No description available.

No description available.

Our work is driven by a class of practical problems of sequential decision making in the context of electric power generation under uncertainties. These problems are usually treated as receding horizon deterministic optimization problems, and/or as scenario-based stochastic programs. Stochastic programming allows to compute a first stage decision that is hedged against the possible futures and -- if a possibility of recourse exists -- this decision can then be particularized to possible future scenarios thanks to the information gathered until the recourse opportunity. Although many decomposition techniques exist, stochastic programming is currently not tractable in the context of day-ahead electric power generation and furthermore does not provide an explicit recourse strategy. The latter observation also makes this approach cumbersome when one wants to evaluate its value on independent scenarios. We propose a supervised learning methodology to learn an explicit recourse strategy for a given generation schedule, from optimal adjustments of the system under simulated perturbed conditions. This methodology may thus be complementary to a stochastic programming based approach. With respect to a receding horizon optimization, it has the advantages of transferring the heavy computation offline, while providing the ability to quickly infer decisions during online exploitation of the generation system. Furthermore the learned strategy can be validated offline on an independent set of scenarios. On a realistic instance of the intra-day electricity generation rescheduling problem, we explain how to generate disturbance scenarios, how to compute adjusted schedules, how to formulate the supervised learning problem to obtain a recourse strategy, how to restore feasibility of the predicted adjustments and how to evaluate the recourse strategy on independent scenarios. We analyze different settings, namely either to predict the detailed adjustment of all the generation units, or to predict more qualitative variables that allow to speed up the adjustment computation procedure by facilitating the ``classical' optimization problem. Our approach is intrinsically scalable to large-scale generation management problems, and may in principle handle all kinds of uncertainties and practical constraints. Our results show the feasibility of the approach and are also promising in terms of economic efficiency of the resulting strategies. The solutions of the optimization problem of generation (re)scheduling must satisfy many constraints. However, a classical learning algorithm that is (by nature) unaware of the constraints the data is subject to may indeed successfully capture the sensitivity of the solution to the model parameters. This has nevertheless raised our attention on one particular aspect of the relation between machine learning algorithms and optimization algorithms. When we apply a supervised learning algorithm to search in a hypothesis space based on data that satisfies a known set of constraints, can we guarantee that the hypothesis that we select will make predictions that satisfy the constraints? Can we at least benefit from our knowledge of the constraints to eliminate some hypotheses while learning and thus hope that the selected hypothesis has a better generalization error? In the second part of this thesis, where we try to answer these questions, we propose a generic extension of tree-based ensemble methods that allows incorporating incomplete data but also prior knowledge about the problem. The framework is based on a convex optimization problem allowing to regularize a tree-based ensemble model by adjusting either (or both) the labels attached to the leaves of an ensemble of regression trees or the outputs of the observations of the training sample. It allows to incorporate weak additional information in the form of partial information about output labels (like in censored data or semi-supervised learning) or -- more generally -- to cope with observations of varying degree of precision, or strong priors in the form of structural knowledge about the sought model. In addition to enhancing the precision by exploiting information that cannot be used by classical supervised learning algorithms, the proposed approach may be used to produce models which naturally comply with feasibility constraints that must be satisfied in many practical decision making problems, especially in contexts where the output space is of high-dimension and/or structured by invariances, symmetries and other kinds of constraints.

Understanding the relationships between biodiversity and climate is essential for predicting the impact of climate change on broad-scale landscape processes. Utilizing indirect indicators of biodiversity derived from remotely sensed imagery, we present an approach to forecast shifts in the spatial distribution of biodiversity. Indirect indicators, such as remotely sensed plant productivity metrics, representing landscape seasonality, minimum growth, and total greenness have been linked to species richness over broad spatial scales, providing unique capacity for biodiversity modeling. Our goal is to map future spatial distributions of plant productivity metrics based on expected climate change and to quantify anticipated change to park habitat in British Columbia. Using an archival dataset sourced from the Advanced Very High Resolution Radiometer (AVHRR) satellite from the years 1987 to 2007 at 1km spatial resolution, corresponding historical climate data, and regression tree modeling, we developed regional models of the relationships between climate and annual productivity growth. Historical interconnections between climate and annual productivity were coupled with three climate change scenarios modeled by the Canadian Centre for Climate Modeling and Analysis (CCCma) to predict and map productivity components to the year 2065. Results indicate we can expect a warmer and wetter environment, which may lead to increased productivity in the north and higher elevations. Overall, seasonality is expected to decrease and greenness productivity metrics are expected to increase. The Coastal Mountains and high elevation edge habitats across British Columbia are forecasted to experience the greatest amount of change. In the future, protected areas may have potential higher greenness and lower seasonality as represented by indirect biodiversity indicators. The predictive model highlights potential gaps in protection along the central interior and Rocky Mountains. Protected areas are expected to experience the greatest change with indirect indicators located along mountainous elevations of British Columbia. Our indirect indicator approach to predict change in biodiversity provides resource managers with information to mitigate and adapt to future habitat dynamics. Spatially specific recommendations from our dataset provide information necessary for management. For instance, knowing there is a projected depletion of habitat representation in the East Rocky Mountains, sensitive species in the threatened Mountain Hemlock ecozone, or preservation of rare habitats in the decreasing greenness of the southern interior region is essential information for managers tasked with long term biodiversity conservation. Forecasting productivity levels, linked to the distribution of species richness, presents a novel approach for understanding the future implications of climate change on broad scale biodiversity. / Graduate

[pt] Esta dissertação tem como objetivo principal introduzir uma formulação de modelo não-linear multivariado, a qual combina o modelo STVAR (Smooth Transition Vector Autoregressive) com a metodologia CART (Classification and Regression Tree) a fim de utilizá-lo para geração de cenários e de previsões. O modelo resultante é um Modelo Vetorial Auto-Regressivo com Transição Suave Estruturado por Árvores, denominado STVAR-Tree e tem como base o conceito de múltiplos regimes, definidos por árvore binária. A especificação do modelo é feita através do teste LM. Desta forma, o crescimento da árvore é condicionado à existência de não-linearidade nas séries, que aponta a divisão do nó e a variável de transição correspondente. Em cada divisão, são estimados os parâmetros lineares, por Mínimos Quadrados Multivariados, e os parâmetros não-lineares, por Mínimos Quadrados Não-Lineares. Como forma de avaliação do modelo STVARTree, foram realizados diversos experimentos de Monte Carlo com o objetivo de constatar a funcionalidade tanto do teste LM quanto da estimação do modelo. Bons resultados foram obtidos para amostras médias e grandes. Além dos experimentos, o modelo STVAR-Tree foi aplicado às séries brasileiras de Vazão de Rios e Preço Spot de energia elétrica. No primeiro estudo, o modelo foi comparado estatisticamente com o Periodic Autoregressive (PAR) e apresentou um desempenho muito superior ao concorrente. No segundo caso, a comparação foi com a modelagem Neuro-Fuzzy e ganhou em uma das quatro séries. Somando os resultados dos experimentos e das duas aplicações conclui-se que o modelo STVAR-Tree pode ser utilizado na solução de problemas reais, apresentando bom desempenho. / [en] The main goal of the dissertation is to introduce a nonlinear multivariate model, which combines the model STVAR (Smooth Transition Vector Autoregressive) with the CART (Classification and Regression Tree) method and use it for generating scenarios and forecasting. The resulting model is a Tree- Structured Vector Autoregressive model with Smooth Transition, called STVARTree, which is based on the concept of multiple regimes, defined by binary tree. The model specification is based on Lagrange Multiplier tests. Thus, the growth of the tree is conditioned on the existence of nonlinearity in the time series, which indicates the node to be split and the corresponding transition variable. In each division, linear parameters are estimated by Multivariate Least Squares, and nonlinear parameters by Non-Linear Least Squares. As a way of checking the STVAR-Tree model, several Monte Carlo experiments were performed in order to see the functionality of both the LM test and the model estimation. Best results were obtained with medium and large samples. Besides, the STVAR-Tree model was applied to Brazilian time series of Rivers Flow and electricity spot price. In the first study, the model was statistically compared to the Periodic Autoregressive (PAR) model and had a much higher performance than the competitor. In the second case, the model comparison was with Neural-Fuzzy Modeling and the STVAR-Tree model won in one of the four series. Adding both the experiments and the two applications results we conclude that the STVARTree model may be applied to solve real problems, having good results.

Under the Basel accords for measuring regulatory capital requirements, the set of credit risk parameters probability of default (PD), exposure at default (EAD) and loss given default (LGD) are measured with own estimates by the internal rating based approach. The estimated parameters are also the foundation of understanding the actual risk in a banks credit portfolio. The predictive performance of such models are therefore interesting to examine. The credit risk parameter LGD has been seen to give low performance for predictive models and LGD values are generally hard to estimate. The main purpose of this thesis is to analyse the predictive performance of a multi-gene genetic programming approach to symbolic regression compared to three benchmark regression models. The goal of multi-gene symbolic regression is to estimate the underlying relationship in the data through a linear combination of a set of generated mathematical expressions. The benchmark models are Logit Transformed Regression, Beta Regression and Regression Tree. All benchmark models are frequently used in the area. The data used to compare the models is a set of randomly selected, de-identified loans from the portfolios of underlying U.S. residential mortgage-backed securities retrieved from International Finance Research. The conclusion from implementing and comparing the models is that, the credit risk parameter LGD is continued difficult to estimated, the symbolic regression approach did not yield a better predictive ability than the benchmark models and it did not seem to find the underlying relationship in the data. The benchmark models are more user-friendly with easier implementation and they all requires less calculation complexity than symbolic regression.

Reduction of crash occurrence on the various roadway locations (mid-block segments; signalized intersections; un-signalized intersections) and the mitigation of injury severity in the event of a crash are the major concerns of transportation safety engineers. Multi lane arterial roadways (excluding freeways and expressways) account for forty-three percent of fatal crashes in the state of Florida. Significant contributing causes fall under the broad categories of aggressive driver behavior; adverse weather and environmental conditions; and roadway geometric and traffic factors. The objective of this research was the implementation of innovative, state-of-the-art analytical methods to identify the contributing factors for crashes and injury severity. Advances in computational methods render the use of modern statistical and machine learning algorithms. Even though most of the contributing factors are known a-priori, advanced methods unearth changing trends. Heuristic evolutionary processes such as genetic programming; sophisticated data mining methods like conditional inference tree; and mathematical treatments in the form of sensitivity analyses outline the major contributions in this research. Application of traditional statistical methods like simultaneous ordered probit models, identification and resolution of crash data problems are also key aspects of this study. In order to eliminate the use of unrealistic uniform intersection influence radius of 250 ft, heuristic rules were developed for assigning crashes to roadway segments, signalized intersection and access points using parameters, such as 'site location', 'traffic control' and node information. Use of Conditional Inference Forest instead of Classification and Regression Tree to identify variables of significance for injury severity analysis removed the bias towards the selection of continuous variable or variables with large number of categories. For the injury severity analysis of crashes on highways, the corridors were clustered into four optimum groups. The optimum number of clusters was found using Partitioning around Medoids algorithm. Concepts of evolutionary biology like crossover and mutation were implemented to develop models for classification and regression analyses based on the highest hit rate and minimum error rate, respectively. Low crossover rate and higher mutation reduces the chances of genetic drift and brings in novelty to the model development process. Annual daily traffic; friction coefficient of pavements; on-street parking; curbed medians; surface and shoulder widths; alcohol / drug usage are some of the significant factors that played a role in both crash occurrence and injury severities. Relative sensitivity analyses were used to identify the effect of continuous variables on the variation of crash counts. This study improved the understanding of the significant factors that could play an important role in designing better safety countermeasures on multi lane highways, and hence enhance their safety by reducing the frequency of crashes and severity of injuries. Educating young people about the abuses of alcohol and drugs specifically at high schools and colleges could potentially lead to lower driver aggression. Removal of on-street parking from high speed arterials unilaterally could result in likely drop in the number of crashes. Widening of shoulders could give greater maneuvering space for the drivers. Improving pavement conditions for better friction coefficient will lead to improved crash recovery. Addition of lanes to alleviate problems arising out of increased ADT and restriction of trucks to the slower right lanes on the highways would not only reduce the crash occurrences but also resulted in lower injury severity levels.

Mémoire numérisé par la Division de la gestion de documents et des archives de l'Université de Montréal.

O recrutamento é extremamente importante no ambiente recifal, sendo o principal responsável pelo reabastecimento de populações adultas de peixes. Esse fenômeno é altamente complexo, não sendo claro se é influenciado apenas por processos estocásticos ou também por processos determinísticos. No presente estudo avaliamos a dinâmica temporal do recrutamento de diversas espécies de peixes recifais, identificando sítios de berçário (i.e. recrutamento estável e alto) e a influência de variáveis ambientais. Para tal, utilizamos dados de um monitoramento de médio prazo (i.e. 2001 a 2014) realizado no Banco dos Abrolhos (BA-Brasil). Foram amostrados mais de 45 sítios, sendo levantados dados sobre a comunidade de peixes, comunidade bentônica e outras variáveis ambientais. A partir desses dados, avaliamos a variação do recrutamento por sítio em dois períodos distintos (i.e. 2001-2008/2006-2014) e a influência de variáveis ambientais no recrutamento, através da técnica Boosted Regression Trees. Constatamos que diversas espécies de peixe apresentam-se com recrutamento estável em distintos sítios de amostragem. Também observamos um efeito positivo da densidade de peixes recifais coespecíficos adultos e da cobertura relativa de algas frondosas no recrutamento de diversas espécies analisadas. No geral, observamos que há certa espécie especificidade no processo de recrutamento, porém, em escalas espaciais maiores, os padrões podem estar ligados a características mais gerais, relacionadas a um grupo taxonômico mais elevado. Em relação aos sítios de berçário, um se destacou, sendo berçário de 5 diferentes espécies, incluindo Scarus trispinosus, uma das espécies prioritárias para conservação na região de Abrolhos. Assim, recomendamos a criação de uma área marinha de proteção integral que englobe o sítio em questão. Além disso, as descobertas deste trabalho nos permitem reforçar a teoria de que o recrutamento de peixes recifais pode ser influenciado por fenômenos determinísticos e não varia simplesmente de maneira estocástica. / Recruitment is extremely important in the reef environment, because it is the main source of population replenishment. Reef fish recruitment is a highly complex process and it is not clear whether it is influenced only by stochastic processes or also by deterministic processes. Herein, we aimed to investigate temporal dynamics of reef fish recruitment, identify nursery sites (i.e. predictably high recruitment sites) and evaluate the influence of environmental variables on recruitment. We used data from a medium-term time series (i.e. 2001-2014) of scientific surveys in Abrolhos Bank (BA-Brazil). We sampled more than 45 sites, for several consecutive years and recorded data about fish community, benthic community and other environmental variables. We assessed the variation of recruitment on each site, during two distinct periods (i.e. 2001-2008 / 2006-2014), and used the Boosted Regression Trees technique to evaluate the influence of environmental variables in recruitment. We found that several reef fish species present a low variable recruitment at different sampling sites. BRT showed a positive effect of the coverage of flesh algae and abundance of conspecific in the abundance of recruits (i.e. young-of-year) of many species. Overall, we notice that the recruitment traits seems to be species specific, but we also found indications that in larger spatial scales, recruitment spatial and temporal patterns may be related to general characteristics among species of the higher taxa. Nursery sites varied among species and one site was a nursery to 5 different reef fish species, including Scarus trispinosus, a species that require priority conservation in the Abrolhos Bank. Therefore, we recommend the creation of a new no-take marine protected area that encompasses this site. Our results also indicated that reef fish recruitment may be influenced by deterministic processes and do not vary only stochastically.

L'agriculture actuelle est impliquée dans de multiples problématiques environnementales, sociales et économiques, aux échelles locales et globales. En agronomie, de nombreux travaux à l'échelle du champ et de l'exploitation visent aujourd'hui à concevoir des systèmes de culture et des systèmes de production en lien avec ces problématiques. En revanche, peu de travaux portent sur la conception et l'évaluation de systèmes agricoles à l'échelle du territoire, alors que cette échelle apparaît pourtant incontournable pour faire face à des enjeux de développement durable. Pour combler ce manque, nous proposons un ensemble méthodologique permettant i) de simuler les conséquences de scénarios de politiques agricoles sur les choix d'assolement des agriculteurs, décrits individuellement, en modélisant l'évolution de leurs système de production et ii) d'évaluer l'impact de ces changements d'assolements à l'échelle du territoire, à l'aide d'indicateurs qui apportent de l'information spatiale sur la contribution de l'agriculture au développement durable. L'ensemble méthodologique proposé débute par la construction d'une typologie des exploitations agricoles du territoire sur la base de la similarité de leur assolement. Parallèlement, l'adaptation d'indicateurs à l'échelle du territoire permet d'évaluer les impacts des externalités des systèmes de culture en mobilisant des procédures de changements d'échelles. Un modèle bioéconomique générique, multi-échelle, spatialement explicite, appelé MOSAICA, qui utilise la typologie et les indicateurs d'impact de l'agriculture à l'échelle régionale, est créé pour produire des mosaïques de systèmes de culture et évalue leur contribution au développement durable du territoire. Ce modèle, couplé à un itinéraire de définition de scénarios exploratoires et normatifs permet de tester l'impact de différents types de leviers agronomiques, socio-économiques, environnementaux, organisationnels et techniques sur les choix des exploitants et in fine sur la contribution de la mosaïque de systèmes de culture au développement durable du territoire. Nous avons appliqué cet ensemble méthodologique à la conception de scénarios de développement agricoles durables en Guadeloupe. Nous avons dans un premier temps développé une typologie des systèmes de production comprenant huit types distincts et relevant de processus décisionnel différents. Puis nous avons adapté à l'échelle du territoire 19 indicateurs pour l'évaluation des mosaïques de systèmes de culture. L’évaluation de la mosaïque actuelle nous a permis de repérer de faibles niveaux de contribution aux enjeux d’autonomie alimentaire et énergétique. Différents scénarios normatifs et exploratoires intégrant des leviers de changement de la mosaïque ont été testés avec MOSAICA. Les évaluations réalisées nous ont permis d'identifier que des leviers agronomiques comme le développement du maraîchage sans intrants chimiques et des leviers sociaux comme la formation de main-d'oeuvre supplémentaire permettraient d'améliorer la contribution de l’agriculture au développement durable du territoire Guadeloupéen. La modélisation mécaniste de l’évolution du territoire agricole permet d'intégrer des connaissances sur la localisation, les performances, les impacts des systèmes de culture et sur les processus décisionnels des exploitants régissant l’orientation productive et le fonctionnement des exploitations. Cette démarche permet de visualiser les changements de système de culture et leurs impacts de manière spatialement explicite, ce qui permet de générer des connaissances sur les leviers susceptibles de faire évoluer positivement l'agriculture du territoire. La démarche et les outils mis en oeuvre sont donc particulièrement utiles pour l'aide à la décision publique pour améliorer la durabilité de l'agriculture dans son ensemble. / Current agricultural systems are responsible for many different environmental, social and economic issues at both local and global scales. Agricultural sciences have contributed to the design of several methods at the farm and field scale in order to prototype cropping systems and farming systems to address these issues. However, few methods have been designed at the regional scale, while this scale seems to be essential in order to address these issues. In order to fill this gap, we here propose a new methodological framework for i) simulating the consequences of policy changes on farmer's cropping plan, described individually, by modeling the evolution of farming systems and to ii) assess the impacts of cropping system changes at the regional scale, with a set of indicators that generate spatially explicit information on the contribution of agriculture to sustainable development. The methodological framework starts with the design of a farm typology over the territory based on the similarity of farmer's crop acreages. In parallel, a set of indicators is adapted to the landscape scale in order to assess the impacts of cropping system externalities by integrating a set of scale change procedures. A generic, multi-scale, spatially explicit bioeconomic model called MOSAICA, which uses the farm typology and the indicators, is created for generating cropping system mosaics and assessing their contribution to sustainable development. This model coupled to a scenario approach composed of exploratory and normative scenarios can simulate the impact of several types of agronomic, socio-economic, environmental, organizational and technical levers of change on the farmer's choices in terms of cropping systems and in fine the impacts of new cropping system mosaics on the contribution to sustainable development of territories. We applied this methodological framework for building scenarios of sustainable agricultural development in Guadeloupe. We first developed a typology of farming systems encompassing eight types of farming systems that revealed several different farmer's decision processes. Then, we developed 19 indicators to assess cropping system mosaics. The assessment of the current cropping system mosaic showed low levels of response of the current mosaic to economic and social issues especially the food and energy self-sufficiency. Different normative and exploratory scenarios integrating levers of change have been simulated with MOSAICA. The assessment of cropping system mosaics from these scenarios highlighted the positive effect of agronomic levers of change such as organic crop-gardening and social levers such as the vocational training of supplementary workforce for improving the contribution of agriculture to sustainable development of the guadeloupean territory. The mechanistic modeling of the agricultural territory allows us to integrate a wide range of knowledge on the location of cropping systems, their levels of performance , their impacts and the decision process of farmer's that drive the farming system characteristics and the farm functioning. This methodological framework helps visualize the cropping system changes at the regional scale and their associated impacts at the landscape scale which is helpful in order to produce knowledge on the levers of change that can improve the response of local agriculture to local and global issues. The framework and tools designed are particularly useful for decision-aid on the future levels of contribution of agriculture to sustainable development.