-linear systems; to control systems subject to parametric uncertainties; and to control electrical machines that may be subject to linear and nonlinear disturbances and uncertainties. In the first application of ANN, it is verified the per- formance of tracking signals in systems of 1 st , 2 nd and 3 rd order through computer simulations results. In this regard, it estimates a performance index using the Integral of the Absolute va- lue of the Error (IAE), which indicates the difference between the system real output and the reference signal value. The proposed structure with the neural network is able to work with clas- sical compensators, the proportional, integral and derivative (PID) controller. Evaluation tests are performed using controllers with variable structure and sliding mode. This strategies pre- sents robustness to a class of parametric uncertainties, called matched parametric uncertainty. However, this technique is not robust related to unmatched uncertainty class. Thus, in this paper a control strategy is proposed based on ANN through sliding mode control technique to mini- mize the uncertainties and disturbances effects. In order to show the effectiveness of proposed method, simulation results are performed using a lateral axis model of an L-1011 in cruise flight conditions subject to the uncertainties and external disturbances. Initially, it is accompli- shed of direct current (DC) motors, and after that, the technique is applied to alternating current (AC) motors (three-phase induction). Through the combination of PID controller and ANN, some evaluations tests are performed in DC motors. The performance of induction motor has been addressed ... / Doutor

University of Macau / Faculty of Science and Technology. / Department of Civil and Environmental Engineering

This thesis examines the use of motion detection and analysis systems to detect falls and repetitive motion patterns of at-risk individuals. Three classes of motion are examined: Activities of daily living (ADL), falls, and repetitive motion. This research exposes a simple relationship between ADL and non-ADL movement, and shows how to use Principal Component Analysis and a kNN classifier to tell the 2 classes of motion apart with 100% sensitivity and specificity. It also identifies a more complex relationship between falls and repetitive motion, which both produce bodily accelerations exceeding 3G but differ with regard to their periodicity. This simplifies the classification problem of falls versus repetitive motion when taking into account that their data representations are similar except that repetitive motion displays a high degree of periodicity as compared to falls. / by Clyde Carryl. / Thesis (M.S.C.S.)--Florida Atlantic University, 2013. / Includes bibliography. / Mode of access: World Wide Web. / System requirements: Adobe Reader.

During the Leukocytapheresis (LCAP) process used to treat patients suffering from acute Ulcerative Colitis, medical practitioners have to continuously monitor the Hematocrit (Ht) level in the blood to ensure it is within the acceptable range. The work done, as a part of this thesis, attempts to create an early warning system that can be used to predict if and when the Ht values will deviate from the acceptable range. To do this we have developed an algorithm based on the Group Method of Data Handling (GMDH) and compared it to other Neural Network algorithms, in particular the Multi Layer Perceptron (MLP). The standard GMDH algorithm captures the fluctuation very well but there is a time lag that produces larger errors when compared to MLP. To address this drawback we modified the GMDH algorithm to reduce the prediction error and produce more accurate results. / by Jay Thakkar. / Pagination error. "References" should be leaves 63-67, and pagination end with leaf 67. / Thesis (M.S.C.S.)--Florida Atlantic University, 2011. / Includes bibliography. / Electronic reproduction. Boca Raton, Fla., 2011. Mode of access: World Wide Web.

The focus of this research is concerned with performing forecasting in technoeconomic contexts using a set of certain novel artificial neural networks (ANNs). Relevant efforts in general, entail the task of quantitatively estimating the details about the likelihood of future events (or unknown outcomes/effects) based on past and current information on the observed events (or known causes). Commensurate with the scope and objectives of the research, the specific topics addressed are as follows: A review on various methods adopted in technoeconomic forecasting and identified are econometric projections that can be used for forecasting via artificial neural network (ANN)-based simulations Developing and testing a compatible version of ANN designed to support a dynamic sigmoidal (squashing) function that morphs to the stochastical trends of the ANN input. As such, the network architecture gets pruned for reduced complexity across the span of iterative training schedule leading to the realization of a constructive artificial neural-network (CANN). Formulating a training schedule on an ANN with sparsely-sampled data via sparsity removal with cardinality enhancement procedure (through Nyquist sampling) and invoking statistical bootstrapping technique of resampling applied on the cardinality-improved subset so as to obtain an enhanced number of pseudoreplicates required as an adequate ensemble for robust training of the test ANN: The training and prediction exercises on the test ANN corresponds to optimally elucidating output predictions in the context of the technoeconomics framework of the power generation considered Prescribing a cone-of-error to alleviate over- or under-predictions toward prudently interpreting the results obtained; and, squeezing the cone-of-error to get a final cone-of-forecast rendering the forecast estimation/inference to be more precise Designing an ANN-based fuzzy inference engine (FIE) to ascertain the ex ante forecast details based on sparse sets of ex post data gathered in technoeconomic contexts - Involved thereof a novel method of .fusing fuzzy considerations and data sparsity.Lastly, summarizing the results with essential conclusions and identifying possible research items for future efforts identified as open-questions. / Includes bibliography. / Dissertation (Ph.D.)--Florida Atlantic University, 2014. / FAU Electronic Theses and Dissertations Collection

The design and construction of a tri-cable, planar robotic device for use in neurophysical rehabilitation is presented. The criteria for this system are based primarily on marketability factors, rather than ideal models or mathematical outcomes. The device is designed to be low cost and sufficiently safe for a somewhat disabled individual to use unsupervised at home, as well as in a therapist's office. The key features are the use of a barrier that inhibits the user from coming into contact with the cables as well as a "break-away" joystick that the user utilizes to perform the rehabilitation tasks. In addition, this device is portable, aesthetically acceptable and easy to operate. Other uses of this system include sports therapy, virtual reality and teleoperation of remote devices. / by Melissa M. Morris. / Includes a thesis demonstration video (QuickTImeMovie ; time [2:25] ; size [16.6MB] ; frame width [640] ; frame height [480]. / Thesis (M.S.C.S.)--Florida Atlantic University, 2007. / Includes bibliography.