<p>Saab Ericsson Space AB develops products for space for a predetermined price. Since the price is fixed, it is crucial to have a reliable prediction model to estimate the effort needed to develop the product. In general software effort estimation is difficult, and at the software department this is a problem.</p><p>By analyzing metrics, collected from former projects, different prediction models are developed to estimate the number of person hours a software project will require. Models for predicting the effort before a project begins is first developed. Only a few variables are known at this state of a project. The models developed are compared to a current model used at the company. Linear regression models improve the estimate error with nine percent units and nonlinear regression models improve the result even more. The model used today is also calibrated to improve its predictions. A principal component regression model is developed as well. Also a model to improve the estimate during an ongoing project is developed. This is a new approach, and comparison with the first estimate is the only evaluation.</p><p>The result is an improved prediction model. There are several models that perform better than the one used today. In the discussion, positive and negative aspects of the models are debated, leading to the choice of a model, recommended for future use.</p>

This thesis investigates the effects of the earth's rotation on internal waves from two perspectives of nonlinear internal wave theory: near-resonant triads and weakly nonlinear models. We apply perturbation theory (multiple scale analysis) to the governing equations of internal waves and develop a near-resonant internal wave triad theory. This theory explains a resonant-like phenomenon in the numerical results obtained from simulating internal waves generated by tide topography interaction. Furthermore, we find that the inclusion of the earth's rotation (nonzero $f$) in the numerical runs leads to a very special type of resonance: parametric subharmonic instability. Through using perturbation expansion to solve separable solutions to the governing equations of internal waves, we derive a new rotation modified KdV equation (RMKdV). Of particular interest, the dispersion relation of the new equation obeys the exact dispersion relation for internal waves for both small and moderate wavenumbers ($k$). Thus this new RMKdV is able to model wea kly nonlinear internal waves with various wavenumbers ($k$), better than the Ostrovsky equation which fails at describing waves of small $k$.

In this thesis, modeling and optimization in the field of storage management under stochastic condition will be investigated using two different methodologies: Simulation Optimization Techniques (SOT), which are usually categorized in the area of Reinforcement Learning (RL), and Nonlinear Modeling Techniques (NMT). For the first set of methods, simulation plays a fundamental role in evaluating the control policy: learning techniques are used to deliver sub-optimal policies at the end of a learning process. These iterative methods use the interaction of agents with the stochastic environment through taking actions and observing different states. To converge to the steady-state condition where policies and value functions do not change significantly with the continuation of the learning process, all or most important states must be visited sufficiently. This might be prohibitively time-consuming for large-scale problems. To make these techniques more efficient both in terms of computation time and robust optimal policies, the idea of Opposition-Based Learning (OBL-Type I and Type II) is employed to modify/extend popular RL techniques including Q-Learning, Q(λ), sarsa, and sarsa(λ). Several new algorithms are developed using this idea. It is also illustrated that, function approximation techniques such as neural networks can contribute to the process of learning. The state-of-the-art implementations usually consider the maximization of expected value of accumulated reward. Extending these techniques to consider risk and solving some well-known control problems are important contributions of this thesis. Furthermore, the new nonlinear modeling for reservoir management using indicator functions and randomized policy introduced by Fletcher and Ponnambalam, is extended to stochastic releases in multi-reservoir systems. In this extension, two different approaches for defining the release policies are proposed. In addition, the main restriction of considering the normal distribution for inflow is relaxed by using a beta-equivalent general distribution. A five-reservoir case study from India is used to demonstrate the benefits of these new developments. Using a warehouse management problem as an example, application of the proposed method to other storage management problems is outlined.

Saab Ericsson Space AB develops products for space for a predetermined price. Since the price is fixed, it is crucial to have a reliable prediction model to estimate the effort needed to develop the product. In general software effort estimation is difficult, and at the software department this is a problem. By analyzing metrics, collected from former projects, different prediction models are developed to estimate the number of person hours a software project will require. Models for predicting the effort before a project begins is first developed. Only a few variables are known at this state of a project. The models developed are compared to a current model used at the company. Linear regression models improve the estimate error with nine percent units and nonlinear regression models improve the result even more. The model used today is also calibrated to improve its predictions. A principal component regression model is developed as well. Also a model to improve the estimate during an ongoing project is developed. This is a new approach, and comparison with the first estimate is the only evaluation. The result is an improved prediction model. There are several models that perform better than the one used today. In the discussion, positive and negative aspects of the models are debated, leading to the choice of a model, recommended for future use.

This thesis investigates the effects of the earth's rotation on internal waves from two perspectives of nonlinear internal wave theory: near-resonant triads and weakly nonlinear models. We apply perturbation theory (multiple scale analysis) to the governing equations of internal waves and develop a near-resonant internal wave triad theory. This theory explains a resonant-like phenomenon in the numerical results obtained from simulating internal waves generated by tide topography interaction. Furthermore, we find that the inclusion of the earth's rotation (nonzero $f$) in the numerical runs leads to a very special type of resonance: parametric subharmonic instability. Through using perturbation expansion to solve separable solutions to the governing equations of internal waves, we derive a new rotation modified KdV equation (RMKdV). Of particular interest, the dispersion relation of the new equation obeys the exact dispersion relation for internal waves for both small and moderate wavenumbers ($k$). Thus this new RMKdV is able to model wea kly nonlinear internal waves with various wavenumbers ($k$), better than the Ostrovsky equation which fails at describing waves of small $k$.

In this thesis, modeling and optimization in the field of storage management under stochastic condition will be investigated using two different methodologies: Simulation Optimization Techniques (SOT), which are usually categorized in the area of Reinforcement Learning (RL), and Nonlinear Modeling Techniques (NMT). For the first set of methods, simulation plays a fundamental role in evaluating the control policy: learning techniques are used to deliver sub-optimal policies at the end of a learning process. These iterative methods use the interaction of agents with the stochastic environment through taking actions and observing different states. To converge to the steady-state condition where policies and value functions do not change significantly with the continuation of the learning process, all or most important states must be visited sufficiently. This might be prohibitively time-consuming for large-scale problems. To make these techniques more efficient both in terms of computation time and robust optimal policies, the idea of Opposition-Based Learning (OBL-Type I and Type II) is employed to modify/extend popular RL techniques including Q-Learning, Q(λ), sarsa, and sarsa(λ). Several new algorithms are developed using this idea. It is also illustrated that, function approximation techniques such as neural networks can contribute to the process of learning. The state-of-the-art implementations usually consider the maximization of expected value of accumulated reward. Extending these techniques to consider risk and solving some well-known control problems are important contributions of this thesis. Furthermore, the new nonlinear modeling for reservoir management using indicator functions and randomized policy introduced by Fletcher and Ponnambalam, is extended to stochastic releases in multi-reservoir systems. In this extension, two different approaches for defining the release policies are proposed. In addition, the main restriction of considering the normal distribution for inflow is relaxed by using a beta-equivalent general distribution. A five-reservoir case study from India is used to demonstrate the benefits of these new developments. Using a warehouse management problem as an example, application of the proposed method to other storage management problems is outlined.

Μελέτη μη γραμμικών μοντέλων παλινδρόμησης(λογιστικό, εκθετικό, Poisson,γενικευμένα γραμμικά μοντέλα) όσον αφορά διαστημα εμπιστοσύνης, έλεγχο υποθέσεων και καλή προσαρμογή. / Analysis of nonlinear regression models (logistic, exponential, Poisson, generalized linear models) regarding confidence interval estimation, tests and good fit.

Model-based control and process optimization technologies are becoming more commonly used by chemical engineers. These algorithms rely on fundamental or empirical models that are frequently described by systems of differential equations with unknown parameters. It is, therefore, very important for modellers of chemical engineering processes to have access to reliable and efficient tools for parameter estimation in dynamic models. The purpose of this thesis is to develop an efficient and easy-to-use parameter estimation algorithm that can address difficulties that frequently arise when estimating parameters in nonlinear continuous-time dynamic models of industrial processes. The proposed algorithm has desirable numerical stability properties that stem from using piece-wise polynomial discretization schemes to transform the model differential equations into a set of algebraic equations. Consequently, parameters can be estimated by solving a nonlinear programming problem without requiring repeated numerical integration of the differential equations. Possible modelling discrepancies and process disturbances are accounted for in the proposed algorithm, and estimates of the process disturbance intensities can be obtained along with estimates of model parameters and states. Theoretical approximate confidence interval expressions for the parameters are developed. Through a practical two-phase nylon reactor example, as well as several simulation studies using stirred tank reactors, it is shown that the proposed parameter estimation algorithm can address difficulties such as: different types of measured responses with different levels of measurement noise, measurements taken at irregularly-spaced sampling times, unknown initial conditions for some state variables, unmeasured state variables, and unknown disturbances that enter the process and influence its future behaviour. / Thesis (Ph.D, Chemical Engineering) -- Queen's University, 2008-06-20 16:34:44.586

Made available in DSpace on 2015-03-26T13:32:08Z (GMT). No. of bitstreams: 1 texto completo.pdf: 790188 bytes, checksum: 1cb94f6a1ebb6d10f561e9184b3f44f3 (MD5) Previous issue date: 2010-02-11 / The main objective of this work was used the cluster analysis in order to classify nonlinear growth models in relation to different quality fit evaluators when utilized data from the following beef lambs genetic groups: Dorper x Morada Nova (DMN), Dorper x Rabo Largo (DRL) e Dorper x Santa Inês (DSI). After the choice of the best model, we aimed also apply the model identity in order to identify the most efficient group. The proposed methodology was considered in two experimental conditions: with repetitions, using data of all animals from each group; and without repetitions, using average data from each group. Twelve nonlinear models were used, whose fit quality was measured by determination coefficient (R2 aj), Akaike information criterion (AIC), Bayesian information criterion (BIC), mean quadratic error of prediction (MEP) and predicted determination coefficient (R2 p). The Richards and von Bertalanffy models, respectively, presented the best fit for the mean and individual data sets. The model identity tests revealed that the DSI group presented higher adult weight, therefore this group is recommend for meat production. / O objetivo principal desse trabalho foi utilizar a análise de agrupamento para classificar modelos de crescimento não-lineares tendo em vista os resultados de diferentes avaliadores de qualidade de ajuste ao considerar dados dos seguintes grupos genéticos de ovinos de corte: Dorper x Morada Nova (DMN), Dorper x Rabo Largo (DRL) e Dorper x Santa Inês (DSI). Após a indicação do modelo comum adequado aos três grupos, objetivou-se também aplicar a identidade de modelos com o intuito de identificar o grupo genético com maior eficiência de crescimento. Toda a metodologia foi aplicada a duas situações experimentais distintas: com repetição, considerando todos os animais de cada grupo genético, e sem repetição, considerando dados médios de cada um destes grupos. Ajustaram-se doze modelos não-lineares, cuja qualidade de ajuste foi medida pelo coeficiente de determinação ajustado (R2 aj), critério de informação de Akaike (AIC), critério de informação Bayesiano (BIC), erro quadrático médio de predição (MEP) e coeficiente de determinação de predição (R2 p). Os modelos Richards e von Bertalanffy foram, respectivamente, os que apresentaram os melhores ajustes para os conjuntos de dados médios e individuais. De acordo com testes de identidade de modelos, o grupo genético DSI foi o que apresentou maior peso adulto, sendo este, portanto, o mais recomendado para exploração de carne.

[pt] Este trabalho tem como objetivo principal adaptar o modelo STR-Tree, o qual é a combinação de um modelo Smooth Transition Regression com Classification and Regression Tree (CART), a fim de utilizá-lo em Classificação. Para isto algumas alterações foram realizadas em sua forma estrutural e na estimação. Devido ao fato de estarmos fazendo classificação de variáveis dependentes binárias, se faz necessária a utilização das técnicas empregadas em Regressão Logística, dessa forma a estimação dos parâmetros da parte linear passa a ser feita por Máxima Verossimilhança. Assim o modelo, que é paramétrico não-linear e estruturado por árvore de decisão, onde cada nó terminal representa um regime os quais têm seus parâmetros estimados da mesma forma que em uma Regressão Logística, é denominado Smooth Transition Logistic Regression-Tree (STLR-Tree). A inclusão dos regimes, determinada pela divisão dos nós da árvore, é feita baseada em testes do tipo Multiplicadores de Lagrange, que em sua forma para o caso Gaussiano utiliza a Soma dos Quadrados dos Resíduos em suas estatísticas de teste, aqui são substituídas pela Função Desvio (Deviance), que é equivalente para o caso dos modelos não Gaussianos, cuja distribuição da variável dependente pertença à família exponencial. Na aplicação a dados reais selecionou-se dois conjuntos das variáveis explicativas de cada uma das duas bases utilizadas, que resultaram nas melhores taxas de acerto, verificadas através de Tabelas de Classificação (Matrizes de Confusão). Esses conjuntos de variáveis foram usados com outros métodos de classificação existentes, são eles: Generalized Additive Models (GAM), Regressão Logística, Redes Neurais, Análise Discriminante, k-Nearest Neighbor (K-NN) e Classification and Regression Trees (CART). / [en] The main goal of this work is to adapt the STR-Tree model, which is the combination of a Smooth Transition with Regression model with Classi cation and Regression Tree (CART), in order to use it in Classification. Some changes were made in its structural form and in the estimation. Due to the fact we are doing binary dependent variables classification, is necessary to use the techniques employed in Logistic Regression, so the estimation of the linear part will be made by Maximum Likelihood. Thus the model, which is nonlinear parametric and structured by a decision tree, where each terminal node represents a regime that have their parameters estimated in the same way as in a Logistic Regression, is called Smooth Transition Logistic Regression Tree (STLR-Tree). The inclusion of the regimes, determined by the splitting of the tree's nodes, is based on Lagrange Multipliers tests, which for the Gaussian cases uses the Residual Sum-of-squares in their test statistic, here are replaced by the Deviance function, which is equivalent to the case of non-Gaussian models, that has the distribution of the dependent variable in the exponential family. After applying the model in two datasets chosen from the bibliography comparing with other methods of classi cation such as: Generalized Additive Models (GAM), Logistic Regression, Neural Networks, Discriminant Analyses, k-Nearest Neighbor (k-NN) and Classification and Regression Trees (CART). It can be seen, verifying in the Classification Tables (Confusion Matrices) that STLR-Tree showed the second best result for the overall rate of correct classification in three of the four applications shown, being in all of them, behind only from GAM.