Typical, potential-driven force fields have been usefully applied to small molecules for decades. However, complex effects such as polarisation, π systems and hydrogen bonding remain difficult to model while these effects become increasingly relevant. In fact, these complex electronic effects become crucial when considering larger biological molecules in solution. Instead, machine learning can be used to recognise patterns in chemical behaviour and predict them, sacrificing computational efficiency for accuracy and completeness of the force field. The kriging machine learning method is capable of taking the geometric features of a molecule and predicting its electrostatic properties after being trained using ab initio data of the same system. We present significant improvements in functionality, application and understanding of the kriging machine learning as part of an electrostatic force field. These improvements are presented alongside an up-scaling of the problems the force field is applied to. The force field predicts electrostatic energies for all common amino acids with a mean error of 4.2 kJmol-1 (1 kcal mol-1), cholesterol with a mean error of 3.9 kJmol-1 and a 10-alanine helix with a mean error of 6.4 kJmol-1. The kriging machine learning has been shown to work identically with charged systems, π systems and hydrogen bonded systems. This work details how different chemical environments and parameters affect the kriging model quality and assesses optimal methods for computationally-efficient kriging of multipole moments. In addition to this, the kriging models have been used to predict moments for atoms they have had no training data for with little loss in accuracy. Thus, the kriging machine learning has been shown to produce transferable models.

Orientadores: Paulo Sérgio Graziano Magalhãe, Armando Zaupa Remacre / Dissertação (mestrado) - Universidade Estadual de Campinas, Faculdade de Engenharia Agrícola / Made available in DSpace on 2018-08-27T21:34:19Z (GMT). No. of bitstreams: 1 Sanches_GuilhermeMartineli_M.pdf: 13551439 bytes, checksum: 1aa534b602f92044d4ea34bee5e63027 (MD5) Previous issue date: 2015 / Resumo: Umas das ferramentas utilizadas na Agricultura de Precisão (AP) é a geoestatística, cujo principal objetivo é a descrição dos padrões espaciais e a estimativa de dados em locais não amostrados. Um dos fatores limitantes para se fazer um adequado mapeamento do solo e atender os requisitos mínimos dos métodos de interpolação é a necessidade de uma amostragem densa da área, o que inviabiliza muitas vezes o mapeamento do solo, devido ao demorado e custoso processo de retirada de amostras. Dentro deste contexto os métodos de interpolação geoestatísticos vislumbram uma solução para o presente desafio, tornando possível a descrição da variabilidade espacial do solo com uma pequena amostragem da variável a qual se deseja conhecer, utilizando para isto outros atributos que são mais facilmente mensuráveis e a um custo menor. Uma das técnicas possíveis para otimizar a quantidade de pontos amostrais consiste na utilização de dados obtidos através de sensores de solo para orientação da amostragem. Este trabalho tem como hipótese que, utilizando dados provenientes de sensores de Condutividade Elétrica Aparente (CEa) do solo em conjunto com técnicas de geoestatística, é possível, através de uma amostragem direcionada e reduzida, conhecer a descrição da variabilidade espacial da fertilidade e do estado físico dos solos com adequada precisão. A presente pesquisa teve como objetivo obter mapas da variabilidade espacial dos atributos químicos e físicos do solo utilizando um número reduzido de amostras e aplicando métodos de interpolação geoestatísticos (krigagem ordinária e com deriva externa), tendo como base dados de condutividade elétrica aparente do solo. A metodologia utilizada para a obtenção dos mapas de variabilidade espacial dos atributos do solo indicam que é possível prever mapas que podem ser utilizados para recomendação de fertilizantes à taxa variável. Esta abordagem abre novas possibilidades para que atributos agronômicos importantes possam ser estimados em grandes áreas a partir de um número reduzido de amostras, auxiliando os agricultores no manejo da cultura e tomada de decisão / Abstract: One of the tools used in precision agriculture (PA) is geostatistics, which main objective is to describe the spatial patterns and the estimated data on non-sampled locations. One of the limiting factors for making a proper soil mapping and meet the minimum requirements of interpolation methods is the need for a dense sampling grid, which often makes it impossible, as the process are time consuming and expensive. Within this context, the geostatistical interpolation methods envision a solution for this challenge, making it possible to describe the soil spatial variability with a small sampling of the primary variable (which you want to know), using other attributes that are easily measured. One of the possible techniques to optimize the number of soil sampling is the use of data obtained from soil sensors. This work the assumption that, using data from the Apparent Electrical Conductivity (ECa) together with geostatistical techniques, it is possible, through a targeted and reduced sampling, know the spatial variability of soils fertility and physical condition with adequate precision. Therefore, this research aims to obtain maps of the spatial variability of chemical and physical soil properties using a reduced number of samples and applying geostatistical interpolation methods (ordinary kriging and kriging with external drift), based on data of apparent electrical conductivity. The methodology used to obtain the maps of spatial variability of soil attributes indicate that it is possible to provide maps that can be used for fertilizer recommendation to the variable rate. This approach opens new possibilities for important agronomic attributes be estimated in large areas from a small number of samples, assisting farmers in crop management and decision-making / Mestrado / Maquinas Agricolas / Mestre em Engenharia Agrícola

El desarrollo de procesos químicos eficientes, tanto desde un punto de vista económico como desde un punto de vista ambiental, es uno de los objetivos principales de la Ingeniería Química. Para conseguir este propósito, durante los últimos años, se están empleando herramientas avanzadas para el diseño, simulación, optimización y síntesis de procesos químicos, las cuales permiten obtener procesos más eficientes y con el menor impacto ambiental posible. Uno de los aspectos más importantes a tener en cuenta para diseñar procesos más eficientes es la disminución del consumo energético. El consumo energético del sector industrial a nivel global representa aproximadamente el 22.2 % del consumo energético total, y dentro de este sector, la industria química representa alrededor del 27 %. Por lo tanto, el consumo energético de la industria química a nivel global constituye aproximadamente el 6 % de toda la energía consumida en el mundo. Además, teniendo en cuenta que la mayor parte de la energía consumida es generada principalmente a partir de combustibles fósiles, cualquier mejora de los procesos químicos que reduzca el consumo energético supondrá una reducción del impacto ambiental. El trabajo recopilado en esta Tesis Doctoral se ha llevado a cabo dentro del grupo de investigación COnCEPT, perteneciente al Instituto Universitario de Ingeniería de los Procesos Químicos de la Universidad de Alicante, durante los años 2014 y 2017. El objetivo principal de la presente Tesis Doctoral se centra en el desarrollo de herramientas y modelos de simulación y optimización de procesos químicos con el fin de mejorar la eficiencia energética de éstos, lo que conlleva a la disminución del impacto ambiental de los procesos. Más concretamente, esta Tesis Doctoral se compone de dos estudios principales, que son los objetivos concretos que se pretenden conseguir: - Estudio y evaluación de los modelos surrogados para la mejora en la optimización basada en simuladores de procesos químicos. - Desarrollo de nuevos modelos para la optimización de procesos químicos y la integración de energía simultánea, para redes de intercambiadores de calor.

Windstorms result in significant damage and economic loss and are a major recurring threat in many countries. Estimating surface-level wind speeds resulting from windstorms is a complicated problem, but geostatistical spatial interpolation methods present a potential solution. Maximum sustained and peak gust weather station data from two historic windstorms in Europe were analyzed to predict surface-level wind speed surfaces across a large and topographically varied landscape. Disjunctively sampled maximum sustained wind speeds were adjusted to represent equivalent continuously sampled 10-minute wind speeds and missing peak gust station data were estimated by applying a gust factor to the recorded maximum sustained wind speeds. Wind surfaces were estimated based on anisotropic and isotropic kriging interpolation methodologies. The study found that anisotropic kriging is well-suited for interpolating wind speeds in meso- and macro-scale areas because it accounts for wind direction and trends in wind speeds across a large, heterogeneous surface, and resulted in interpolation surface improvement in most models evaluated. Statistical testing of interpolation error for stations stratified by geographic classification revealed that stations in coastal and/or mountainous locations had significantly higher prediction errors when compared with stations in non-coastal/non-mountainous locations. These results may assist in mitigating losses to structures due to excessive wind events.

With the massive increase in usage of some parts of the electromagnetic spectrum during the last decades, the ability to create real time maps of signal coverage is now more important than ever before. This Masters project is designed to test two different methods of generating such maps with a one second limit to processing time. The interpolation methods under consideration are known as inverse distance weighting and kriging. Several different variants of kriging are considered and compared some of which were implemented specif cally for the project and one variant designed by a third party.The data used is acquired from an antenna array inside a laboratory room at LTU rather than being simulated. The data collection is done with the transmitter at several different positions in the room to make sure the interpolation works consistently. The results show only small differences in both the mean and median of the absolute error when comparing inverse distance weighting and kriging and the variations between transmitter positions are signifcant enough that no single variant is consistently the best using that metric. Using a resolution with 25cm2 pixel size there were no problems reaching significantly lower than the 1sec time limit. If the resolution is increased to apixel size of 1cm2 neither method is able to consistently update the map at the required pace. Kriging however showed that it can generate values outside the range of observed values which could make the extra effort required to implement it worth it since such a characteristic might be very useful for  finding the transmitter.

Quantile function modeling is a more robust, comprehensive, and flexible method of statistical analysis than the commonly used mean-based methods. More and more data are collected in the form of multivariate, functional, and multivariate functional data, for which many aspects of quantile analysis remain unexplored and challenging. This thesis presents a set of quantile analysis methods for multivariate data and multivariate functional data, with an emphasis on environmental applications, and consists of four significant contributions. Firstly, it proposes bivariate quantile analysis methods that can predict the joint distribution of bivariate response and improve on conventional univariate quantile regression. The proposed robust statistical techniques are applied to examine barley plants grown in saltwater and freshwater conditions providing interesting insights into barley’s responses, informing future crop decisions. Secondly, it proposes modeling and visualization of bivariate functional data to characterize the distribution and detect outliers. The proposed methods provide an informative visualization tool for bivariate functional data and can characterize non-Gaussian, skewed, and heavy-tailed distributions using directional quantile envelopes. The radiosonde wind data application illustrates our proposed quantile analysis methods for visualization, outlier detection, and prediction. However, the directional quantile envelopes are convex by definition. This feature is shared by most existing methods, which is not desirable in nonconvex and multimodal distributions. Thirdly, this challenge is addressed by modeling multivariate functional data for flexible quantile contour estimation and prediction. The estimated contours are flexible in the sense that they can characterize non-Gaussian and nonconvex marginal distributions. The proposed multivariate quantile function enjoys the theoretical properties of monotonicity, uniqueness, and the consistency of its contours. The proposed methods are applied to air pollution data. Finally, we perform quantile spatial prediction for non-Gaussian spatial data, which often emerges in environmental applications. We introduce a copula-based multiple indicator kriging model, which makes no distributional assumptions on the marginal distribution, thus offers more flexibility. The method performs better than the commonly used variogram approach and Gaussian kriging for spatial prediction in simulations and application to precipitation data.

The circular dataimage is defined by displaying direction as the color at the same direction in a color wheel composed of a sequence of two-color gradients with color continuity between gradients. The resulting image of circular-spatial data is continuous with high resolution. Examples include ocean wind direction, Earth's main magnetic field, and rocket nozzle internal combustion flow. The cosineogram is defined as the mean cosine of the angle between random components of direction as a function of distance between observation locations. It expresses the spatial correlation of circular-spatial data. A circular kriging solution is developed based on a model fitted to the cosineogram. A method for simulating circular random fields is given based on a transformation of a Gaussian random field. It is adaptable to any continuous probability distribution. Circular random fields were implemented for selected circular probability distributions. An R software package was created with functions and documentation.

The estimation of bedrock level for soil and rock engineering is a challenge associated to many uncertainties. Nowadays, this estimation is performed by geotechnical or geophysics investigations. These methods are expensive techniques, that normally are not fully used because of limited budget. Hence, the bedrock levels in between investigations are roughly estimated and the uncertainty is almost unknown. Machine learning (ML) is an artificial intelligence technique that uses algorithms and statistical models to predict determined tasks. These mathematical models are built dividing the data between training, testing and validation samples so the algorithm improve automatically based on passed experiences. This thesis explores the possibility of applying ML to estimate the bedrock levels and tries to find a suitable algorithm for the prediction and estimation of the uncertainties. Many diferent algorithms were tested during the process and the accuracy level was analysed comparing with the input data and also with interpolation methods, like Kriging. The results show that Kriging method is capable of predicting the bedrock surface with considerably good accuracy. However, when is necessary to estimate the prediction interval (PI), Kriging presents a high standard deviation. The machine learning presents a bedrock surface almost as smooth as Kriging with better results for PI. The Bagging regressor with decision tree was the algorithm more capable of predicting an accurate bedrock surface and narrow PI. / BIG and BeFo project "Rock and ground water including artificial intelligence

No description available.

No description available.