Dans ce mémoire, une nouvelle approche neuronale TLS EXIN est proposée pour estimer la position d'un récepteurGPS. L'idée générale de cette approche est d'avoir une méthode plus robuste pour le calcul de la position.Le pseudorange alone est l'une des techniques les plus simples et les plus utilisées pour le positionnement GPS.Cette technique nécessite la résolution d'un système surdéterminé d'équations linéaires Ax+/-b. En général, lesmoindres carrés ordinaires (OLS) et les moindres carrés pondérés (WLS) sont les méthodes couramment utiliséespour estimer la position d'un récepteur grâce à leur rapidité et leur robustesse, mais la structure particulière de lamatrice de données A et les bruits affectant ses entrées ne sont pas considérés. Au contraire, cette thèse a pourobjectif d'analyser ces problèmes et d'étudier le comportement des méthodes des moindres carrés (LS) enprésence d'une matrice de données A bruitée.L'approche des moindres carrés totaux (TLS) prend en compte le bruit dans la matrice de données A ainsi que dansle vecteur d'observation b. Cette dernière est une technique moins robuste que OLS et plus sensible auchangement des données, elle est en général résolue par une méthode directe. La méthode TLS EXIN basée surles réseaux de neurones est un algorithme itératif (flux gradient) pour résoudre le problème TLS. Elle donne unmeilleur résultat parce qu'elle peut exploiter les informations d'état initial provenant des époques précédentes et, encas de conditions initiales nulles, donne une estimation précise même en cas de problème singulier.Pour réaliser des comparaisons entre les différentes méthodes des moindres carrés (LS), deux jeux de données ontété collectés. Le premier jeu de données est issu du réseau TERIA et comporte des données collectées depuisdifférentes stations de référence situées dans toute la France. Le deuxième jeu de données est le résultat d'unecampagne de mesures utilisant un appareil GPS (Ublox NL-6002U).Grâce à ces données réelles, un nombre de conditionnement bas a été estimé. Dans ce cas, toutes les méthodesLS donnent des estimations équivalentes, et le choix du meilleur algorithme (OLS, et surtout, WLS) est privilégiépour leur rapidité de calcul. Cependant, le pire scénario qui puisse se produire a été étudié (dans le cas d'unsatellite éloigné), et ont été observés des mauvais conditionnements du problème de GPS (nombre deconditionnement élevé). Cette situation extrême justifie l'utilisation du réseau neuronal TLS EXIN. Les résultatsobtenus confirment cette approche, même pour un nombre de conditionnement élevé.____________________________________________________________________________________________ / In this thesis, the neural approach TLS EXIN was proposed in a new way in order to estimate the position of a GPSreceiver. The general idea of this approach is to develop a more robust method for calculating the position.The ¿pseudorange alone¿ method is one of the simplest techniques and most widely used for estimating the GPSpositioning and it results in solving an overdetermined system of linear equations Ax¿b. In general, the ordinaryleast squares (OLS) and weighted least squares (WLS) are the commonly used methods to estimate the position ofa receiver for their quickness and robustness, but the particular structure of the data matrix A and the noise affectingits entries are not considered. This thesis, instead, aims to address these problems and study the behavior of leastsquares (LS) methods in the presence of a noisy data matrix A.The approach of total least squares (TLS) takes into account the noise in A and b. It is a less robust technique thanOLS and more sensitive to data changes. It is general solved as a direct method. Instead, the neural network TLSEXIN, which is an iterative algorithm (gradient flow) for solving TLS, works better both because it may exploit theinitial condition information derived from the previous epochs, and, in case of null initial conditions, yields anaccurate estimate even in case of close-to-degenerate data matrix.To perform tests between different methods of least squares, two sets of data were collected. The first one comesfrom the TERIA network and includes data collected from different reference stations located throughout France.While the second one is the result of a measurement campaign using GPS (Ublox NL-6002U).Using the real data, a low condition number has been estimated: in this case all LS methods yield equivalentestimates: as a consequence, OLS and WLS are to be preferred for their low computational cost. However, theworst scenario has been investigated which may occur in case of a distant satellite, resulting in the ill-conditioning ofthe GPS problem. This extreme situation justifies the use of the TLS EXIN neural network. The results obtainedconfirm of this approach even for very large condition numbers.

Bariatric surgery, also known as weight loss surgery (WLS) is an intervention for individuals who are suffering from obesity and weight-related health complications which often accompany being 100 pounds or more overweight (Smith et al., 2011). Even though bariatric surgery has been shown to be a life-saving and life-enhancing operation, efforts to seek out surgery options, qualify and prepare for the procedure, recover from surgery, and then adapt new routines to support surgery are challenging not only for the patient, but also for the people with whom the patient spends the most time—their romantic partners (Applegate & Friedman, 2008; Bylund, Benzein, & Carina, 2013; Moore & Cooper, 2016; Sarwer, Dilks, & West-Smith, 2011). This dissertation was a means to explore relational, food, and WLS success experiences that take place within couples when at least one person has had bariatric surgery. Comparisons between life before, during, and after WLS were discussed with focused attention given to relationship dynamics and daily food routines (Bocchieri, Meana, & Fisher, 2002). The couple’s definition of WLS success and the means by which they have been successful were launching points for more in-depth conversation. Data was generated through 2 interviews per couple (n = 11) with patients who met selection criteria for the study—committed long-term relationship and of the same residence for at least the past 5 years with at least 1 person having been successful with WLS. Success was defined by the WLS patient, however they had to be at least 2 years post-WLS, the critical time period where postoperative weight regain has been shown to occur, in order to participate (Magro et al., 2008; Ogden, Avenell, & Ellis, 2011; Pories et al., 2016). A thematic analysis with multiple rounds of coding was conducted after data saturation was met and couples indicated their agreeableness with results through a short, follow-up survey which also functioned as a form of member-checking. Overall, couples’ relationship dynamics were characterized as secure and WLS gave them another way to give support, engage in teamwork, and ultimately become closer. Patients and spouses explained that they loved each other unconditionally, no matter what the patient weighed, and this had been the reality for their entire relationship. Thus, security was the theme for relationship dynamics with support, teamwork, and closer as subthemes. Spouses expressed their desire to help the WLS patient when it came to being open to change and then making necessary modifications in habits and lifestyle. The commitment to change happened before WLS and a mind-shift happened after WLS that enabled both people to adjust their thinking, consistently evaluate their routines, and continue to change their behaviors. As a result, commit and mind-shift were the themes and subsequent changes (diet, exercise, and mindsets) were the subthemes. A secure relationship and commitment to making “better choices” assisted the patient in experiencing WLS success and this meant that their spouse experienced success, too; “it’s our success together.” The theme for WLS success was follow-through and subthemes were results, comfort, happy, and freedom. Hope was also a by-product of success and it was the grand-theme of this study.

SEB Merchant Banking provides to its institutional customers a true market neutral product called Dynamic Manager Alpha (DMA). The DMA is constructed by a long position in an exceptionally well performing mutual fund and a beta adjusted short position in an appropriate index. The key to making the product market neutral is adjusting with the correct beta, since the beta changes, it is very important to have a good model for predicting beta in the future. This master thesis begins with describing what beta is in a CAPM sense. It then continues with recognizing the so called “Two Beta Trap”, which separates two kinds of beta. The first is in CAPM sense, with a market portfolio represented by the whole market. The second is a “best fit” beta where the market portfolio is the index which explains as much as possible of the fund returns. It is this second way of calculating beta that is used in this thesis and therefore beta can be viewed upon like a hedgeratio. The purpose of this thesis is to predict the future beta for mutual funds with as high accuracy as possible. The starting point has been historic OLS (Ordinary Least Squares) estimation of beta. From earlier studies and own studies in this thesis a lot of different techniques for predicting beta has been tested. For example the eriodicity in the returns, the interval length, different regression methods as LAD (Least Absolute Deviation) and IRLS (Iteratively Reweighted Least Squares). Also different adjustments to beta have been tested for better catching the momentum in beta and general mean reverting tendencies. The results of the studies show that when possible, returns calculated with daily compounding is not favorable. For daily but especially weekly returns, LAD and IRLS are superior to OLS in predicting beta. Adjusting techniques have a positive effect in predicting beta, especially for weekly returns. Monthly returns seem to be most stable and have the smallest prediction errors, but with the right model and adjustment, betas with weekly returns have almost as good characteristics. Since the prediction model needs to have a fast response to market changes, returns calculated with short compounding is favorable. It is therefore very encouraging that the results from this thesis have showed great improvements in prediction of beta for returns calculated with weekly compounding.

<p>SEB Merchant Banking provides to its institutional customers a true market neutral product called Dynamic Manager Alpha (DMA). The DMA is constructed by a long position in an exceptionally well performing mutual fund and a beta adjusted short position in an appropriate index. The key to making the product market neutral is adjusting with the correct beta, since the beta changes, it is very important to have a good model for predicting beta in the future.</p><p>This master thesis begins with describing what beta is in a CAPM sense. It then continues with recognizing the so called “Two Beta Trap”, which separates two kinds of beta. The first is in CAPM sense, with a market portfolio represented by the whole market. The second is a “best fit” beta where the market portfolio is the index which explains as much as possible of the fund returns. It is this second way of calculating beta that is used in this thesis and therefore beta can be viewed upon like a hedgeratio.</p><p>The purpose of this thesis is to predict the future beta for mutual funds with as high accuracy as possible. The starting point has been historic OLS (Ordinary Least Squares) estimation of beta. From earlier studies and own studies in this thesis a lot of different techniques for predicting beta has been tested. For example the eriodicity in the returns, the interval length, different regression methods as LAD (Least Absolute Deviation) and IRLS (Iteratively Reweighted Least Squares). Also different adjustments to beta have been tested for better catching the momentum in beta and general mean reverting tendencies.</p><p>The results of the studies show that when possible, returns calculated with daily compounding is not favorable. For daily but especially weekly returns, LAD and IRLS are superior to OLS in predicting beta. Adjusting techniques have a positive effect in predicting beta, especially for weekly returns. Monthly returns seem to be most stable and have the smallest prediction errors, but with the right model and adjustment, betas with weekly returns have almost as good characteristics. Since the prediction model needs to have a fast response to market changes, returns calculated with short compounding is favorable. It is therefore very encouraging that the results from this thesis have showed great improvements in prediction of beta for returns calculated with weekly compounding.</p>

The modeling of freight travel demand has gained increasing attention in the recent past due to the importance of efficient and safe freight transportation to regional economic growth. Despite the attention paid to the modeling of freight travel demand, advances in modeling methods and the development of practical tools for forecasting freight flows have been limited. The development of freight demand models that incorporate the behavioral aspects of freight demand face significant hurdles, partially due to the data requirements, which are a consequence of the inherent complexity of the mechanisms driving freight demand. This research attempts to make a contribution in this context by proposing a relatively data simple, but behaviorally robust statewide modeling framework for the state of Florida, in the spirit of an aggregate level four-step planning process. The modeling framework that is developed in this research can be applied to the modeling of freight travel demand using data contained in readily available commercial databases such as the Reebie TRANSEARCH database and the InfoUSA employer database. The modeling methodology consists of a structural equations modeling framework that can accommodate multiple dependent variables simultaneously. This framework predicts freight flows on various modes between two zipcodes based on the socio-economic characteristics and the modal level of service characteristics. Separate models have been developed for various commodity groups. The estimated models for various commodity groups are found to offer statistically valid indications and plausible interpretations suggesting that these models may be suitable for application in freight transportation demand forecasting applications. The sensitivity analysis conducted on these models clearly added evidence to the fact that employment is the key factor influencing freight flows between two regions.

[pt] Este trabalho lida com o problema de radiolocalização de terminais em um ambiente de comunicações móveis celulares. Desenvolve-se novas alternativas para a estimação da posição, admitindo-se que as medidas de tempo de chegada (ToA) obtidas no enlace-rádio estão corrompidas por ruído aditivo e apresentam erro médio positivo durante períodos aleatórios, devido à ausência de linha de visada (NLOS) entre terminal e estações radio-bases. Em uma das alternativas desenvolve-se um estimador assintoticamente eficiente do erro de NLOS, sob o critério de mínimos quadrados ponderados (WLS). Para esta estimativa, admite- se o conhecimento a priori do espalhamento temporal do canal, e que o perfil de potência do sinal pode ser calculado por uma média temporal de medidas independentes em um receptor RAKE. O esquema de localização apresentado incorpora também um teste de hipóteses desenvolvido sob o critério de Neyman-Pearson, para detectar, a cada instante de tempo, a ocorrência de transições entre os estados LOS/NLOS do canal. Em outra contribuição do trabalho, as coordenadas do terminal são estimadas recursivamente utilizando-se algoritmos bayesianos, com a dimensão do espaço de estados aumentada para incluir o efeito do erro de NLOS sobre as medidas de ToA. Resultados de simulação obtidos sob diferentes cenários comprovam a eficiência dos esquemas de estimação aqui desenvolvidos, quando comparados à única solução de que se tem conhecimento na literatura. Apresenta-se ainda nesta tese uma análise para o problema de ambigüidade em métodos hiperbólicos de localização, cujo objetivo é identificar a região do plano em que este método fornece duas soluções fisicamente admissíveis. A área desta região é comparada com a área total de triangulação. / [en] This work addresses the radiolocation problem of a moving terminal in a cellular mobile communications environment. New alternatives are developed for position estimation, assuming that the Time of Arrival (ToA) measurements obtained from radio link are corrupted by additive noise and have positive mean error during random periods of time due to the non-line of sight (NLOS) propagation condition between the terminal and base stations. In one of the proposals, an asymptotically efficient WLS estimator of the NLOS error is developed under the Weighted Least Squares criterion. It is assumed that the channel temporal scattering model is known and the mean power delay profile can be evaluated by time averaging independent measurements from a RAKE receiver. The location estimation scheme also includes a hypothesis testing based on Neyman- Pearson approach to detect at each instant of time the LOS/NLOS states transitions. In another contribution, the terminal coordinates are recursively estimated using bayesian algorithms, with the state-space dimension augmented to include the NLOS error effect over ToA measurements. Simulation results obtained under different scenarios show the effectiveness of the estimation schemes developed here when compared to the only alternative known from the literature. An analysis concerning the ambiguity problem in hyperbolic location methods is also presented, aiming to determine the regions where this method gives two physically admissible solutions, and compare them to the total trilateration area.

Underwater communication within underwater sensor network is crucial for surveillance of coast and ocean areas. The aim of this report was to examine whether it is realistic to use the communication signal which is sent from one node to another as a sonar pulse, and in such case at what distances. To examine the problem, a system consisting of two nodes and one approaching target was simulated in Matlab. At first, the system tries to detect the target by using a generalized likelihood ratio test, which calculates the probability of a present target from the surrounding sounds. When a target is detected by a node, it estimates the bearing to the target by using beamforming and sends out a communication signal to the other node. The communication signal spreads out in the water, and bounces on the target before it is received by the second node. To calculate the distance, the second node decodes the signal to get the time difference, from which the distance is calculated. In the end, the target's position was estimated with a weighted least square estimator with measurements of the bearing and distance. The result shows that the distance to the target could be estimated with high precision in the given scenario, and that the width of the Cramér-Rao lower bound depends mainly on the variance of the beamforming algorithm. The maximum distance reached up to two kilometers but was mainly restricted by the detection algorithm. In conclusion, the result shows that the communication pulse can be used as a sonar pulse at the tested distances. However, the simulated scenario is a simplified version of the real world so more testing should be performed before a final conclusion can be made. / För övervakning av kust- och havsområden, vid exempelvis militära operationer eller för oceanografska observationer, används ofta ett undervattenssystem som är uppbyggt av flera noder som finns utplacerade på botten. Noderna lyssnar efter mål såsom ubåtar, fartyg etc, med syftet att kunna detektera och lokalisera dessa. Om en nod lyckas detektera ett mål så skickar den ut en akustisk kommunikationssignal till övriga noder i systemet. Målet med detta examensarbete var att undersöka om den kommunikationssignal som skickas mellan noderna också kan användas som en sonarpuls för att bestämma avståndet till målet, och därmed förbättra lokaliseringen av målets position. Under antagandet att kommunikationssignalen kan användas som sonarpuls, så undersöktes dessutom vid vilka avstånd mellan noden och målet som det var möjligt att använda signalen som sonarpuls. Resultatet visar att det är möjligt att använda kommunikationssignalen som en sonarpuls. Bäst funkar det på nära avstånd, då är den estimerade positionen i stort sett lika med det riktiga positionen. I takt med att avståndet till målet ökar så ökar även osäkerheten i vilken rikting målet befinner sig, estimeringen av avståndet höll sig däremot väldigt nära den faktiska distansen i alla simuleringar som gjordes. Simuleringen som gjordes var dock en förenkling av verkligheten, och flera av de störningsmoment som finns ute i naturen har inte tagits med i beräkningarna. För att undersöka detta så simulerades ett sensorsystem bestående av två noder tillsammans med ett mål som närmade sig noderna. Noderna försöker detektera målet genom att lyssna efter ljud som tillhör målet. Genom att mäta energinivåer i de ljudsignaler som noderna hör, så kan man utifrån sannolikhetslära bestämma hur troligt det är att det finns ett mål i närheten. När sannolikheten är tillräckligt hög säger man att ett mål detekterats. För att bestämma positionen så uppskattades målets riktning och avstånd i förhållande till noderna, som i sin tur användes för att beräkna målets position.

Balance between energy consumption and production of electricityis a very important for the electric power system operation and planning. Itprovides a good principle of effective operation, reduces the generation costin a power system and saves money. Two novel approaches to reduce thetotal errors between forecast and real electricity consumption wereproposed. An Adaptive Linear Regression of Original Predictions (ALROP)was constructed to modify the existing predictions by using simple linearregression with estimation by the Ordinary Least Square (OLS) method.The Weighted Least Square (WLS) method was also used as an alternativeto OLS. The Modeling of Prediction Errors (MPE) was constructed in orderto predict errors for the existing predictions by using the Autoregression(AR) and the Autoregressive-Moving-Average (ARMA) models. For thefirst approach it is observed that the last reported value is of mainimportance. An attempt was made to improve the performance and to getbetter parameter estimates. The separation of concerns and the combinationof concerns were suggested in order to extend the constructed approachesand raise the efficacy of them. Both methods were tested on data for thefourth region of Sweden (“elområde 4”) provided by Bixia. The obtainedresults indicate that all suggested approaches reduce the total percentageerrors of prediction consumption approximately by one half. Resultsindicate that use of the ARMA model slightly better reduces the total errorsthan the other suggested approaches. The most effective way to reduce thetotal consumption prediction errors seems to be obtained by reducing thetotal errors for each subregion.

This thesis is, to my knowledge, the first study utilising data from the Swedish population and housing censuses between 1960 and 1990 merged with other data from the same period in order to estimate extensive margin labour supply responses to changes in municipal tax rate changes. Given that women historically have not faced the same structural labour market preconditions as men, the empirical strategy is designed to allow for an analysis of gender-based heterogeneity in labour supply responses. Using a weighted fixed effects framework, estimates of the average over time between municipal effects of tax rate increases are presented. Using the preferred main model specification, the estimate for the average tax rate elasticity is -0.165 for men and 0.3513 for women. Additionally, an attempt is made to estimate an effect using a difference-in-difference framework, treating the overall largest municipal tax rate changes as a form of quasi-experimental treatment. The results of the main analysis indicate the presence of gender-based heterogeneity in extensive margin labour supply responses during 1960 to 1990 within the administrative region in question.

Esta dissertação trata do problema de processamento de Erros Grosseiros (EGs) com base na aplicação do chamado Índice de Não-Detecção de Erros, ou apenas UI (Undetectability Index), na análise dos resíduos do estimador de estado por mínimos quadrados ponderados. O índice UI foi desenvolvido recentemente e possibilita a classificação das medidas de acordo com as suas características de não refletirem grande parcela de seus erros nos resíduos daquele estimador. As medidas com maiores UIs são aquelas cujos erros são mais difíceis de serem detectados através de métodos que fazem uso da análise dos resíduos, pois grande parcela do erro dessas medidas não aparece no resíduo. Inicialmente demonstrou-se, nesta dissertação, que erros das estimativas das variáveis de estado em um sistema com EG não-detectável (em uma medida de alto índice UI) podem ser mais significativos que em medidas com EGs detectáveis (em medidas com índices UIs baixos). Justificando, dessa forma, a importância de estudos para tornar possível o processamento de EGs em medidas com alto índice UI. Realizou-se, então, nesta dissertação, diversas simulações computacionais buscando analisar a influência de diferentes ponderações de medidas no UI e também nos erros das estimativas das variáveis de estado. Encontrou-se, então, uma maneira que destacou-se como a mais adequada para ponderação das medidas. Por fim, ampliaram-se, nesta dissertação, as pesquisas referentes ao UI para um estimador de estado por mínimos quadrados ponderados híbrido. / This dissertation deals with the problem of Gross Errors processing based on the use of the so-called Undetectability Index, or just UI. This index was developed recently and it is capable to classify the measurements according to their characteristics of not reflecting their errors into the residuals of the weighted least squares state estimation process. Gross errors in measurements with higher UIs are very difficult to be detected by methods based on the residual analysis, as the errors in those measurements are masked, i.e., they are not reflected in the residuals. Initially, this dissertation demonstrates that a non-detectable gross error (error in a measurement with high UI) may affect more the accuracy of the estimated state variables than a detectable gross error (error in a measurement with low UI). Therefore, justifying the importance of studies that make possible gross errors processing in measurements with high UI. In this dissertation, several computational simulations are carried out to analyze the influence of different weights of measurements in the UI index and also in the accuracy of the estimated state variables. It is chosen a way that stood out as the most appropriate for weighing the measurements. Finally, in this dissertation, the studies referring to the UI is extended for a hybrid weighted least squares state estimator.