This paper exposes problems of the commonly used technique of splitting the available data in neural spatial interaction modelling into training, validation, and test sets that are held fixed and warns about drawing too strong conclusions from such static splits. Using a bootstrapping procedure, we compare the uncertainty in the solution stemming from the data splitting with model specific uncertainties such as parameter initialization. Utilizing the Austrian interregional telecommunication traffic data and the differential evolution method for solving the parameter estimation task for a fixed topology of the network model [ i.e. J = 9] this paper illustrates that the variation due to different resamplings is significantly larger than the variation due to different parameter initializations. This result implies that it is important to not over-interpret a model, estimated on one specific static split of the data. (authors' abstract) / Series: Discussion Papers of the Institute for Economic Geography and GIScience

The aim of the thesis is to address issues concerning modelling and evaluation of monetary policy by obtaining targeting rules from optimisation techniques using welfare loss functions that capture asymmetries and zone targeting behaviours. The motivation is that the specification of the most widely used monetary policy rule, i.e. the Taylor rule, may not adequately capture the stylised key features of monetary policy practice as has been shown by Nobay and Peel (2003), Aksoy et al. (2006) and Boinet and Martin (2008). The thesis also addresses the importance of the behaviour of certain financial asset prices and their implications in monetary policy decision making. It also analyses the impact of uncertainty about the true state of the economy on domestic interest rates. First, the response of monetary policy to deviations of inflation and output from their target values based on a framework that allows asymmetric and zone targeting monetary authorities’ preferences is estimated.1 Second, the monetary policy reaction function, which is augmented with a comprehensive index that collects and synthesises information from the financial asset markets is estimated for South Africa based on a framework that allows asymmetric and zone targeting monetary authorities’ preferences.2 Third, the impact of uncertainty about the state of the economy on monetary policy in South Africa using a framework that allows asymmetric and zone targeting monetary authorities’ preferences is analysed. The main findings are that the monetary authorities’ response towards inflation is zone symmetric and their response to output fluctuations is asymmetric. The second major finding is that the conditions in the financial asset markets form an important information set for the monetary authorities and that the monetary authorities pay close attention to the conditions in these markets by placing an equal weight on financial asset markets booms and recessions. The empirical results also reveal a significant impact of uncertainty about the state of the economy on domestic interest rates during the inflation targeting period and that the monetary authorities exhibit discretionary behaviour when implementing monetary policy under uncertainty. The thesis contributes to the body of knowledge in the field of economics by addressing important issues in monetary policy design and conduct using a framework that capture the stylised key features of monetary policy practice. All these issues are important in design and conduct of monetary policy. They are currently debated at many central banks including South Africa. / Thesis (PhD)--University of Pretoria, 2011. / Economics / unrestricted

Au cours des dernières décennies, de grands efforts en recherche et développement ont été faits pour développer et promouvoir les véhicules électriques (VEs). La plupart de ces recherches portent essentiellement sur le développement des moteurs électriques de ces véhicules et des technologies de batteries de recharge. Cependant, un des obstacles majeurs pour le déploiement des VEs à grande échelle réside dans l'incertitude d’assister et de guider les conducteurs de ce type de véhicule d’une façon appropriée pour atteindre les stations de recharge tout en satisfaisant leurs souhaits (points de recharge disponibles, moins d’attente possible, proposition d’autres points d’intérêts : restaurant, shopping, etc.). Afin de remédier à ce manque, nous proposons dans ce travail de thèse une approche distribuée et adaptative orientée modèles pour la gestion de l'énergie pour la recharge des VEs. Pour ce faire, nous nous somme focalisés sur la modélisation des processus de recharge en utilisant une approche formelle basée sur des outils de systèmes à événements discrets, à savoir l'algèbre (max, +) et les réseaux de Petri. Les modèles développés ont permis d’étudier, d’analyser et d’évaluer le comportement du système de recharge. De plus, une approche d'optimisation basée sur la programmation linéaire est proposée afin d’affecter et d’orienter d'une façon optimale les VEs vers les stations de recharge appropriées et ordonnancer leurs opérations de recharge. Afin de prédire le taux et la durée de recharge moyens des VEs compte tenu des dates d’arrivée des demandes de recharge et l'état de recharge de chaque véhicule, une approche dédiée basée sur une fonction prédictive est proposée. En utilisant cette approche, les opérations de recharge pourraient être planifiées en minimisant les temps d'attente des VEs au sein des stations de recharge et en assurant un taux de recharge acceptable pour chaque demande. Les résultats d’analyse et de simulations obtenus ont montré que les approches de modélisation, d’optimisation et de prédiction proposées permettent d’affecter de façon adéquate et optimale les VEs aux stations de recharge tout en satisfaisant toutes les contraintes du processus de recharge. / In the last decades, very great research and development efforts have been made to develop and promote electric vehicles (EVs). Most efforts have been made to further develop the power engine of these vehicles and batteries technologies. However, one of the major obstacles to the large deployment of EVs is the uncertainty of drivers to get a suitable and vacant place at a charging station (CS). In this manuscript, we focus on the charging process modelling using formal approaches based on discrete event system tools namely (max,+) algebra and Petri nets. In addition, an optimization approach based on linear programming is proposed to optimally assign and reroute EVs to the suitable CSs and schedule their charging operations. In order to predict, manage and handle charging needs of EVs, a dedicated model based on a predictive function is introduced. The aim is to predict the average charging rate and time while considering the inter-arrival of charging requests and the state of charging of EVs. Using this approach, charging operations could be planned while minimizing waiting times of EVs and avoiding queuing situations within CSs. Simulation results showed that the proposed approaches allow assigning adequately and optimally EVs to CSs while satisfying all process constraints.

A novel fuzzy system named as 2-way fuzzy system is developed by combining the intuitionistic fuzzy set theory with the fuzzy systems theory. The developed system is used in modelling and minimizing uncertainty and inconsistency. Uncertainty is the width of the interval introduced by the independent assignment of membership and nonmembership functions of the intuitionistic fuzzy sets / and inconsistency is the violation of the consistency inequality in this assignment. The uncertainty and inconsistency is reduced through a 2 phase training. An evaluation of the degree of reduction of inconsistency is carried out at the end of the rst phase of training by forming the shadowed set patterns of the membership and nonmembership functions. The system is further trained for a second phase in order to reduce uncertainty. There are three dierent methods developed for the stability analysis of fuzzy systems. The rst method is based on the approximating sequences technique, and the design turns into an optimal control problem. In the sec ond analysis, describing function of a 2-way fuzzy system is evaluated analytically, and a systematic design approach is developed using describing function technique. The last analysis technique employs the Lie algebra theory in the stability analysis of Takagi-Sugeno fuzzy systems. The theoretical results are simulated on an application system, which is a exible-joint robot arm system.

The proven ability to ferment Saccharomyces cerevisiae on a large scale presents an attractive target for producing chemicals and fuels from sustainable sources. Efficient and predominant carbon flux through to ethanol is a significant engineering issue in the development of this yeast as a multi-product cell chassis used in biorefineries. In order to evaluate diversion of carbon flux away from ethanol, combinatorial deletions were investigated in genes encoding the six isozymes of alcohol dehydrogenase (ADH), which catalyse the terminal step in ethanol production. The scarless, dominant and counter- selectable amdSYM gene deletion method was optimised for generation of a combinatorial ADH knockout library in an industrially relevant strain of S. cerevisiae. Current understanding of the individual ADH genes fails to fully evaluate genotype-by-genotype and genotype-by-environment interactions: rather, further research of such a complex biological process requires a multivariate mathematical modelling approach. Application of such an approach using the Design of Experiments (DoE) methodology is appraised here as essential for detailed empirical evaluation of complex systems. DoE provided empirical evidence that in S. cerevisiae: i) the ADH2 gene is not associated with producing ethanol under anaerobic culture conditions in combination with 25 g l-1 glucose substrate concentrations; ii) ADH4 is associated with increased ethanol production when the cell is confronted with a zinc-limited [1 μM] environment; and iii) ADH5 is linked with the production of ethanol, predominantly at pH 4.5. A successful metabolic engineering strategy is detailed which increases the product portfolio of S. cerevisiae, currently used for large-scale production of bioethanol. Heterologous expression of the cytochrome P450 fatty acid peroxygenase from Jeotgalicoccus sp., OleTJE, fused to the RhFRED reductase from Rhodococcus sp. NCIMB 978 converted free fatty acid precursors to C13, C15 and C17 alkenes (3.81 ng μl-1 total alkene concentration).