[pt] Esta dissertação investiga a utilização de um sistema Neuro-Fuzzy Hierárquico para previsão de séries e a extração de regras fuzzy em aplicações de Mineração de Dados. O objetivo do trabalho foi estender o modelo Neuro- Fuzzy Hierárquico BSP para a classificação de registros e a previsão de séries temporais. O processo de classificação de registros no contexto de Mineração de Dados consiste na extração de regras de associação que melhor caracterizem, através de sua acurácia e abrangência, um determinado grupo de registros de um banco de dados (BD). A previsão de séries temporais, outra tarefa comum em Mineração de Dados tem como objetivo prever o comportamento de uma série temporal no instante t+k (k ? 1).O trabalho consistiu de 5 etapas principais: elaborar um survey dos principais sistemas e modelos mais utilizados nas aplicações de Mineração de Dados; avaliar o desempenho do sistema NFHB original em aplicações de Mineração de Dados; desenvolver uma extensão do modelo NFHB dedicado à classificação de registros em uma BD; desenvolver um novo modelo híbrido Neuro-Fuzzy Genético para o ajuste automático dos parâmetros do sistema dedicado a previsão de séries temporais; e o estudo dos casos. O estudo da área resultou num survey sobre os principais modelos para Mineração de Dados. São apresentados os modelos mais utilizados em tarefas de classificação e extração de regras tais como: redes neurais, árvores de decisão crisp e fuzzy, algoritmos genéticos, estatística e sistemas neuro-fuzzy. Na etapa de avaliação do modelo NFHB original, foi verificado que além do tradicional aprendizado dos parâmetros, comuns às redes neurais e aos sistemas neuro-fuzzy, o modelo possui as seguintes aracterísticas: aprendizado da estrutura, a partir do uso de particionamentos recursivos; número maior de entradas que o habitualmente encontrado nos sistemas neuro-fuzzy; e regras com hierarquia, características adequadas para as aplicações de Mineração de Dados. Entretanto, o processo de extração de regras e a seleção de atributos não são adequados para este tipo de aplicação, assim como a excessiva complexidade da parametrização do modelo para aplicações de previsão de séries temporais. Uma extensão ao modelo NFHB original foi então proposta para aplicações de classificação de registros no contexto da Mineração de Dados onde se têm como objetivo principal a extração de informação em forma de regras interpretáveis. Foi necessário modificar a seleção de atributos e o processo original de extração de regras. O sistema fuzzy do tipo Takagi-Sugeno do modelo NFHB original fornece regras inadequadas do ponto de vista da Mineração de Dados. O novo modelo NFHB, dotado das modificações necessárias, mostrou um ótimo desempenho na extração de regras fuzzy válidas que descrevem a informação contida no banco de dados. As medidas de avaliação normalmente usadas para analisar regras crisp (Se x1 é <14.3 e...), como abrangência e acurácia, foram modificadas para poderem ser aplicadas ao caso de avaliação das regras fuzzy (Se x1 é Baixo e..) extraídas pelo sistema NFHB após da fase de aprendizado. A quantidade e a qualidade das regras extraídas é um ponto fundamental dos sistemas voltados para aplicações de Mineração de Dados, que buscam sempre obter o menor número de regras e da maior qualidade possível. Nesse sentido, o processo de seleção das características de entrada foi alterado para evitar particionamentos excessivos, ou seja regras desnecessárias. Foram implementadas duas estratégias de seleção (Fixa e Adaptativa) em função de diferentes medidas de avaliação como a Entropia e o método de Jang. Um novo modelo híbrido neuro-fuzzy genético para previsão de séries temporais foi criado para resolver o problema da excessiva complexidade de parametrização do sistema, o qual conta com mais de 15 parâmetros.Foi proposto um novo modelo híbrido neuro-fuzzy genético capaz de evoluir e obter um conjunto de parâmetros adequado par / [en] This dissertation investigates the use of a Neuro-Fuzzy Hierarchical system for time series forecasting and fuzzy rule extraction for Data Mining applications. The objective of this work was to extend the Neuro-Fuzzy BSP Hierarchical model for the classification of registers and time series forecasting. The process of classification of registers in the Data Mining context consists of extracting association rules that best characterise, through its accuracy and coverage measures, a certain group of registers of database (DB). The time series forecasting other common task in Data Mining, has a main objective to foresee the behavior of a time series in the instant t+k (k>=1). The work consisted of 5 main stages: to elaborate a survey of the main systems and the most common models in Data Mining applications; to evaluate the performance of the original NFHB system in Data Mining applicatons; to develop an extension of the NFHB model dedicated to the classification of registers in a DB; to develop a new Neuro-Fuzzy Genetic hybrid model for the automatic adjustment of the parameters of the system for time series forecasting applicatons; and the case estudies. The study of the area resulted in a survey of the main Data Mining models. The most common methods used in Data Mining application are presented such as: neural nets, crisp and fuzzy decision trees, genetic algorithms, statistics and neuro-fuzzy systems. In the stage of evaluation of the original NFHB model, it verified that besides the traditional learning of the parameters, common to the neural nets and the neuro-fuzzy systems, the model possesses the following characteristics: learning of the structure; recursive partitioning; larger number of inputs than usually found on the neuro-fuzzy systems; rule with hierarchy; which are characteristics adapted for Data Mining applications. However the rule extraction process and attributes selection are not appropriate for this type of applications, as well as the excessive complexity of the tuning of the model for time series forecasting applicatons. An extension of the original NFHB model was then proposed for applicatons of classification of registers in the Data Mining context, where the main objective in the extraction of information in form of interpratable rules. It was necessary to modify the attributes selection and the original rule extraction process. The Takagi-Sugeno fuzzy system of the original NFHB model supplies inadequate rules, from the Data Mining point of view. The new NFHB models, endowed with necessary modifications, showed good performance in extracting valid fuzzy rules that describe the information contained in the database. The evaluation metrics, usually used to analyse crips rules (If x1 is <14.3 and), as coverage and accuracy, were modified to be applied to the evaluation of the fuzzy rules (If x1 is Low and) extracted from the NFHB system after the learning process. The amount and quality of the extracted rules are important points of the systems dedicated for Data Mining applicatons, where the target is to obtain the smallest number of rules and of the best quality. In that sense, the input selection strategies were implemented (Static and Adaptive), using different evaluation measures as Entropy and the jang algorithm. A new genetic neuro-fuzzy hybrid model for time series forecasting was created to solve the problem of the excessive complexity of the model tuning, which comprises more than 15 parameters. A new model wes proposed, a genetic neuro-fuzzy hybrid, model capable to develop and to obtain an appropriate set of parameters for the forecasting of time series. The new hybrid, model capable to develop and to obtain an appropriate set of parameters for the forecasting of time series. The new hybrid model presented good results with different types of series. A tool based on the NFHB model was developed for classification and forecasting applications. Th

This research is focused on the integration problem of process planning and scheduling in steel heat treatment operations environment using artificial intelligent techniques that are capable of dealing with such problems. This work addresses the issues involved in developing a suitable methodology for scheduling heat treatment operations of steel. Several intelligent algorithms have been developed for these propose namely, Genetic Algorithm (GA), Sexual Genetic Algorithm (SGA), Genetic Algorithm with Chromosome differentiation (GACD), Age Genetic Algorithm (AGA), and Mimetic Genetic Algorithm (MGA). These algorithms have been employed to develop an efficient intelligent algorithm using Algorithm Portfolio methodology. After that all the algorithms have been tested on two types of scheduling benchmarks. To apply these algorithms on heat treatment scheduling, a furnace model is developed for optimisation proposes. Furthermore, a system that is capable of selecting the optimal heat treatment regime is developed so the required metal properties can be achieved with the least energy consumption and the shortest time using Neuro-Fuzzy (NF) and Particle Swarm Optimisation (PSO) methodologies. Based on this system, PSO is used to optimise the heat treatment process by selecting different heat treatment conditions. The selected conditions are evaluated so the best selection can be identified. This work addresses the issues involved in developing a suitable methodology for developing an NF system and PSO for mechanical properties of the steel. Using the optimisers, furnace model and heat treatment system model, the intelligent system model is developed and implemented successfully. The results of this system were exciting and the optimisers were working correctly.

A new adaptive learning Artificial Immune System (AIS) based committee machine is developed in this thesis. The new proposed approach efficiently tackles the general problem of clustering high-dimensional data. In addition, it helps on deriving useful decision and results related to other application domains such classification and prediction. Artificial Immune System (AIS) is a branch of computational intelligence field inspired by the biological immune system, and has gained increasing interest among researchers in the development of immune-based models and techniques to solve diverse complex computational or engineering problems. This work presents some applications of AIS techniques to health problems, and a thorough survey of existing AIS models and algorithms. The main focus of this research is devoted to building an ensemble model integrating different AIS techniques (i.e. Artificial Immune Networks, Clonal Selection, and Negative Selection) for classification applications to achieve better classification results. A new AIS-based ensemble architecture with adaptive learning features is proposed by integrating different learning and adaptation techniques to overcome individual limitations and to achieve synergetic effects through the combination of these techniques. Various techniques related to the design and enhancements of the new adaptive learning architecture are studied, including a neuro-fuzzy based detector and an optimizer using particle swarm optimization method to achieve enhanced classification performance. An evaluation study was conducted to show the performance of the new proposed adaptive learning ensemble and to compare it to alternative combining techniques. Several experiments are presented using different medical datasets for the classification problem and findings and outcomes are discussed. The new adaptive learning architecture improves the accuracy of the ensemble. Moreover, there is an improvement over the existing aggregation techniques. The outcomes, assumptions and limitations of the proposed methods with its implications for further research in this area draw this research to its conclusion.

Student Number : 9702018F - MSc(Eng) Dissertation - School of Electrical and Information Engineering - Faculty of Engineering and the Built Environment / Due to the volatile nature of the world economies, investing is crucial in ensuring an individual is prepared for future financial necessities. This research proposes an application, which employs computational intelligent methods that could assist investors in making financial decisions. This system consists of 2 components. The Forecasting Component (FC) is employed to predict the closing index price performance. Based on these predictions, the Stock Quantity Selection Component (SQSC) recommends the investor to purchase stocks, hold the current investment position or sell stocks in possession. The development of the FC module involved the creation of Multi-Layer Perceptron (MLP) as well as Radial Basis Function (RBF) neural network classifiers. TCategorizes that these networks classify are based on a profitable trading strategy that outperforms the long-term “Buy and hold” trading strategy. The Dow Jones Industrial Average, Johannesburg Stock Exchange (JSE) All Share, Nasdaq 100 and the Nikkei 225 Stock Average indices are considered. TIt has been determined that the MLP neural network architecture is particularly suited in the prediction of closing index price performance. Accuracies of 72%, 68%, 69% and 64% were obtained for the prediction of closing price performance of the Dow Jones Industrial Average, JSE All Share, Nasdaq 100 and Nikkei 225 Stock Average indices, respectively. TThree designs of the Stock Quantity Selection Component were implemented and compared in terms of their complexity as well as scalability. TComplexity is defined as the number of classifiers employed by the design. Scalability is defined as the ability of the design to accommodate the classification of additional investment recommendations. TDesigns that utilized 1, 4 and 16 classifiers, respectively, were developed. These designs were implemented using MLP neural networks, RBF neural networks, Fuzzy Inference Systems as well as Adaptive Neuro-Fuzzy Inference Systems. The design that employed 4 classifiers achieved low complexity and high scalability. As a result, this design is most appropriate for the application of concern. It has also been determined that the neural network architecture as well as the Fuzzy Inference System implementation of this design performed equally well.

Pesquisas científicas no campo da engenharia de reabilitação estão proporcionando cada vez mais mecanismos que visam ajudar pessoas portadoras de alguma deficiência física a executar tarefas simples do dia-a-dia. Com isso em mente, esse trabalho tem a finalidade de desenvolver um sistema que utiliza sinais musculares e redes neuro-fuzzy para a caracterização de determinados movimentos de um braço humano, com o objetivo de possibilitar futuramente a integração em sistemas de reabilitação. Ensaios preliminares demonstraram que para a caracterização de movimentos simples realizados por um braço humano, o uso exclusivo de técnicas simples de processamento de sinal é suficiente, como a utilização do valor rms. No entanto, para a caracterização de movimentos complexos é necessário um processamento mais robusto do sinal. Para isso foi desenvolvido um sistema experimental que adquire, através de um eletromiógrafo (EMG) de 8 canais, o sinal mioelétrico com eletrodos de superfície posicionados em lugares estratégicos do braço. O sinal é adquirido utilizando como estímulo um modelo virtual que demonstra ao usuário os movimentos do segmento mão-braço que devem ser executados de forma aleatória. Finalmente, com o uso de uma rede neuro-fuzzy, que possibilita a distinção tanto de movimentos simples como de movimentos compostos, se adaptando a diferentes usuários, os movimentos executados foram caracterizados em 12 movimentos distintos, previamente definidos, com uma taxa de acerto médio de 65%. / The scientific researches in the field of rehabilitation engineering are increasingly providing mechanisms to help people with a disability to perform simple tasks of day-to-day. With that in mind, this work aims to develop an experimental robotic prosthesis in order to implement, in the same, a control system that uses muscle signals and neuro-fuzzy networks for characterization of certain movements of a human arm, in order to enable further integration in rehabilitation systems. Preliminary tests showed that for the characterization of simple movements performed by a human arm, the exclusive use of simple techniques of signal processing is sufficient, as the use of the rms value. However, for the characterization of complex movements is required a more robust signal processing. For this was developed an experimental system that acquires through an electromyography (EMG) of 8 channels, the myoelectric signal with surface electrodes positioned in strategic places of the arm. The acquired signal uses, as a stimulus, a virtual model that demonstrates the hand-arm segment movements to be executed by the user at random. Finally, through a neuro-fuzzy network, which enables the distinction of both simple and compound movements, self-adapting to different users, the movements performed were characterized in 12 distinct movements, previously defined, with an average accuracy of 65%.

This thesis deals with enhancing the performance of Fuzzy Logic (FL) based PI controllers for High Voltage Direct Current Transmission Systems (HVDC) by optimizing the key parameters i.e. membership functions (MFs) and fuzzy rule base in the controllers design. In the first part of the thesis, an adaptive Fuzzy PI controller is designed and the effect of various MF shapes, widths and distribution on the performance of a FL controlled HVDC system under different system conditions is studied with the aim of selecting a MF which minimizes the total control error. Simulated results show that the shape, width and distribution of a MF influences the performance of the FL controller and concludes that nonlinear MFs (i.e. Gaussian) offer a more better choice than linear (i.e. Triangular) MFs as the former provides a smoother transition at the switching points and thus propose a better controller. In the second part of the thesis, a Neuro-Fuzzy (NF) controller to update the fuzzy rule base with changing system conditions is proposed, which in turn adjusts the PI gains of a conventional PI controller. Results from simulations illustrate the potential of the proposed control scheme as the NF controller successfully adapts to different system conditions and is able to minimize the total current error. / UOIT

There are many situations where a linguistic description of complex phenomena allows better assessments. It is well known that the assessment of water quality continues depending heavily upon subjective judgments and interpretation, despite the huge datasets available nowadays. In that sense, the aim of this study has been to introduce intelligent linguistic operations to analyze databases, and produce self interpretable water quality indicators, which tolerate both imprecision and linguistic uncertainty. Such imprecision typically reflects the ambiguity of human thinking when perceptions need to be expressed. Environmental management concepts such as: "water quality", "level of risk", or "ecological status" are ideally dealt with linguistic variables. In the present Thesis, the flexibility of computing with words offered by fuzzy logic has been considered in these management issues. Firstly, a multipurpose hierarchical water quality index has been designed with fuzzy reasoning. It integrates a wide set of indicators including: organic pollution, nutrients, pathogens, physicochemical macro-variables, and priority micro-contaminants. Likewise, the relative importance of the water quality indicators has been dealt with the analytic hierarchy process, a decision-aiding method. Secondly, a methodology based on a hybrid approach that combines fuzzy inference systems and artificial neural networks has been used to classify ecological status in surface waters according to the Water Framework Directive. This methodology has allowed dealing efficiently with the non-linearity and subjective nature of variables involved in this classification problem. The complexity of inference systems, the appropriate choice of linguistic rules, and the influence of the functions that transform numerical variables into linguistic variables have been studied. Thirdly, a concurrent neuro-fuzzy model based on screening ecological risk assessment has been developed. It has considered the presence of hazardous substances in rivers, and incorporates an innovative ranking and scoring system, based on a self-organizing map, to account for the likely ecological hazards posed by the presence of chemical substances in freshwater ecosystems. Hazard factors are combined with environmental concentrations within fuzzy inference systems to compute ecological risk potentials under linguistic uncertainty. The estimation of ecological risk potentials allows identifying those substances requiring stricter controls and further rigorous risk assessment. Likewise, the aggregation of ecological risk potentials, by means of empirical cumulative distribution functions, has allowed estimating changes in water quality over time. The neuro-fuzzy approach has been validated by comparison with biological monitoring. Finally, a hierarchical fuzzy inference system to deal with sediment based ecological risk assessment has been designed. The study was centered in sediments, since they produce complementary findings to water quality analysis, especially when temporal trends are required. Results from chemical and eco-toxicological analyses have been used as inputs to two parallel inference systems which assess levels of contamination and toxicity, respectively. Results from both inference engines are then treated in a third inference engine which provides a final risk characterization, where the risk is provided in linguistic terms, with their respective degrees of certitude. Inputs to the risk system have been the levels of potentially toxic substances, mainly metals and chlorinated organic compounds, and the toxicity measured with a screening test which uses the photo-luminescent bacteria Vibrio fischeri. The Ebro river basin has been selected as case study, although the methodologies here explained can easily be applied to other rivers. In conclusion, this study has broadly demonstrated that the design of water quality indexes, based on fuzzy logic, emerges as suitable and alternative tool to support decision makers involved in effective sustainable river basin management plans. / Existen diversas situaciones en las cuales la descripción en términos lingüísticos de fenómenos complejos permite mejores resultados. A pesar de los volúmenes de información cuantitativa que se manejan actualmente, es bien sabido que la gestión de la calidad del agua todavía obedece a juicios subjetivos y de interpretación de los expertos. Por tanto, el reto en este trabajo ha sido la introducción de operaciones lógicas que computen con palabras durante el análisis de los datos, para la elaboración de indicadores auto-interpretables de calidad del agua, que toleren la imprecisión e incertidumbre lingüística. Esta imprecisión típicamente refleja la ambigüedad del pensamiento humano para expresar percepciones. De allí que las variables lingüísticas se presenten como muy atractivas para el manejo de conceptos de la gestión medioambiental, como es el caso de la "calidad del agua", el "nivel de riesgo" o el "estado ecológico". Por tanto, en la presente Tesis, la flexibilidad de la lógica difusa para computar con palabras se ha adaptado a diversos tópicos en la gestión de la calidad del agua. Primero, se desarrolló un índice jerárquico multipropósito de calidad del agua que se obtuvo mediante razonamiento difuso. El índice integra un extenso grupo de indicadores que incluyen: contaminación orgánica, nutrientes, patógenos, variables macroscópicas, así como sustancias prioritarias micro-contaminantes. La importancia relativa de los indicadores al interior del sistema de inferencia se estimó con un método de análisis de decisiones, llamado proceso jerárquico analítico. En una segunda fase, se utilizó una metodología híbrida que combina los sistemas de inferencia difusos y las redes neuronales artificiales, conocida como neuro-fuzzy, para el estudio de la clasificación del estado ecológico de los ríos, de acuerdo con los lineamientos de la Directiva Marco de Aguas. Esta metodología permitió un manejo adecuado de la no-linealidad y naturaleza subjetiva de las variables involucradas en este problema clasificatorio. Con ella, se estudió la complejidad de los sistemas de inferencia, la selección apropiada de reglas lingüísticas y la influencia de las funciones que transforman las variables numéricas en lingüísticas. En una tercera fase, se desarrolló un modelo conceptual neuro-fuzzy concurrente basado en la metodología de evaluación de riesgo ecológico preliminar. Este modelo consideró la presencia de sustancias peligrosas en los ríos, e incorporó un mapa auto-organizativo para clasificar las sustancias químicas, en términos de su peligrosidad hacia los ecosistemas acuáticos. Con este modelo se estimaron potenciales de riesgo ecológico por combinación de factores de peligrosidad y de concentraciones de las sustancias químicas en el agua. Debido a la alta imprecisión e incertidumbre lingüística, estos potenciales se obtuvieron mediante sistemas de inferencia difusos, y se integraron por medio de distribuciones empíricas acumuladas, con las cuales se pueden analizar cambios espacio-temporales en la calidad del agua. Finalmente, se diseñó un sistema jerárquico de inferencia difuso para la evaluación del riesgo ecológico en sedimentos de ribera. Este sistema estima los grados de contaminación, toxicidad y riesgo en los sedimentos en términos lingüísticos, con sus respectivos niveles de certeza. El sistema se alimenta con información proveniente de análisis químicos, que detectan la presencia de sustancias micro-contaminantes, y de ensayos eco-toxicológicos tipo "screening" que usan la bacteria Vibrio fischeri. Como caso de estudio se seleccionó la cuenca del río Ebro, aunque las metodologías aquí desarrolladas pueden aplicarse fácilmente a otros ríos. En conclusión, este trabajo demuestra ampliamente que el diseño y aplicación de indicadores de calidad de las aguas, basados en la metodología de la lógica difusa, constituyen una herramienta sencilla y útil para los tomadores de decisiones encargados de la gestión sostenible de las cuencas hidrográficas.

We propose a novel approach for segmentation of human objects, including face and body, in image sequences. In modern video coding techniques, e.g., MPEG-4 and MPEG-7, human objects are usually the main focus for multimedia applications. We combine temporal and spatial information and employ a neuro-fuzzy mechanism to extract human objects. A fuzzy self-clustering technique is used to divide the video frame into a set of segments. The existence of a face within a candidate face region is ensured by searching for possible constellations of eye-mouth triangles and verifying each eye-mouth combination with the predefined template. Then rough foreground and background are formed based on a combination of multiple criteria. Finally, human objects in the base frame and the remaining frames of the video stream are precisely located by a fuzzy neural network which is trained by a SVD-based hybrid learning algorithm. Through experiments, we compare our system with two other approaches, and the results have shown that our system can detect face locations and extract human objects more accurately.

We develop a neuro-fuzzy network technique to extract TSK-type fuzzy rules from a given set of input-output data for classification problems. Fuzzy clusters are generated incrementally from the training data set, and similar clusters are merged dynamically together through input-similarity, output-similarity, and output-variance tests. The associated membership functions are defined with statistical means and deviations. Each cluster corresponds to a fuzzy IF-THEN rule, and the obtained rules can be further refined by a fuzzy neural network with a hybrid learning algorithm which combines a recursive SVD-based least squares estimator and the gradient descent method. The proposed technique has several advantages. The information about input and output data subspaces is considered simultaneously for cluster generation and merging. Membership functions match closely with and describe properly the real distribution of the training data points. Redundant clusters are combined and the sensitivity to the input order of training data is reduced. Besides, generation of the whole set of clusters from the scratch can be avoided when new training data are considered.

Die zuverlässige Einhaltung von Lieferzusagen stellt ein wichtiges Kriterium bei der Auswahl der Teilnehmer eines auftragsbezogenen Produktionsnetzes dar. Für die objektive Bewertung der Lieferzuverlässigkeit der potenziellen Netzwerkteilnehmer bedarf es der Quantifizierung der relevanten Unsicherheiten integriert in einen allgemein gültigen Ansatz der Verfügbarkeitsprüfung. Die Arbeit stellt daraus resultierend Ansätze zur Berechnung der Unsicherheit vor. Durch die Quantifizierung der Unsicherheit innerhalb der Unternehmen ergibt sich zudem die Möglichkeit der flexiblen, situationsabhängigen Nutzung des für langfristige Rahmenverträge reservierten Sicherheitsbestandes zur Befriedigung kurzfristiger Anfragen. Diese Aufgabe unterstützt ein konfigurierbares Modell zur Entscheidungsunterstützung, das auf einem Neuro-Fuzzy-System basiert. Die Kennzahlen der Lieferzuverlässigkeit unterliegen einem dynamischen Verhalten während des Wertschöpfungsprozesses in dem auftragsbasierten Produktionsnetz. Durch die Integration dieser Kennzahlen in das Management dieses Prozesses ergibt sich die Möglichkeit, aus der Zunahme der Unsicherheit mögliche Störungen und deren Auswirkungen bereits vor ihrem Eintreten zu erfassen und im Rahmen eines präventiven Störungsmanagements zu agieren.