O objetivo principal deste trabalho foi testar redes neurais artificiais (RNAs) do tipo multilayer perceptron (MLP) na estimativa da evapotranspiração de referência e da evapotranspiração na linha de plantio de limeira ácida. As RNAs foram treinadas sob algoritmo de gradiente conjugado de erros, com funções de ativação sigmóide na camada intermediária e linear na camada de saída. Foram conduzidas análises comparativas com modelos de regressão. Valores diários de evapotranspiração de referência foram calculados usando o modelo Penman-Monteith (EToPM) a partir de dados meteorológicos (1997-2006) observados em Piracicaba, estado de São Paulo, Brasil (latitude: 22º 42 30 S; longitude: 47º 38 30 W; altitude: 546 m). Os modelos foram desenvolvidos a partir de dados de radiação solar global (Rg), saldo de radiação (Rn) ou radiação no topo da atmosfera (RTA) em combinação com temperatura do ar (Tar), déficit de pressão de vapor no ar (DPV) e velocidade do vento (u). Bom desempenho foi obtido quando os dados de Rg ou Rn estavam disponíveis, mesmo com a falta de uma ou mais das outras variáveis exigidas pelo modelo Penman- Monteith. As RNAs mostraram melhor desempenho do que os modelos de regressão, especialmente quando RTA foi considerada na entrada. O erro absoluto médio (MAE) das RNAs variou de 0,1 a 0,2 mm d-1, representando de 4 a 6 % dos valores médios de EToPM. A evapotranspiração na linha de plantio, condutância difusiva e transpiração foliar foram obtidas em pomar adulto de limeira ácida (Citrus latifolia Tan.), com espaçamento 7 m × 4 m , orientação Leste-Oeste das linhas de plantio e sem limitação hídrica, em Piracicaba, Brasil. A condutância à difusão de vapor (gs) e transpiração foliar (T) foram determinadas com porômetro de equilíbrio constante e balanço nulo, em folhas completamente expandidas, na parte média da copa nas faces expostas da linha de plantio, a intervalos horários ao longo de 42 dias. A densidade de fluxo de fótons fotossintéticos (DFFF) incidentes sobre a folha, temperatura e déficit de pressão de vapor no ar (Tar e DPV) no interior do pomar e o horário de observação (h) foram combinados nos modelos de estimativa de gs e T. Somente os modelos ajustados para o inverno apresentaram bom desempenho. Medidas lisimétricas foram utilizadas na determinação da evapotranspiração diurna na linha de plantio (ETli 9-17h). Saldo de radiação (Rn), temperatura do ar (Tar), déficit de pressão de vapor (DPV), evapotranspiração de referência estimada pelo modelo Penman-Monteith (EToPM) e dia do ano foram combinados na estimativa de ETli 9-17h. O desempenho das RNAs foi superior ao dos modelos com base em regressão. O erro médio absoluto (MAE) nos modelos RNAs variou entre 3,6 e 10,6 L planta-1, representando de 6 a 18% dos valores médios de ETli 9-17h. Os modelos incluindo o efeito temporal apresentaram melhor desempenho. A estimativa da evapotranspiração de referência na escala diária e da evapotranspiração diurna na linha de plantio pelos modelos propostos mostrou-se adequada. Ficou evidente a existência de outros efeitos temporais operando concomitantemente com o ambiente atmosférico na determinação de gs e ETli 9-17h. / The main objective of this study was to test artificial neural networks (ANNs) of multilayer perceptron type (MLP) for estimating reference evapotranspiration, diffusive leaf conductance and crop evapotranspiration of a mature and irrigated citrus orchard. The ANNs were trained under conjugate gradient algorithm. The sigmoid and linear activation functions were used for the hidden and output nodes, respectively. Comparative analyses with regression models were carried out. Daily values of reference evapotranspiration were computed using the Penman-Monteith method (EToPM) from climatic data (1997-2006) at Piracicaba, Brazil. All models were developed considering global radiation (Rg), net radiation (Rn) or extraterrestrial radiation (Ra) in combination with air temperature (Tar), air vapor pressure deficit (VPD) and wind velocity (u) as input data. Good performance was obtained for any model when net radiation or solar radiation were available, even missing one or more of other variables required by the Penman-Monteith equation. The performance of ANNs were improved when compared to those obtained with regression model basis, especially when Ra was considered as input data. Mean absolute error (MAE) from ANNs varied from 0.1 to 0.2 mm d-1, representing between 4 and 6 % of the mean EToPM values. Crop evapotranspiration, leaf diffusive conductance and leaf transpiration data were obtained from an acid lime (Citrus latifolia Tan.) mature orchard, located at the same region. The orchard, with East-West planting rows and 7 m × 4 m spacing, was drip irrigated to maintain non-limiting water conditions. Leaf diffusive conductance to water vapor (gs) and transpiration (T) were measured on fully expanded leaves, in the middle height of the canopy, at Northen and Southern exposed faces, in hourly intervals along 42 selected days, using a steady-state null-balance porometer. Variability of gs and T values were described as function of the exposition faces of the planting rows, time of day and season. Significant differences between exposition faces for gs and T values were only observed in the spring. The relationship between gs or T values and leaf environmental conditions varied according to the season. Photosynthetic photon flux density (PPFD) incident on the leaf, air temperature (Tar) and vapor pressure deficit (VPD) and time of day (h) were used as inputs. Adequate performance was only observed for winter models. Lysimetric data were used to determine diurnal evapotranspiration from orchard row (ETli 9-17h). Net radiation (Rn), air temperature and deficit pressure vapor (Tar, DPV) and Penman-Monteith reference evapotranspiration (EToPM) data were combined in the regression analyses and developing process of ANNs. Also any other temporal effect was taken into account by including day of the year (DOY). Mean absolute error (MAE) for ANNs models varied from 3.6 to 10.6 L plant-1, representing between 6 and 18% of mean ETli 9-17h values. Errors decreased when DOY was included. According to the results, it can be concluded that it is possible to estimate daily EToPM and diurnal citrus orchard evapotranspiration (ETli 9-17h) accurately by the proposed models. Relevance of other temporal effects operating on gs and ETli 9-17h determination, in addition to environmental variations, was evident.

O objetivo principal deste trabalho foi testar redes neurais artificiais (RNAs) do tipo multilayer perceptron (MLP) na estimativa da evapotranspiração de referência e da evapotranspiração na linha de plantio de limeira ácida. As RNAs foram treinadas sob algoritmo de gradiente conjugado de erros, com funções de ativação sigmóide na camada intermediária e linear na camada de saída. Foram conduzidas análises comparativas com modelos de regressão. Valores diários de evapotranspiração de referência foram calculados usando o modelo Penman-Monteith (EToPM) a partir de dados meteorológicos (1997-2006) observados em Piracicaba, estado de São Paulo, Brasil (latitude: 22º 42 30 S; longitude: 47º 38 30 W; altitude: 546 m). Os modelos foram desenvolvidos a partir de dados de radiação solar global (Rg), saldo de radiação (Rn) ou radiação no topo da atmosfera (RTA) em combinação com temperatura do ar (Tar), déficit de pressão de vapor no ar (DPV) e velocidade do vento (u). Bom desempenho foi obtido quando os dados de Rg ou Rn estavam disponíveis, mesmo com a falta de uma ou mais das outras variáveis exigidas pelo modelo Penman- Monteith. As RNAs mostraram melhor desempenho do que os modelos de regressão, especialmente quando RTA foi considerada na entrada. O erro absoluto médio (MAE) das RNAs variou de 0,1 a 0,2 mm d-1, representando de 4 a 6 % dos valores médios de EToPM. A evapotranspiração na linha de plantio, condutância difusiva e transpiração foliar foram obtidas em pomar adulto de limeira ácida (Citrus latifolia Tan.), com espaçamento 7 m × 4 m , orientação Leste-Oeste das linhas de plantio e sem limitação hídrica, em Piracicaba, Brasil. A condutância à difusão de vapor (gs) e transpiração foliar (T) foram determinadas com porômetro de equilíbrio constante e balanço nulo, em folhas completamente expandidas, na parte média da copa nas faces expostas da linha de plantio, a intervalos horários ao longo de 42 dias. A densidade de fluxo de fótons fotossintéticos (DFFF) incidentes sobre a folha, temperatura e déficit de pressão de vapor no ar (Tar e DPV) no interior do pomar e o horário de observação (h) foram combinados nos modelos de estimativa de gs e T. Somente os modelos ajustados para o inverno apresentaram bom desempenho. Medidas lisimétricas foram utilizadas na determinação da evapotranspiração diurna na linha de plantio (ETli 9-17h). Saldo de radiação (Rn), temperatura do ar (Tar), déficit de pressão de vapor (DPV), evapotranspiração de referência estimada pelo modelo Penman-Monteith (EToPM) e dia do ano foram combinados na estimativa de ETli 9-17h. O desempenho das RNAs foi superior ao dos modelos com base em regressão. O erro médio absoluto (MAE) nos modelos RNAs variou entre 3,6 e 10,6 L planta-1, representando de 6 a 18% dos valores médios de ETli 9-17h. Os modelos incluindo o efeito temporal apresentaram melhor desempenho. A estimativa da evapotranspiração de referência na escala diária e da evapotranspiração diurna na linha de plantio pelos modelos propostos mostrou-se adequada. Ficou evidente a existência de outros efeitos temporais operando concomitantemente com o ambiente atmosférico na determinação de gs e ETli 9-17h. / The main objective of this study was to test artificial neural networks (ANNs) of multilayer perceptron type (MLP) for estimating reference evapotranspiration, diffusive leaf conductance and crop evapotranspiration of a mature and irrigated citrus orchard. The ANNs were trained under conjugate gradient algorithm. The sigmoid and linear activation functions were used for the hidden and output nodes, respectively. Comparative analyses with regression models were carried out. Daily values of reference evapotranspiration were computed using the Penman-Monteith method (EToPM) from climatic data (1997-2006) at Piracicaba, Brazil. All models were developed considering global radiation (Rg), net radiation (Rn) or extraterrestrial radiation (Ra) in combination with air temperature (Tar), air vapor pressure deficit (VPD) and wind velocity (u) as input data. Good performance was obtained for any model when net radiation or solar radiation were available, even missing one or more of other variables required by the Penman-Monteith equation. The performance of ANNs were improved when compared to those obtained with regression model basis, especially when Ra was considered as input data. Mean absolute error (MAE) from ANNs varied from 0.1 to 0.2 mm d-1, representing between 4 and 6 % of the mean EToPM values. Crop evapotranspiration, leaf diffusive conductance and leaf transpiration data were obtained from an acid lime (Citrus latifolia Tan.) mature orchard, located at the same region. The orchard, with East-West planting rows and 7 m × 4 m spacing, was drip irrigated to maintain non-limiting water conditions. Leaf diffusive conductance to water vapor (gs) and transpiration (T) were measured on fully expanded leaves, in the middle height of the canopy, at Northen and Southern exposed faces, in hourly intervals along 42 selected days, using a steady-state null-balance porometer. Variability of gs and T values were described as function of the exposition faces of the planting rows, time of day and season. Significant differences between exposition faces for gs and T values were only observed in the spring. The relationship between gs or T values and leaf environmental conditions varied according to the season. Photosynthetic photon flux density (PPFD) incident on the leaf, air temperature (Tar) and vapor pressure deficit (VPD) and time of day (h) were used as inputs. Adequate performance was only observed for winter models. Lysimetric data were used to determine diurnal evapotranspiration from orchard row (ETli 9-17h). Net radiation (Rn), air temperature and deficit pressure vapor (Tar, DPV) and Penman-Monteith reference evapotranspiration (EToPM) data were combined in the regression analyses and developing process of ANNs. Also any other temporal effect was taken into account by including day of the year (DOY). Mean absolute error (MAE) for ANNs models varied from 3.6 to 10.6 L plant-1, representing between 6 and 18% of mean ETli 9-17h values. Errors decreased when DOY was included. According to the results, it can be concluded that it is possible to estimate daily EToPM and diurnal citrus orchard evapotranspiration (ETli 9-17h) accurately by the proposed models. Relevance of other temporal effects operating on gs and ETli 9-17h determination, in addition to environmental variations, was evident.

The thesis deals with the trading on capital markets, the use of artificial intelligence, artificial neural networks, for modeling the behavior of stocks. The work contains a description of the capital markets, stock trading, methods of artificial intelligence. The main part of the thesis is the model for predicting the course and trend of shares, working in MATLAB, which serves as a support for trading decisions.

Tato diplomová práce se zabývá problematikou obchodování na komoditních trzích. Řešení problematiky spočívá ve využití umělé inteligence, konkrétně neuronových sítí, k technické analýze vývoje ceny vybrané komodity a snaze o co nejpřesnější predikci budoucího vývoje ceny pro podporu investičního rozhodování. Model neuronové sítě je vytvořen a použit pro predikci v programu MATLAB.

The thesis is concerned with optical character recognition from image data with different methods used for character classification. In the first theoretical part it focuses on explanation of all important parts of system for optical character recognition. The latter practical part of the thesis describes an example of image segmentation, the implementation of artificial neural networks for image recognition and create simple training set of data for the evaluation of the network. It also describes the process of training Tesseract tool and its implementation in a simple application EasyTessOCR for character recognition.