No description available.

No description available.

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

This Master's thesis investigates a method to measure atomic nuclei with record precision using muonic atom X-ray spectroscopy. In particular, 6Li is measured experimentally. The method used is independent from the previous most precise measurement of the 6Li nuclear charge radius which uses electron scattering. Measuring low-Z elements using muonic X-ray transitions requires excellent detectors which have so far been mostly optimised for higher energies. This project investigates methods to reach precision requirements for low-Z elements which can yield insight into nuclear structure models, and uses a Silicon Drift Detector (SDD) which is here characterised in detail and found to allow for significantly improved results over previous attempts. So far, the SDD and developed calibration scheme demonstrates a 3.7 eV precision compared to the target 0.5 eV. It appears to be limited by detector resolution, which also makes curve fitting difficult for complex line structures. A new method for generating calibration lines, X-ray fluorescence, is tested and shows good promise for future use. The planned use of a Metallic Magnetic Microcalorimeter will potentially improve results significantly, having a much-improved resolution over SDD's. Preliminary experimental results find ΔEµLi-6, 2p-1s = 18780.6 ± 15.7 eV, which is a factor of 4 improvement over the previous best measurement of this transition and the world's most precise measurement to date. While the uncertainty is larger than seen in designated calibration runs, it demonstrates the ability to perform high-precision muonic atom spectroscopy. With new detector technologies, this thesis finds no immediate obstacles to the target 0.5 eV precision.

No description available.

Drift is a point of contention with pesticide applications, causing the need to research application methods that provide consistent efficacy while minimizing off-target movement. Experiments were conducted to evaluate eight undiluted herbicides on invasive woody plants, Chinese tallow (Triadica sebifera) and Callery pear (Pyrus calleryana), when applied individual plant treatment (IPT) via hack-and-squirt. Applications of undiluted aminocyclopyrachlor or imazapyr at 1 ml per 7.6 cm of tree diameter at breast height (DBH) made in the spring provided superior control over other herbicides or application timings. CamelBak® hydration reservoirs were evaluated for storage durability with eight undiluted herbicides. A third study was conducted to assess droplet size and distribution of Roadside Inc.’s new sprayer head for driftable fines. All nozzles were evaluated in a wind tunnel and produced droplet sizes above the benchmark for driftable fines (≤150 μm). The spray head also distributed droplets effectively from 2-30 feet from spray origin.

For neutral genes, uniparental inheritance is expected to reduce effective population size relative to biparentally inherited genes. In finite populations, the ensuing genetic drift can cause stronger spatial and temporal differentiation. An intrapopulation polymorphism in chloroplast DNA was used to examine relative spatial and temporal population structure of chloroplast and allozyme markers in the annual plant Phacelia dubia. There was significant differentiation among populations at chloroplast markers but not for allozyme loci. A fine-scale analysis showed significant structure among sites within populations for chloroplast markers and local heterozygote deficiencies at allozyme loci. These spatial analyses suggest that gene flow via pollen exceeds that via seed. Temporal variation in chloroplast markers, assessed over a 10-year period, was evident in two of four populations, and allozyme loci were characterized by temporal variation in rare-allele frequencies. Population structure appeared to be related to the intensity and type of human disturbance influencing each population. Habitat destruction promoted isolation and enhanced differentiation, whereas mowing increased seed dispersal and reduced differentiation for chloroplast markers. At this time, genetic drift appears to be the primary force shaping chloroplast gene frequencies.

Electromagnetic induction (EMI) has been used in archaeology for decades, but still lags in use and development when compared to magnetometry and ground-penetrating radar. While it has become more popular than electrical resistivity area survey, it is now less commonly used than electrical resistivity tomography. The EMI method is likely underutilized due to drift problems and a lack of multi-sensor, vehicle-towed systems capable of rapid, high-density data collection. In this article we demonstrate not only the effectiveness of EMI survey, but a case where entire villages would have remained undetected without it. At the Singer-Hieronymus Site in central Kentucky, USA, a vehicle-towed frequency domain EMI survey detected the location of plazas, residential areas, and trash disposal areas across multiple Fort Ancient villages that contained both intact and heavily disturbed deposits. Additionally, three new villages were revealed. Through this process, we discovered how Fort Ancient village dynamics may be studied through a geophysical investigation of village shape, size, and spatial organization.

No description available.

Severe earthquakes cause damage to buildings. One measure of damage is the residual drift. Large residual drifts suggest expensive repairs and could lead to complete loss of the building. As such, research has been conducted on how to reduce the residual drift. Recent research has focused on self-centering frames and dual systems, both of which increase the post-yield stiffness of the building during and after an earthquake. Self-centering systems have yet to be adopted into standard practice but dual systems are used regularly. Dual systems in steel buildings typically combine two types of traditional lateral force resisting systems such as bucking restrained braced frames (BRBFs) and moment resisting frames (MRFs). However, the cost of making the moment connections for the MRFs can make dual systems costly. An alternative to MRFs is to use gravity columns as the secondary system in a dual system. The gravity columns can be used to help resist the lateral loads and limit the residual drifts if the lateral stiffness of the gravity columns can be activated. By restraining the displacement of the gravity columns, the stiffness of the columns adds to the stiffness of the brace frame, thus engaging the lateral stiffness of the gravity columns. Three methods of engaging the stiffness of the gravity columns are investigated in this thesis; one, fixed ground connections, two, a heavy elastic brace in the top story, and three, a heavy elastic brace in the middle bay. Single and multiple degree of freedom models were analyzed to determine if gravity columns can be effective in reducing residual drift. In the single degree of freedom system (SDOF) models, the brace size was varied to get a range of periods. The column size was varied based on a predetermined range of post-yield stiffness to determine if the residual drift decreased with higher post-yield stiffness. Three and five story models were analyzed with a variety of brace and column sizes and with three different configurations to activate the gravity columns. Using gravity columns as part of a dual system decreases the residual drift in buildings. The results from the SDOF system show that the residual drift decreased with increased post-yield stiffness. The three and five story models showed similar results with less residual drift when larger columns were used. Further, the models with a heavy gravity column in the top story had the best results.