Esta tese apresenta um estudo para a utilização de Redes Neurais Artificiais (RNA) no processo de avaliação e ordenamento de alternativas de projetos de transporte. Partindo-se da ideia de que esse processo constitui-se em um padrão que pode ser captado pelas RNA, a verificação deste argumento foi feita selecionando-se um contexto de avaliação, definindo-se variáveis a serem consideradas no processo de avaliação, e criando-se estruturas de RNA para treinamento com base em outras avaliações já realizadas. Nesta pesquisa foram utilizados 180 \"Estudos de Casos\" recebidos de 32 Estados americanos. Esses dados serviram de entrada para um processo de aprendizagem utilizando-se o simulador \"Neural Planner 4.52\", que baseia-se em redes \"Multilayer Perceptron (MLP)\" e no treinamento em \"Backpropagation\". Várias redes foram treinadas para que fosse definida aquela com um melhor desempenho para o reconhecimento dos padrões existentes nesses casos apresentados. Os 486 experimentos demonstraram índices de acertos superiores a 92% que podem ser visualizados no programa computacional denominado \"EVALUATOR\", uma interface entre o simulador de RNA e usuários. Conclui-se, portanto, que as RNA podem reconhecer os padrões implícitos em avaliações anteriores e servem para avaliar e ordenar alternativas de outros projetos apresentados que pertençam ao mesmo contexto utilizado para treinamento. / This thesis presents a research aimed at the use of Artificial Neural Networks (ANN) for appraising and ranking transportation project alternatives. Based on the principle that this process of appraisal and ranking constitutes a pattern that can be perceived by ANN, the verification of this hypothesis was conducted selecting an evaluation context, defining variables to be considered in the process, and creating ANN structures for training based on other evaluation cases. In this research, 180 \"Case Studies\" from 32 American states were used. These data were used as input to a learning process using the simulator \"Neural Planner 4.52\", which is based on \"Multilayer Perceptron (MLP)\" networks and uses a \"Backpropagation\" training algorithm. Several networks were trained to obtain the one most capable of recognizing the patterns of the projects analyzed. More than 92% of the 486 experiments presented right indexes, as shown by a software called \"EVALUATOR\", a user interface between ANN simulator. The conclusion is that ANN can recognize the implicit patterns in previous evaluations and can be used to appraise and rank alternatives from other projects belonging to the same context used for the ANN training.

A irrigação é considerada uma das mais importantes tecnologias empregadas para aumentar a produtividade e permitir maior eficiência e estabilidade na produção agrícola. A sua adoção deve ser baseada na viabilidade técnica e econômica do projeto, obtida por meio da análise detalhada e cuidadosa dos fatores climáticos, agronômicos e econômicos envolvidos. O manejo eficiente pode ser definido como o uso criterioso dos recursos disponíveis para se atingir um determinado objetivo. No caso da irrigação, o manejo eficiente objetiva maximizar a produção vegetal com o menor custo possível, tanto em quesitos ambientais quanto econômicos. Para tanto, necessita-se do desenvolvimento de um sistema tecnologicamente eficiente, de reduzido custo e de facilidade de instalação e manutenção. Assim, uma rede inteligente de estações meteorológicas, capaz de monitorar o ambiente em tempo real, de adaptar-se aos diversos estágios fenológicos da planta, aos diversos solos e culturas e de comunicar-se entre si e com um servidor torna-se necessário. Este projeto teve como objetivo o desenvolvimento e emprego de uma rede de estações meteorológicas sem fio, de baixo custo e de fácil manutenção e implantação, para a determinação da evapotranspiração de referência, e do conteúdo de água no solo para o eficiente manejo de irrigação. Para que tal rede de sensores seja implantada, há a necessidade de que os módulos eletrônicos possuam microcontroladores de baixo consumo energético, uma vez que eles serão alimentados por baterias e painéis solares, e com capacidade para executar os algoritmos de inferência das variáveis de medida, de calibração e correção de tais medidas, de comunicação com os demais elementos da rede e de executar o controlador de irrigação, baseado em Redes Neurais Artificiais. A abordagem da inteligência artificial utilizada possui a capacidade de aprender e estimar parâmetros a partir de sua base de treinamento e das condições que a cercam. Além das capacidades do microcontrolador, o módulo dos sensores deverá possuir elementos para aferir a temperatura e a umidade do ambiente, a radiação solar, a temperatura e o conteúdo de água no solo, além de um módulo de comunicação sem fio. O sistema desenvolvido foi testado nas estufas do INCT-EI/ESALQ/USP manejando a irrigação da cultura do pimentão vermelho Capsicum annuum L. cv. All Big, durante o período de 25/01/2018 a 31/07/2018. Após o treinamento da rede neural artificial, o sistema desenvolvido determinou a evapotranspiração de referência com um coeficiente de determinação de 0,968, com um erro médio absoluto de 0,055 mm e com um valor-P de 1,02 10-45 para um intervalo de confiança de 95%. Sendo assim, verificou-se que a rede de estações meteorológicas desenvolvida é capaz de estimar a evapotranspiração de referência a partir de dados provenientes de sensores de reduzido custo financeiro e com dados meteorológicos faltantes. / Irrigation is considered one of the most important technologies used to increase productivity and to allow greater efficiency and stability in agricultural production. Its adoption must be based on technical and economic feasibility of the project, obtained by means of a detailed and careful analysis of the climatic, agronomic and economic factors involved. Efficient management can be defined as the judicious use of available resources to achieve a given goal. In the case of irrigation, efficient management aims to maximize plant production at the lowest possible cost, both in environmental and economic aspects. This requires the development of a technologically efficient system, which is low cost and ease of installation and maintenance. Thus, an intelligent network of weather stations capable of monitoring the environment in real time, adapting to the plant\'s various phenological stages, diverse soils and cultures and communicating with each other and with a server becomes necessary. This project aimed at the development and use of the wireless station network which is low cost, easy maintenance, and deployment for the determination of reference evapotranspiration and soil water content for efficient irrigation management. In order for such a sensor network to be implemented, there is a need for such modules to have low energy microcontrollers, since they will be powered by batteries and solar panels, and capable of performing the inference algorithms of the measurement, calibration, and correction of such measures, of communication with the other elements of the network and of executing the irrigation controller, based on Artificial Neural Networks. The artificial intelligence approach used has the capacity to learn and estimate parameters based on its training base and the conditions surrounding it. In addition to the capabilities of the microcontroller, the sensor module must have elements to measure ambient temperature and humidity, brightness, temperature and soil water content, as well as a wireless communication module. The developed system was tested in the greenhouses of INCT-EI/ESALQ/USP managing the irrigation of the red bell pepper crop Capsicum annuum L. cv. All Big, during the days of 01/25/2018 to 31/07/2010. After the artificial neural network training, the developed system determined reference evapotranspiration with a determination coefficient of 0.968, with an absolute mean error of 0.055 mm and a P-value of 1.02 10-45 for a confidence interval of 95%. Thus, it was verified that the developed weather-station network is able to estimate the reference evapotranspiration with low-cost sensors and with missing meteorological data.

A robótica móvel autônoma é uma área relativamente recente que tem como objetivo a construção de mecanismos capazes de executar tarefas sem a necessidade de um controlador humano. De uma forma geral, a robótica móvel defronta com três problemas fundamentais: mapeamento de ambientes, localização e navegação do robô. Sem esses elementos, o robô dificilmente poderia se deslocar autonomamente de um lugar para outro. Um dos problemas existentes nessa área é a atuação de robôs móveis em ambientes externos como parques e regiões urbanas, onde a complexidade do cenário é muito maior em comparação aos ambientes internos como escritórios e casas. Para exemplificar, nos ambientes externos os sensores estão sujeitos às condições climáticas (iluminação do sol, chuva e neve). Além disso, os algoritmos de navegação dos robôs nestes ambientes devem tratar uma quantidade bem maior de obstáculos (pessoas, animais e vegetações). Esta dissertação apresenta o desenvolvimento de um sistema de classificação da navegabilidade de terrenos irregulares, como por exemplo, ruas e calçadas. O mapeamento do cenário é realizado através de uma plataforma robótica equipada com um sensor laser direcionado para o solo. Foram desenvolvidos dois algoritmos para o mapeamento de terrenos. Um para a visualização dos detalhes finos do ambiente, gerando um mapa de nuvem de pontos e outro para a visualização das regiões próprias e impróprias para o tráfego do robô, resultando em um mapa de navegabilidade. No mapa de navegabilidade, são utilizados métodos de aprendizado de máquina supervisionado para classificar o terreno em navegável (regiões planas), parcialmente navegável (grama, casacalho) ou não navegável (obstáculos). Os métodos empregados foram, redes neurais artificais e máquinas de suporte vetorial. Os resultados de classificação obtidos por ambos foram posteriormente comparados para determinar a técnica mais apropriada para desempenhar esta tarefa / Autonomous mobile robotics is a recent research area that focus on the construction of mechanisms capable of executing tasks without a human control. In general, mobile robotics deals with three fundamental problems: environment mapping, robot localization and navigation. Without these elements, the robot hardly could move autonomously from a place to another. One problem of this area is the operation of the mobile robots in outdoors (e.g. parks and urban areas), which are considerably more complex than indoor environments (e.g. offices and houses). To exemplify, in outdoor environments, sensors are subjected to weather conditions (sunlight, rain and snow), besides that the navigation algorithms must process a larger quantity of obstacles (people, animals and vegetation). This dissertation presents the development of a system that classifies the navigability of irregular terrains, like streets and sidewalks. The scenario mapping has been done using a robotic platform equipped with a laser range finder sensor directed to the ground. Two terrain mapping algorithms has been devolped. One for environment fine details visualization, generating a point cloud map, and other to visualize appropriated and unappropriated places to robot navigation, resulting in a navigability map. In this map, it was used supervised learning machine methods to classify terrain portions in navigable (plane regions), partially navigable (grass, gravel) or non-navigable (obstacles). The classification methods employed were artificial neural networks and support vector machines. The classification results obtained by both were later compared to determine the most appropriated technique to execute this task

Este trabalho apresenta os resultados obtidos com o desenvolvimento de um sistema robusto como uma alternativa de reconhecimento da qualidade de vapor de álcool combustível e do poder calorífico do gás combustível GLP em um nariz eletrônico. Foram implementadas duas metodologias experimentais para a extração de atributos dos padrões de vapor de álcool combustível e de gás GLP. Na primeira abordagem de tratamento dos dados, foram usados um Sistema de Inferência Fuzzy (FIS), e dois algoritmos de treinamento de Redes Neurais Artificiais (RNA) para reconhecer padrões de vapor de álcool combustível: a Backpropagation e Learning Vector Quantization. A segunda abordagem para o tratamento dos dados foi desenvolver um sistema reconhecedor do poder calorífico do gás GLP robusto à perda aleatória de um dos sensores. Foram usados três sistemas. No primeiro foi implementada uma RNA para reconhecer todos os dados que simulavam a falha de um sensor aleatório. O resultado desse sistema foi de 97% de acertos. O segundo implementou sete RNA’s treinadas com subconjuntos dos dados de entrada, tais que seis RNA’s foram treinadas com um sensor diferente com falha; e a sétima RNA foi treinada com dados dos sensores sem falhas. O resultado desse sistema foi de 99% de acertos. O terceiro implementou uma Máquina de Comitê Estática Ensemble constituída de dez RNA’s em paralelo para resolver o problema. O resultado foi de 97% de acertos. As RNA’s tiveram melhores respostas que os FIS. Foram sugeridas algumas formas de implementação em hardware do sistema reconhecedor em sistemas pré-fabricados com DSP’s e micro-controladores. / This work shows the results of a robust system development as an alternative to recognize the quality of an alcohol fuel vapor sample and Liquid Petrol Gas (LPG) heat power in an electric nose. Two experimental methodologies were implemented to extract the features of alcohol fuel vapor and LPG gas patterns. The first approach to process the data used an Fuzzy Inference System (FIS) and two training algorithms of Artificial Neural Networks (ANN) to recognize alcohol fuel vapor patterns: Backpropagation and Learning Vector Quantization. The second approach consists of process data to develop an LPG heat power recognizing system robust to one-random-sensor-loss. Three systems were used. The first implemented an ANN to recognize all data that simulated the failure of a random sensor. This system had 97% of right responses. The second implemented seven ANN’s trained with input data subsets, such that six ANN’s were trained with a different failure sensor, and the seventh ANN was trained with data of all sensors without failure. This system had 99% of right responses. The third implemented an Ensemble Static Learning Machine containing ten parallel RNA’s to solve the problem. The result were 97% of right responses. RNA’s had better results than FIS. Some ways of hardware implementation of the recognizing system were suggested in DSP and micro-controllers pre-built systems.

Esta tese consiste em desenvolver um sistema de identificação e classificação de falhas em motores de indução trifásico. As falhas analisadas foram simuladas em laboratório e envolvem problemas elétricos, como curto-circuito no estator, e problemas mecânicos, como barras quebradas no rotor. O sistema computacional proposto é formado pela transformada discreta wavelet, pelo cálculo de variáveis estatísticas e por redes neurais artificiais. A partir dos sinais elétricos da corrente do estator, a transformada wavelet produz os coeficientes característicos das falhas, os quais são usados no cálculo das variáveis estatísticas, como a média, root mean square, skewness e kurtosis. Estes valores são transmitidos como dados de entrada para as redes neurais que identificam as falhas e classificam a natureza das mesmas. Por fim, resultados obtidos visam validar a metodologia sugerida, que buscou nos sistemas inteligentes soluções eficazes para diagnosticar falhas em máquinas elétricas. / This thesis consists in developing a system for the identification and classification of faults in three-phase electric motors. The faults were analyzed and simulated in the laboratory and involve electrical problems, such as short circuit in the stator, and mechanical problems, such as broken rotor bars. The proposed computer system is formed by discrete wavelet transform, by calculation of statistical variables and for artificial neural networks. From the electrical signals of the stator current, the wavelet transform produces characteristic coefficients of faults, which are extracted by calculating of statistics variables, such as mean, root mean square, skewness and kurtosis. These values are passed as input to the neural networks that identify faults and the severity of it. Finally, results aimed at validating the methodology suggested that sought effective solutions in intelligent systems to diagnose faults in electrical machines.

Este trabalho apresenta uma alternativa ao controle vetorial de motores de indução, sem a utilização de sensores para realimentação da velocidade mecânica do motor. Ao longo do tempo, diversas técnicas de controle vetorial têm sido propostas na literatura. Dentre elas está a técnica de controle por orientação de campo (FOC), muito utilizada na indústria e presente também neste trabalho. A principal desvantagem do FOC é a sua grande sensibilidade às variações paramétricas da máquina, as quais podem invalidar o modelo e as ações de controle. Nesse sentido, uma estimativa correta dos parâmetros da máquina, torna-se fundamental para o acionamento. Este trabalho propõe o desenvolvimento e implementação de um estimador baseado em um sistema de inferência neuro-fuzzy adaptativo (ANFIS) para o controle de velocidade do motor de indução trifásico em um acionamento sem sensores. Pelo fato do acionamento em malha fechada admitir diversas velocidades de regime estacionário para o motor, uma nova metodologia de treinamento por partição de frequência é proposta. Ainda, faz-se a validação do sistema utilizando a orientação de campo magnético no referencial de campo de entreferro da máquina. Simulações para avaliação do desempenho do estimador mediante o acionamento vetorial do motor foram realizadas utilizando o programa Matlab/Simulink. Para a validação prática do modelo, uma bancada de testes foi implementada; o acionamento do motor foi realizado por um inversor de frequência do tipo fonte de tensão (VSI) e o controle vetorial, incluindo o estimador neuro-fuzzy, foi realizado pelo pacote de tempo real do programa Matlab/Simulink, juntamente com uma placa de aquisição de dados da National Instruments. / This work presents an alternative sensorless vector control of induction motors. Several techniques for induction motor control have been proposed in the literature. Among these is the field oriented control (FOC), strongly used in industries and also in this work. The main drawback of the FOC technique is its sensibility to deviations of the parameters of the machine, which can deteriorate the control actions. Therefore, an accurate determination of the machines parameters is mandatory to the drive system. This work proposes the development of an adaptive neuro-fuzzy inference system (ANFIS) estimator to control the angular speed of a three-phase induction motor in a sensorless drive. In a closed loop configuration, several speed commands can be imposed to the motor. Thus, a new frequency partition training of ANFIS is proposed. Moreover, the ANFIS speed estimator is validated in a magnetizing flux oriented control scheme. Simulations to evaluate the performance of the estimator considering the vector drive system were done by the Matlab/Simulink. To determine the benefits of the proposed model a practical system was implemented using a voltage source inverter (VSI) and the vector control including the ANFIS estimator, carried out by the Real Time Toolbox from Matlab/Simulink and a data acquisition card from National Instruments.

Cette thèse présente une méthode pour désagréger les appareils électriques dans le profil des bâtiments résidentiels de charge. Au cours des dernières années, la surveillance de l’énergie a obtenu beaucoup de popularité dans un environnement privé et industriel. Avec des algorithmes de la désagrégation, les données mesurées à partir de soi-disant compteurs intelligents peuvent être utilisés pour fournir de plus amples informations de la consommation d’énergie. Une méthode pour recevoir ces données est appelé non-intrusifs charge identification. La majeure partie de la thèse peut être divisée en trois parties. Dans un premier temps, un système de mesure propre a été développé et vérifié. Avec ce système, les ensembles de données réelles peuvent être générés pour le développement et la vérification des algorithmes de désagrégation. La deuxième partie décrit le développement d’un détecteur de flanc. Différentes méthodes sont présentées et évaluées, avec lequel les temps de commutation des appareils peuvent être détectés dans le profil de la charge. La dernière partie décrit un procédé de classification. Différents critères sont utilisés pour la classification. Le classificateur reconnaît et étiquette les appareils individuels de la courbe de charge. Pour les classifications différentes structures de réseaux de neurones artificiels sont comparés. / This thesis presents a method to disaggregate electrical appliances in the load profile of residential buildings. In recent years, energy monitoring has obtained significantly popularity in private and industrial environment. With algorithms of the disaggregation, the measured data from so-called smart meters can be used to provide more information of the energy usage. One method to receive these data is called non-intrusive appliance load monitoring.The main part of the thesis can be divided into three parts. At first, an own measurement system was developed and verified. With that system, real data sets can be generated for the development and verification of the disaggregation algorithms. The second part describes the development of an event detector. Different methods are presented and evaluated, with which the switching times of the appliances can be detected in the load profile. The last part describes a classification method. Different features are used for the classification. The classifier recognizes and labels the individual appliances in the load profile. For the classification different structures of artificial neural network (ANN) are compared.

Face à la limitation de la modélisation paramétrique, nous avons proposé dans cette thèse une procédure standard pour combiner les données reçues a partir de Réseaux de capteurs sans fils (WSN) pour modéliser a l'aide de Réseaux de Neurones Artificiels (ANN). Des expériences sur la modélisation thermique ont permis de démontrer que la combinaison de WSN et d'ANN est capable de produire des modèles thermiques précis. Une nouvelle méthode de formation "Multi-Pattern Cross Training" (MPCT) a également été introduite dans ce travail. Cette méthode permet de fusionner les informations provenant de différentes sources de données d'entraînements indépendants (patterns) en un seul modèle ANN. D'autres expériences ont montré que les modèles formés par la méthode MPCT fournissent une meilleure performance de généralisation et que les erreurs de prévision sont réduites. De plus, le modèle de réseau neuronal basé sur la méthode MPCT a montré des avantages importants dans le multi-variable Model Prédictive Control (MPC). Les simulations numériques indiquent que le MPC basé sur le MPCT a surpassé le MPC multi-modèles au niveau de l'efficacité du contrôle. / A Wireless Sensor Network (WSN) consisting of autonomous sensor nodes can provide a rich stream of sensor data representing physical measurements. A well built Artificial Neural Network (ANN) model needs sufficient training data sources. Facing the limitation of traditional parametric modeling, this paper proposes a standard procedure of combining ANN and WSN sensor data in modeling. Experiments on indoor thermal modeling demonstrated that WSN together with ANN can lead to accurate fine grained indoor thermal models. A new training method "Multi-Pattern Cross Training" (MPCT) is also introduced in this work. This training method makes it possible to merge knowledge from different independent training data sources (patterns) into a single ANN model. Further experiments demonstrated that models trained by MPCT method shew better generalization performance and lower prediction errors in tests using different data sets. Also the MPCT based Neural Network Model has shown advantages in multi-variable Neural Network based Model Predictive Control (NNMPC). Software simulation and application results indicate that MPCT implemented NNMPC outperformed Multiple models based NNMPC in online control efficiency.

Climatic variations, together with large-scale environmental forces and human development affect the quality of coastal recreational waters, creating potential risks to human health. These environmental forces, including increased temperature and precipitation, often promote specific vector-borne diseases in the Caribbean and Gulf of Mexico. Human activities affect water quality through discharges from urban areas, including nutrient and other pollutants derived from wastewater systems. Both water quality of recreational beaches and vector-borne diseases can be better managed by understanding their relationship with local environmental forces. I evaluated how changes in vector-borne diseases and poor recreational water quality were related to specific environmental factors through the application of satellite-derived observations, field observations, and public health records. Variability in dengue fever incidence rates in coastal towns of the Yucatan Peninsula (Mexico) was evaluated with respect to environmental factors in Chapter Two. Correlations between fecal indicator bacteria concentrations (i.e., culturable enterococci) at Escambron Beach (San Juan, Puerto Rico, USA) and regional environmental factors are discussed in Chapter Three. Predictions of dengue fever occurrences in the Yucatan Peninsula were tested using a nonlinear approach (i.e., Artificial Neural Networks) and are presented in Chapter Four. The Artificial Neural Network (ANN) model was also used to predict culturable enterococci concentration exceeding safe recreational water quality standards in Escambron Beach and results are presented in Chapter Five. Environmental factors assessed to understand their influence on dengue fever occurrences and culturable enterococci concentrations included precipitation, mean sea level (MSL), air temperatures (e.g., maximum, minimum, and average), humidity, and satellite-derived sea surface temperature (SST), dew point, direct normal irradiance (DNI), and turbidity. These factors were combined with demographic data (e.g., population size) and compared with dengue fever incidence rates and culturable enterococci concentration using linear and nonlinear statistical approaches. Dengue incidence rates in Yucatan (Mexico) generally increased in July/August and decreased during November/December. A linear regression model showed that previous dengue incidence rates explained 89% of dengue fever variability (p < 0.05). Dengue incidence two weeks prior (previous incidence) influences future outbreaks by allowing the virus to continue propagating. Yet dengue incidence was best explained by precipitation, minimum air temperature, humidity, and SST (p < 0.05). Dengue incidence variability was best explained by SST and minimum air temperature in our study region (r = 0.50 and 0.48, respectively). Increases in SST preceded increased dengue incidence rate by eight weeks. Dengue incidence time series were positively correlated to SST and minimum air temperature anomalies. This is related to the virus and mosquito behavior. Including oceanographic variables among environmental factors in the model improved modelling skill of dengue fever in Mexico. Chapter Three shows that precipitation, MSL, DNI, SST, and turbidity explained some of the enterococci variation in Escambron Beach surface waters (AIC = 26.76; r = 0.20). Variation in these parameters preceded increased culturable enterococci concentrations, with lags spanning from 24 h up to 11 days. The highest influence on culturable enterococci was precipitation between 480 mm–900 mm. Rainy events often result in overflows of sewage systems and other non-point sources near Escambron Beach in Puerto Rico. A significant decrease in culturable enterococci concentrations was observed during increased irradiance (r = -0.24). This may be due to bacterial inactivation. Increased culturable enterococci concentrations were significantly associated with higher turbidity daily anomalies (r = 0.25), in part because bacteria were protected from light inactivation. Increased culturable enterococci concentrations were related to warmer SST anomalies (r = 0.12); this is likely due to increased bacterial activity and reproduction. Higher culturable enterococci concentrations were also significantly correlated to medium to high values of dew point daily anomalies (r = 0.19). A significant decrease in culturable enterococci during higher daily MSL anomalies (r = -0.19) is possibly due to dilution of bacteria in beach waters, whereas during lower MSL anomalies the back-washing promotes increased bacteria concentrations through mixing from sediments. These environmental variables improve our understanding of the ecology of these bacteria over time. The predictive capability increases by including more than one environmental variable. Chapter Four explains a predictive model of dengue fever occurrences in San Juan, Puerto Rico (1994–2012), and Yucatan (2007–2012). The model was modified to predict dengue fever outbreak occurrences for two population segments: population at risk of infection (i.e., < 24 years old) and vulnerable population (i.e., < 5 years old and > 65 years old). There were a total of four predictive models, two sets for each location using the specified population segments. Model predictions showed previous dengue cases, minimum air temperature, date, and population size as the factors with the most influence to predict dengue fever outbreak occurrences in Mexico. Previous dengue cases, maximum air temperature, date, and population size were the most influential factors for San Juan, Puerto Rico. The models showed an accuracy around 50% and a predictive capability of 70%. These environmental and demographic variables are important primary predictors for dengue fever outbreaks in Puerto Rico and Mexico. Chapter Five shows the application of the ANNs model to predict culturable enterococci exceedance based on the U.S. Environmental Protection Agency (U.S. EPA) Recreational Water Quality Criteria (RWQC) at Escambron Beach, San Juan, Puerto Rico. The model identified DNI, turbidity, 48 h cumulative precipitation, MSL, and SST as the most influential factors to predict enterococci concentration exceedance, based on the U.S. EPA RWQC at Escambron Beach from 2005–2014. The model showed an accuracy of 76%, with a predictive capability greater than 60%, which is higher than linear models. Results showed the applicability of remote sensing data and ANNs to predict recreational water quality and help improve early warning system and public health. This work helps to better understand complex relationships between climatic variations and public health issues in tropical coastal areas and provides information that can be used by public health practitioners.

Historical water quality databases from two South Australian drinking water reservoirs were used, in conjunction with various computational modelling methods for the ordination, clustering and forecasting of complex ecological data. Techniques used throughout the study were: Kohonen artificial neural networks (KANN) for data categorisation and the discovery of patterns and relationships, recurrent supervised artificial neural networks (RANN) for knowledge discovery and forecasting of algal dynamics and hybrid evolutionary algorithms (HEA) for rule-set discovery and optimisation for forecasting algal dynamics. These methods were combined to provide an integrated approach to the analysis of algal populations including interactions within the algal community and with other water quality factors, which results in improved understanding and forecasting of algal dynamics. The project initially focussed on KANN for the patternising and classification of the historical data to reveal links between the physical, chemical and biological components of the reservoirs. This offered some understanding of the system and relationships being considered for the construction of the forecasting models. Specific investigations were performed to examine past conditions and the impacts of different management regimes, as well as to discover sets of conditions that correspond with specific algal functional groups. RANN was then used to build models for forecasting both Chl-a and the main nuisance species, Anabaena, up to 7 days in advance. This method also provided sensitivity analyses to demonstrate the relationship between input and output variables by plotting the reaction of the output to variations in the inputs. Initially one year from the data set was selected for the testing of a model, as per the split-sample technique. To further test the models, it was later decided to select several years for testing to ensure the models were useful under changed conditions, and that test results were not misleading regarding the models true capabilities. RANN were firstly used to create reservoir specific or ad-hoc models. Later, the models were trained with the merged data sets of both reservoirs to create one model that could be applied to either reservoir. Another method of forecasting was trialled and compared to RANN. HEA was found to be equal or superior to RANN in predictive power, also allowed sensitivity analysis and provided an explicit, portable rule set. The HEA rule sets were initially tested on selected years of data, however to fully demonstrate the models potential, a process for k-fold cross-validation was developed to test the rule-set on all years of data. To further extend the applicability of the HEA rule-set; the idea of rule-based agents for specific lake ecosystem categories was examined. The generality of a rule-based agent means that, after successful validation on several lakes from one category, the agent could then be applied to other water bodies from within that category that had not been involved in the training process. The ultimate test of the rule-based agent for the warm monomictic and eutrophic lake ecosystem category was to be applied to a real-time monitoring and forecasting situation. The agent was fed with online, real-time data from a reservoir that belonged to the same ecosystem category but was not used in the training process. These preliminary experiments showed promising results. It can be concluded that the concept of rulebased agents will facilitate real-time forecasting of algal blooms in drinking water reservoirs provided on-line monitoring of relevant variables has been implemented. Contributions of this research include: (1) to offer insight into the capabilities of 3 kinds of computational modelling techniques applied to complex water quality data, (2) novel applications of KANN including the division of data into separate management periods for comparison of management efficiency, (3) to both qualitatively and quantitatively elucidate relationships between water quality parameters, (4) research toward the development of a forecasting tool for algal abundance 7 days in advance that could be generic for a particular lake ecosystem category and implemented in real-time, and (5) to suggest a thorough testing method for such models (k-fold cross validation). / http://proxy.library.adelaide.edu.au/login?url= http://library.adelaide.edu.au/cgi-bin/Pwebrecon.cgi?BBID=1331584 / Thesis (Ph.D.) -- University of Adelaide, School of Earth and Environmental Sciences, 2008