Individuals of the cnidarian genus Physalia are a common sight at New Zealand beaches and are the primary cause of jellyfish stings to beachgoers each year. The identity of the species and the environmental factors that determine its presence are unknown. Lack of knowledge of many marine species is not unusual, as pelagic invertebrates often lack detailed taxonomic descriptions as well as information about their dispersal mechanisms such that meaningful patterns of distribution and dispersal are almost impossible to determine. Molecular systematics has proven to be a powerful tool for species identification and for determining geographical distributions. However, other techniques are needed to indicate the causal mechanisms that may result in a particular species distribution. The aim of this study was to apply molecular techniques to the cnidarian genus Physalia to establish which species occur in coastal New Zealand, and to apply models to attempt to forecast its occurrence and infer some mechanisms of dispersal. Physalia specimens were collected from New Zealand, Australia and Hawaii and sequenced for Cytochrome c oxidase I (COI) and the Internal transcribed spacer 1 (ITS1). Three clans were found: a Pacific-wide clan, an Australasian clan and New Zealand endemic clan with a distribution confined to the Bay of Plenty and the East Coast of the North Island. Forecasting Physalia occurrence directly from presence data using artificial neural networks (ANN) proved unsuccessful and it was necessary to pre-process the presence data using a variable sliding window to reduce noise and improve accuracy. This modelling approach outperformed the time lagged based networks giving improved forecasts in both regions that were assessed. The ANN models were able to indicated significant trends in the data but would require more data at higher resolution to give more accurate forecasts of Physalia occurrence suitable for decision making on New Zealand beaches. To determine possible causal mechanisms of recorded occurrences and to identify possible origins of Physalia the presence and absence of Physalia on swimming beaches throughout the summer season was modelled using ANN and Naϊve Bayesian Classifier (NBC). Both models were trained on the same data consisting of oceanographic variables. The modelling carried out in this study detected two dynamic systems, which matched the distribution of the molecular clans. One system was centralised in the Bay of Plenty matching the New Zealand endemic clan. The other involved a dynamic system that encompassed four other regions on both coasts of the country that matched the distribution of the other clans. By combining the results it was possible to propose a framework for Physalia distribution including a mechanism that has driven clan divergence. Moreover, potential blooming areas that are notoriously hard to establish for jellyfish were hypothesised for further study and/or validation.

<p>Vid analys av fluider med akustisk spektroskopi finns ett behov av att finna multivariata metoder för att utifrån akustiska spektra prediktera storheter såsom viskositet och densitet. Användning av artificiella neurala nätverk och bayesiska nätverk för detta syfte utreds genom teoretiska och praktiska undersökningar. Förbehandling och uppdelning av data samt en handfull linjära och olinjära multivariata analysmetoder beskrivs och implementeras. Prediktionsfelen för de olika metoderna jämförs och PLS (Partial Least Squares) framstår som den starkaste kandidaten för att prediktera de sökta storheterna.</p> / <p>When analyzing fluids using acoustic spectrometry there is a need of finding multivariate methods for predicting properties such as viscosity and density from acoustic spectra. The utilization of artificial neural networks and Bayesian networks for this purpose is analyzed through theoretical and practical investigations. Preprocessing and division of data along with a handful of linear and non-linear multivariate methods of analysis are described and implemented. The errors of prediction for the different methods are compared and PLS (Partial Least Squares) appear to be the strongest candidate for predicting the sought-after properties.</p>

Hidden Markov Models (HMMs) are, undoubtedly, the most employed core technique for Automatic Speech Recognition (ASR). Nevertheless, we are still far from achieving high‐performance ASR systems. Some alternative approaches, most of them based on Artificial Neural Networks (ANNs), were proposed during the late eighties and early nineties. Some of them tackled the ASR problem using predictive ANNs, while others proposed hybrid HMM/ANN systems. However, despite some achievements, nowadays, the dependency on Hidden Markov Models is a fact. During the last decade, however, a new tool appeared in the field of machine learning that has proved to be able to cope with hard classification problems in several fields of application: the Support Vector Machines (SVMs). The SVMs are effective discriminative classifiers with several outstanding characteristics, namely: their solution is that with maximum margin; they are capable to deal with samples of a very higher dimensionality; and their convergence to the minimum of the associated cost function is guaranteed. In this work a novel approach based upon probabilistic kernels in support vector machines have been attempted for speech data classification. The classification accuracy in case of support vector classification depends upon the kernel function used which in turn depends upon the data set in hand. But still as of now there is no way to know a priori which kernel will give us best results The kernel used in this work tries to normalize the time dimension by fitting a probability distribution over individual data points which normalizes the time dimension inherent to speech signals which facilitates the use of support vector machines since it acts on static data only. The divergence between these probability distributions fitted over individual speech utterances is used to form the kernel matrix. Vowel Classification, Isolated Word Recognition (Digit Recognition), have been attempted and results are compared with state of art systems.

Robot manipulators are considered as the key element in flexible manufacturing systems. Nonetheless, for a successful accomplishment of robot integration, the robots need to be accurate. The leading source of inaccuracy is the mismatch between the prediction made by the robot controller and the actual system. This work presents techniques for identification of actual kinematic parameters and pose accuracy compensation using a laser-based 3-D measurement system. In identification stage, both direct search and gradient methods are utilized. A computer simulation of the identification is performed using virtual position measurements. Moreover, experimentation is performed on industrial robot FANUC Robot R-2000iB/210F to test full pose and relative position accuracy improvements. In addition, accuracy obtained by classical parametric methodology is improved by the implementation of artificial neural networks. Neuro-parametric method proves an enhanced improvement in simulation results. The whole proposed theory is reflected by developed simulation software throughout this work while achieving accuracy nine times better when comparing before and after implementation.

In this thesis work, electricty generation modeling of the wind energy -one type of the renewable energy sources- is studied. The wind energy characteristics and the distribution of wind speed in a specific region is also examined. In addition, the power curves of the wind turbines are introduced and the relationship between the wind speed and wind power is explained. The generation characteristics of the wind turbines from various types of producers are also investigated. In this study, the main wind power forecasting methods are presented and the advantages and disadvantages of the methods are analyzed. The physical approaches, statistical methods and the Artificial Neural Network (ANN) methods are introduced. The parameters that affect the capacity factor, the total energy generation and the payback period are examined. In addition, the wind turbine models and their effect on the total energy generation with different wind data from various sites are explained. As a part of this study, a MATLAB-based software about wind speed and energy modelling and payback period calculation has been developed. In order to simplify the calculation process, a Graphical User Interface (GUI) has been designed. In addition, a simple wind energy persistence model for wind power plant operator in the intra-day market has been developed.

Vid analys av fluider med akustisk spektroskopi finns ett behov av att finna multivariata metoder för att utifrån akustiska spektra prediktera storheter såsom viskositet och densitet. Användning av artificiella neurala nätverk och bayesiska nätverk för detta syfte utreds genom teoretiska och praktiska undersökningar. Förbehandling och uppdelning av data samt en handfull linjära och olinjära multivariata analysmetoder beskrivs och implementeras. Prediktionsfelen för de olika metoderna jämförs och PLS (Partial Least Squares) framstår som den starkaste kandidaten för att prediktera de sökta storheterna. / When analyzing fluids using acoustic spectrometry there is a need of finding multivariate methods for predicting properties such as viscosity and density from acoustic spectra. The utilization of artificial neural networks and Bayesian networks for this purpose is analyzed through theoretical and practical investigations. Preprocessing and division of data along with a handful of linear and non-linear multivariate methods of analysis are described and implemented. The errors of prediction for the different methods are compared and PLS (Partial Least Squares) appear to be the strongest candidate for predicting the sought-after properties.

The significant amount of data contained in digital images present barriers to methods of learning from the information they hold. Noise and the subjectivity of image evaluation further complicate such automated processes. In this thesis, we examine a particular area in which these difficulties are experienced. We attempt to control the parameters of a multi-step algorithm that processes visual information. A framework for approaching the parameter selection problem using reinforcement learning agents is presented as the main contribution of this research. We focus on the generation of state and action space, as well as task-dependent reward. We first discuss the automatic determination of fuzzy membership functions as a specific case of the above problem. Entropy of a fuzzy event is used as a reinforcement signal. Membership functions representing brightness have been automatically generated for several images. The results show that the reinforcement learning approach is superior to an existing simulated annealing-based approach. The framework has also been evaluated by optimizing ten parameters of the text detection for semantic indexing algorithm proposed by Wolf et al. Image features are defined and extracted to construct the state space. Generalization to reduce the state space is performed with the fuzzy ARTMAP neural network, offering much faster learning than in the previous tabular implementation, despite a much larger state and action space. Difficulties in using a continuous action space are overcome by employing the DIRECT method for global optimization without derivatives. The chosen parameters are evaluated using metrics of recall and precision, and are shown to be superior to the parameters previously recommended. We further discuss the interplay between intermediate and terminal reinforcement.

An artificial neural network (ANN) method was developed for solving boundary value problems (BVPs) on an arbitrary irregular domain in such a manner that all Dirichlet and/or Neuman boundary conditions (BCs) are automatically satisfied. Exact satisfaction of BCs is not available with traditional numerical solution techniques such as the finite element method (FEM). The ANN is trained by reducing error in the given differential equation (DE) at certain points within the domain. Selection of these points is significantly simpler than the often difficult definition of meshes for the FEM. The approximate solution is continuous and differentiable, and can be evaluated at any location in the domain independent of the set of points used for training. The continuous solution eliminates interpolation required of discrete solutions produced by the FEM. Reducing error in the DE at a particular location in the domain does not necessarily imply improvement in the approximate solution there. A theorem was developed, proving that the solution will improve whenever error in the DE is reduced at all locations in the domain during training. The actual training of ANNs reasonably approximates the assumptions required by the proof. This dissertation offers a significant contribution to the field by developing a method for solving BVPs where all BCs are automatically satisfied. It had already been established in the literature that such automatic BC satisfaction is beneficial when solving problems on rectangular domains, but this dissertation presents the first method applying the technique to irregular domain shapes. This was accomplished by developing an innovative length factor. Length factors ensure BC satisfaction extrapolate the values at Dirichlet boundaries into the domain, providing a solid starting point for ANN training to begin. The resulting method has been successful at solving even nonlinear and non-homogenous BVPs to accuracy sufficient for typical engineering applications.

Large Eddy Simulation (LES) of bluff body stabilized premixed and partially premixed combustion close to the flammability limit is carried out in this thesis. The LES algorithm has no ad-hoc adjustable model parameters and is able to respond automatically to variations in the inflow conditions. Algorithm validation is achieved by comparison with reactive and non-reactive experimental data. In the reactive flow, two scalar closure models, Eddy Break-Up (EBULES) and Linear Eddy Mixing (LEMLES), are used and compared. Over important regions, the flame lies in the Broken Reaction Zone regime. Here, the EBU model assumptions fail. The flame thickness predicted by LEMLES is smaller and the flame is faster to respond to turbulent fluctuations, resulting in a more significant wrinkling of the flame surface. As a result, LEMLES captures better the subtle effects of the flame-turbulence interaction. Three premixed (equivalence ratio = 0.6, 0.65, and 0.75) cases are simulated. For the leaner case, the flame temperature is lower, the heat release is reduced and vorticity is stronger. As a result, the flame in this case is found to be unstable. In the rich case, the flame temperature is higher, and the spreading rate of the wake is increased due to the higher amount of heat release Partially premixed combustion is simulated for cases where the transverse profile of the inflow equivalence ratio is variable. The simulations show that for mixtures leaner in the core the vortical pattern tends towards anti-symmetry and the heat release decreases, resulting also in instability of the flame. For mixtures richer in the core, the flame displays sinusoidal flapping resulting in larger wake spreading. More accurate predictions of flame stability will require the use of detailed chemistry, raising the computational cost of the simulation. To address this issue, a novel algorithm for training Artificial Neural Networks (ANN) for prediction of the chemical source terms has been implemented and tested. Compared to earlier methods, the main advantages of the ANN method are in CPU time and disk space and memory reduction.

For the appropriate operation and correct short term planning, daily demand curve (DDC) of municipal water distribution networks should be forecasted beforehand. For that purpose, artificial neural networks (ANN) is used as a new method. The proposed approach employs already recorded DDCs extracted from the database of ASKI (Ankara Water Authority) SCADA center and related independent parameters such as temperature and relative humidity obtained from DMI (State Meteorological Institute). In this study, a computer model was developed in order to forecast hourly DDCs using Matlab and related modules. Parameters that affect the consumption of the water were determined as temperature, relative humidity, human behavior (weekend or workday) and season. Randomly selected days were taken into account for performance of the ANN model. Forecasted DDC values were compared with recorded data and consequently the model gives relatively satisfactory results, an average of 75% match according to R2 values for Ankara N8-3 network. Same architecture was applied for Antalya network give better results, average of 85%. For planning purposes / total volume and peak water consumption values for the selected recorded days, the day before recorded days, ANN forecasted days and seasonal average was compared and seasonal average gave relatively better results. Using the forecasted DDC, (i) performance analysis of the pressure zone and (ii) optimum valve setting evaluation for pressure management were realized. The results of the study may help water utilities for short term planning of a water distribution network, rehabilitation of elements, taking counter measures and setting the valve openings for minimizing leakage and optimizing customer conformity of the distribution network.