Neural networks demonstrate great potential for discovering non-linear relationships in time-series and extrapolating from them. Results of forecasting using financial data are particularly good [LapFar87, Schöne90, ChaMeh92]. In contrast, traditional statistical methods are restrictive as they try to express these non-linear relationships as linear models. This thesis investigates the use of the Backpropagation neural model for time-series forecasting. In general, neural forecasting research [Hinton87] can be approached in three ways: research into, the weight space, into the physical representation of inputs, and into the learning algorithms. A new method to enhance input representations to a neural network, referred to as model sNx, has been developed. It has been studied alongside a traditional method in model N. The two methods reduce the unprocessed network inputs to a value between 0 and 1. Unlike the method in model N, the variants of model sNx, sN1 and sN2, accentuate the contracted input value by different magnitudes. This different approach to data reduction exploits the characteristics of neural extrapolation to achieve better forecasts. The feasibility of the principle of model sNx has been shown in forecasting the direction of the FFSE-100 Index. The experimental strategy involved optimisation procedures using one data set and the application of the optimal network from each model to make forecasts on different data sets with similar and dissimilar patterns to the first. A Neural Forecasting System (NFS) has been developed as a vehicle for the research. The NFS offers historical and live simulations, and supports: a data alignment facility for standardising data files with non-uniform sampling times and volumes, and merging them into a spreadsheet; a parameter specification table for specifications of neural and system control parameter values; a pattern specification language for specification of input pattern formation using one or more time-series, and loading to a configured network; a snapshot facility for re-construction of a partially trained network to continue or extend a training session, or re-construction of a trained network to forecast for live tests; and a log facility for recording experimental results. Using the NFS, specific pattern features selected from major market trends have been investigated [Pring8O]: triple-top ('three peaks'), double-top ('two peaks'), narrow band ('modulating'), bull ('rising') and recovery ('U-turn'). Initially, the triple-top pattern was used in the N model to select between the logarithmic or linear data form for presenting raw input data. The selected linear method was then used in models sN1, sN2 and N for network optimisations. Experiments undertaken used networks of permutations of sizes of input nodes (I), hidden nodes (H), and tolerance value. Selections were made for: the best method, by value, direction, or value and direction, for measuring prediction accuracy; the best configuration function, H - I 4), with 4) equal to 0.9, 2 or 3; and the better of sN1 and sN2. The evaluation parameters were, among others, the prediction accuracy (%), the weighted return (%), the Relative Threshold Prediction Index (RTPI) indicator, the forecast error margins. The RTPI was developed to filter out networks forecasting above a minimum prediction accuracy with a credit in the weighted return (%). Two optimal networks, one representing model sNx and one N were selected and then tested on the double-top, narrow band, bull and recovery patterns. This thesis made the following research conthbutions. • A new method in model sNx capable of more consistent and accurate predictions. • The new RTPI neural forecasting indicator. • A method to forecast during the consolidation ('non-diversifying') trend which most traditional methods are not good at. • A set of improvements for more effective neural forecasting systems.

Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / The control of asian soybean rust depends on the application of fungicides at the right time. The use of forecasting systems is an important tool in the decision-making process. This work aims to estimate a prediction algorithm that generates risk levels of Phakopsora pachyrhizi infection based on rainfall, minimum temperature, sowing date, growth stage of the crop and local, aimed at applying fungicides products at the correct time. Four experiments were conducted in the experimental area of Phytus Institute, Itaara city, central region of Rio Grande do Sul, in the crop 2014/2015. Each experiment corresponded to a different sowing date and consisted of treatments: control (T1) without fungicide application; application as recommended in the algorithm to be assessed (T2), application of the scheduled program in days after emergence (DAE) (T3), based on growth stage of the crop (T4), application as recommended in the algorithm with seven days delay (T5), application of the scheduled program in days after emergence with seven days delay (T6) and application based on growth stage of the crop with seven days delay (T7). To evaluate the effect of treatments were determined to AUCPD, rate of progress, productivity, weight of a thousand grains, and correlations between the dependent and independent variables related to the first pustule and disease severity. The positioning defined by the use of the algorithm did not provide superiority over AUCPD variable rate of progress, productivity and weight of a thousand grains, compared to other treatments in none of the experiments. Asian rust occurred at different growth stages of soybean and the use of sowing dates may indirectly measure the pressure of inoculum of this pathogen. The seven-day period is not consistent for the calculation of the meteorological variables that precede the disease. Temperature was not relevant to explain the epidemic and its use in the algorithm not justified. Rainfall had decisive influence on the epidemic and the more periods of rain occurred, the higher were the severity levels. / O controle da ferrugem asiática da soja é dependente da aplicação de fungicidas no momento correto. O uso de sistemas de previsão é uma ferramenta importante no processo de tomada de decisão. O objetivo deste trabalho foi aferir um algoritmo de previsão que gera níveis de risco de infecção de Phakopsora pachyrhizi baseado em precipitação, temperatura mínima, época de semeadura, estádio fenológico da cultura e local, visando a aplicação de produtos fungicidas na época correta. Foram conduzidos quatro experimentos na área experimental do Instituto Phytus, município de Itaara, região central do Rio Grande do Sul, na safra 2014/2015. Cada experimento correspondeu a uma época de semeadura diferente e foi constituído dos tratamentos: testemunha (T1) sem aplicação de fungicida; aplicação de acordo com o recomendado no algoritmo a ser aferido (T2), aplicação do programa calendarizado em dias após a emergência (DAE)(T3), programa baseado nos estádios da soja (T4), aplicação com sete dias de atraso do recomendado pelo algoritmo (T5), aplicação calendarizada com sete dias de atraso (T6) e aplicação baseada em estádio fenológico com sete dias de atraso (T7). Para avaliar o efeito dos tratamentos, foram determinados a AACPD, taxa de progresso, produtividade, massa de mil grãos, e correlações entre as variáveis dependentes e independentes relacionadas a primeira pústula e a severidade da doença. O posicionamento definido pelo uso do algoritmo não propiciou superioridade sobre as variáveis AACPD, taxa de progresso, produtividade e peso de mil grãos, em relação aos demais tratamentos em nenhum dos experimentos. A ferrugem asiática ocorreu em diferentes estádios fenológicos da soja e o uso de épocas de semeadura pode medir indiretamente a pressão do inóculo deste patógeno. O período de sete dias não é consistente para cálculo das variáveis meteorológicas que precedem a doença. Temperatura não foi relevante para explicar a epidemia e seu uso no algoritmo não se justificou. A precipitação apresentou influência decisiva na epidemia e quanto mais períodos de chuva ocorreram, maiores foram os níveis de severidade.

Submitted by Liliane Ferreira (ljuvencia30@gmail.com) on 2018-11-19T11:07:15Z No. of bitstreams: 2 Dissertação - Júnio Santos Bulhões - 2018.pdf: 4626907 bytes, checksum: 268499105ec64b2e9abf04faa47a91e2 (MD5) license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) / Approved for entry into archive by Luciana Ferreira (lucgeral@gmail.com) on 2018-11-19T11:41:27Z (GMT) No. of bitstreams: 2 Dissertação - Júnio Santos Bulhões - 2018.pdf: 4626907 bytes, checksum: 268499105ec64b2e9abf04faa47a91e2 (MD5) license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) / Made available in DSpace on 2018-11-19T11:41:27Z (GMT). No. of bitstreams: 2 Dissertação - Júnio Santos Bulhões - 2018.pdf: 4626907 bytes, checksum: 268499105ec64b2e9abf04faa47a91e2 (MD5) license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) Previous issue date: 2018-10-29 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / This work proposes a methodology that uses spectral analysis and system identification in order to fill gaps in time series. The methodology proposes the implementation of predictions in time series of physical and chemical variables that are related with flood areas that are collected with no frequency. It is used predictive neural network with autoregressive model and classification neural network. Collected values are extracted from the original data set in order to later test and validate the proposed methodology. The results demonstrated the effectiveness of the methodology, which is able to predict the behavior of different variables using the previously recognized patterns in the time series. / Este trabalho propõe metodologia que utiliza análise espectral em conjunto com modelo de identificação de sistema para preenchimento de lacunas em séries temporais. A metodologia propõe realizar previsão em séries temporais de variáveis físicas e químicas relacionadas as inundações com frequência de coleta variável. Utiliza-se rede neural artificial de previsão com modelo autorregressivo e rede neural classificatória. Valores coletados são armazenados para posteriormente testar e validar a metodologia proposta. Os resultados alcançados demonstram a eficácia da metodologia, que é capaz de prever o comportamento de diferentes variáveis utilizando os padrões reconhecidos previamente nas séries temporais.

Every year floods are responsible of a significant number of human losses, many of which could be avoided with a broader implementation of flood forecasting systems. Nevertheless, there are still some technological and economic limitations that impede the creation of these systems in many parts of the world. At the core of many flood forecasting systems is a hydrologic model that transforms the weather forecast into a flow forecast. Using real-time modeling for potential floods poses a series of problems: if the model is complex, the computational power required can be significant, and consequently expensive, and if the model is simple enough to run on regular computers in the time allotted, it is likely that the results will not be accurate enough to be useful. I propose the development of a standardized method for using pre-computed scenarios as an alternative to real-time flood modeling. I explain how pre-computing has been used on other realms in the past, and how it is beginning to be implemented in different branches of hydrology, the prediction coastal flooding due to storms or tsunamis being one of the most developed. My research has focused on answering the questions that arise during the design stage of a flood forecasting system not only for rain or snow driven floods, but also by anthropogenic-produced floods. I analyze the number of parameters and their granularity to be used to create the scenarios, the accuracy of the results, different strategies to implement the systems, etc. Finally, I present some test-cases of the application of the method, and assess their results.

A robust evaluation toolset has been designed for Naval Research Laboratory’s Real-Time Ocean Forecasting System RELO with the purpose of facilitating an adaptive sampling strategy and providing a more educated guidance for routing underwater gliders. The major challenges are to integrate into the existing operational system, and provide a bridge between the modeling and operative environments. Visualization is the selected approach and the developed software is divided into 3 packages: The first package is to verify that the glider is actually following the waypoints and to predict the position of the glider for the next cycle’s instructions. The second package helps ensures that the delivered waypoints are both useful and feasible. The third package provides the confidence levels for the suggested path. This software’s implementation is in Python for portability and modularity to allow for easy expansion for new visuals.

<p>In a world that is moving faster and faster, a company’s ability to align to market changes is becoming a major competitive factor. Forecasting enables companies to predict what lies ahead, e.g. trend shifts or market turns, and makes it possible to plan for it. But looking into the future is never an easy task.</p><p>“Prediction is very difficult, especially if it’s about the future.” (Niels Bohr, 1885-1962)</p><p>However, progress in the field of forecasting has shown that it is possible for companies to improve on forecasting practices. This master thesis looks at the sales forecasting practices in MNCs primarily operating in emerging and developing countries. We examine the whole process of sales forecasting, also known as forecasting management, in order to develop a comprehensive model for forecasting in this type of companies. The research is based on a single case study, which is then later generalized into broader conclusions.</p><p>The conclusion of this master thesis is that forecasting is a four-step exercise. The four stages we have identified are: Knowledge creation, knowledge transformation, knowledge use and feedback. In the course of these four stages a company’s sales forecast is developed, changed and used. By understanding how each stage works and what to focus on, companies will be able to improve their forecasting practices.</p>

In a world that is moving faster and faster, a company’s ability to align to market changes is becoming a major competitive factor. Forecasting enables companies to predict what lies ahead, e.g. trend shifts or market turns, and makes it possible to plan for it. But looking into the future is never an easy task. “Prediction is very difficult, especially if it’s about the future.” (Niels Bohr, 1885-1962) However, progress in the field of forecasting has shown that it is possible for companies to improve on forecasting practices. This master thesis looks at the sales forecasting practices in MNCs primarily operating in emerging and developing countries. We examine the whole process of sales forecasting, also known as forecasting management, in order to develop a comprehensive model for forecasting in this type of companies. The research is based on a single case study, which is then later generalized into broader conclusions. The conclusion of this master thesis is that forecasting is a four-step exercise. The four stages we have identified are: Knowledge creation, knowledge transformation, knowledge use and feedback. In the course of these four stages a company’s sales forecast is developed, changed and used. By understanding how each stage works and what to focus on, companies will be able to improve their forecasting practices.

Conselho Nacional de Desenvolvimento Científico e Tecnológico / The tomato (Lycopersicon esculentum Mill) is an important crop in the world and an important product for trade "in nature and for industry. The amount of fungicides used in tomato crop for control of diseases is high and contributes significantly to production costs of tomato, besides the high risk of intoxication of the applicator and environmental problems. So, the use of disease forecasting systems based on mathematical models to manage the applications of fungicides for foliar diseases control in tomato may to reduce production costs, by decreasing the number of fungicide applications and machinery operation during the culture cycle. The objective was to test the systems for predicting occurrence of late blight (Phytophthora infestans) and early blight (Alternaria solani) on irrigated tomato under two climatic conditions and estimate the cost of deployment. Three experiments were conducted, the first experiment in the spring of 2007, in Santa Maria - RS. The other two were accomplished in Cristalina - GO. There were two seasons of transplanting (01/04/08 and 03/05/08). Used the hybrid U2006 (UNILEVER), with growth habit determined. The spacing between the rows of plants was 1.3 m between plants in rows of 0,3 m. The randomized block design with three replications was used, each plot of 5,2 m wide, consisting of 4 rows of plants with 5 m long. The systems for predicting disease FAST, to early blight, and Blitecast to late blight were used. Irrigation was performed by spraying in Santa Maria and sprinkler, surface drip and subsurface in Cristalina-GO. The results showed that the use of disease forecasting systems is the most appropriate way to control diseases in tomato, as with the use of disease forecasting systems that can significantly reduce the number of fungicide applications in relation the calendar, based on weekly applications in areas with conditions generally unfavorable to the development of disease. Considering the scenarios of reductions of the number of fungicide applications (30, 43 and 65%), compared to the weekly application, in climatic conditions unfavorable to disease development, it is possible to recover the costs of implementing the system prediction in all scenarios. / A cultura do tomate (Lycopersicon esculentum Mill) é uma das mais expressivas culturas no cenário agrícola mundial, constituindo importante produto para o comércio in natura e indústria de extratos. O volume de fungicidas utilizado na cultura do tomateiro no controle das principais doenças é elevado e contribui significativamente nos custos de produção de tomate, além do elevado risco de intoxicação do aplicador e dos problemas ambientais. Nesse sentido, utilização de sistemas de previsão e aviso, baseados em modelos matemáticos para gerenciar o manejo das aplicações de fungicidas para o controle das principais doenças foliares no tomateiro poderá auxiliar na redução dos custos de produção, através da diminuição do número de aplicações de fungicidas e de operações de máquinas durante o ciclo de desenvolvimento da cultura. O objetivo do trabalho foi testar os sistemas de previsão de ocorrência de requeima e alternaria em tomateiro industrial irrigado sob duas condições climáticas e estimar o seu custo de implantação. Foram realizados três experimentos, o primeiro experimento no segundo semestre de 2007, em Santa Maria RS. Os outros dois foram realizados no município de Cristalina GO. Realizaram-se duas épocas de transplante (01/04/08 e 03/05/08). Utilizou-se a cultivar híbrida U2006 (UNILEVER), de hábito de crescimento determinado. O espaçamento entre as fileiras de plantas foi de 1,3 m e entre as plantas nas fileiras de 0,3 m. Utilizou-se o delineamento experimental de Blocos ao Acaso com três repetições, sendo cada parcela de 5,2 m de largura, composta de 4 fileiras de plantas com 5 m de comprimento. Utilizaram-se os sistemas de previsão de doenças FAST, para mancha de alternaria, e BLITECAST, para requeima. A irrigação foi realizada por aspersão em Santa Maria e por aspersão, por gotejamento superficial e subsuperficial em Cristalina. Os resultados mostraram que o uso de sistemas de previsão de doenças é a maneira mais adequada para o controle de doenças na cultura do tomateiro, pois com o uso de sistemas de previsão de doenças se consegue reduzir significativamente o número de aplicações de fungicidas, em relação ao calendário semanal, em regiões com condições geralmente desfavoráveis ao desenvolvimento de doenças. Considerando-se os cenários de reduções do número de aplicação de fungicidas (30, 43 e 65%), em relação ao calendário semanal de aplicação, condições climáticas desfavoráveis ao desenvolvimento de doenças, é possível recuperar os custos de implantação do sistema de previsão em todos os cenários.