Philosophiae Doctor - PhD / In this thesis, we present a novel hybrid architecture by combining the most popular sequence recognition models such as Recurrent Neural Networks (RNNs) and Hidden Markov Models (HMMs). Though sequence recognition problems could be potentially modelled through well trained HMMs, they could not provide a reasonable solution to the complicated recognition problems. In contrast, the ability of RNNs to recognize the complex sequence recognition problems is known to be exceptionally good. It should be noted that in the past, methods for applying HMMs into RNNs have been developed by other researchers. However, to the best of our knowledge, no algorithm for processing HMMs through learning has been given. Taking advantage of the structural similarities of the architectural dynamics of the RNNs and HMMs, in this work we analyze the combination of these two systems into the hybrid architecture. To this end, the main objective of this study is to improve the sequence recognition/classi_cation performance by applying a hybrid neural/symbolic approach. In particular, trained HMMs are used as the initial symbolic domain theory and directly encoded into appropriate RNN architecture, meaning that the prior knowledge is processed through the training of RNNs. Proposed algorithm is then implemented on sample test beds and other real time biological applications.

No description available.

The dynamic response of the demand is defined as the time-domain real and reactive power response to a voltage disturbance, and it represents the dynamic load characteristics. This thesis develops a methodology for probabilistic estimation and prediction of dynamic responses of the demand at bulk supply points. The main outcome of the research is being able to predict the contribution of different categories of loads to the total demand mix and their controllability without conducting detailed customer surveys or collecting smart meter data, and to predict the dynamic response of the demand without performing field tests. The prediction of the contributions of different load categories and their controllability and load characteristics in the near future (e.g., day ahead) plays an important role in system analysis and planning, especially in the short-term dispatch and control. However, the research related to this topic is missing in the publically available literature, and an approach needs to be developed to enable the prediction of the participation of different loads in total load mix, their controllability and the dynamic response of the demand. This research contributes to a number of areas, such as load forecasting, load disaggregation and load modelling. First, two load forecasting methodologies which have not been compared before are compared; and based on the results of comparison and considering the actual requirements in this research, a methodology is selected and used to predict both the real and reactive power. Second, a unique methodology for load disaggregation is developed. This methodology enables the estimation of the contributions of different load categories to the total demand mix and their controllability based on RMS measured voltage and real and reactive power. The confidence level of the estimation is also assessed. The methodology for disaggregation is integrated with the load forecasting tool to enable prediction of load compositions and dynamic responses of the demand. The prediction is validated with data collected from real UK power network. Finally, based on the prediction, an example of load shifting is used to demonstrate that different dynamic responses can be obtained based on the availability and redistribution of controllable devices and that load shifting decisions, i.e., demand side management actions, should be made based not only on the amount of demand to be shifted, but also on predicted responses before and after load shifting.

Submitted by Maria Aparecida (cidazen@gmail.com) on 2017-08-02T11:31:43Z No. of bitstreams: 1 Felipe Oliveira.pdf: 3179442 bytes, checksum: 0988804a0a58c2aaf337ea2f5034dc42 (MD5) / Made available in DSpace on 2017-08-02T11:31:43Z (GMT). No. of bitstreams: 1 Felipe Oliveira.pdf: 3179442 bytes, checksum: 0988804a0a58c2aaf337ea2f5034dc42 (MD5) Previous issue date: 2017-02-17 / The locational marginal prices (LMPs) are essential financial guidelines for the electricity industry, which orientates most of the projects and deliberations in electrical market environments. In current scenario of the electricity markets, wind power plants and energy storage systems have been revealing itself as feasible and relevant electrical energy supply alternatives. In this work a generic methodology based on artificial intelligence (AI) techniques is formulated and applied to the calculation and decomposition of LMPs of electric power systems (EPS) with the insertion of energy storage systems and wind farms. In the proposed AI-based methodology the optimal power flow (OPF) model, on which the calculation and decomposition of LMP is based, considers the wind behavior profile volatility, the risks of wind power levels previously scheduled, and the energy storage systems operative peculiarities. The proposed AI-based methodology takes into account the mathematical and computational models of the particle swarm optimization (PSO) algorithm. This proposal was properly implemented and applied for the computation and decomposition of LMPs of test systems and considering different operative scenarios involving conventional power plants, wind farms, and energy storage systems. / Os preços marginais locacionais (LMPs – Locational Marginal Prices) consistem em diretrizes financeiras mercadologicamente indispensáveis para a indústria da eletricidade, os quais norteiam grande parte dos projetos e deliberações no âmbito dos mercados elétricos. No panorama vigente dos mercados elétricos, as plantas de geração eólica e os sistemas de armazenamento de energia vêm progressiva e ininterruptamente se revelando alternativas de suprimento de eletricidade cada vez mais relevantes e viáveis. Neste trabalho, é formulada uma metodologia genérica baseada em técnicas de inteligência artificial (IA) cuja aplicação tem o objetivo de computar e decompor os LMPs associados às barras constituintes de um sistema elétrico de potência (SEP) integrado por geradores convencionais, plantas de geração eólica e por sistemas de armazenamento de energia. Na metodologia IA proposta, o modelo de fluxo de potência ótimo (FPO) sobre o qual se alicerça o cômputo e a decomposição dos LMPs associados às barras de um SEP, leva em consideração a volatilidade inerente ao perfil comportamental dos ventos, os riscos associados à assunção de níveis previamente programados de potência proveniente da geração eólica e as peculiaridades operativas concernentes aos sistemas de armazenamento de energia. Adotando-se os modelos matemáticos e computacionais dos algoritmos de otimização por enxame de partículas (PSO – Particle Swarm Optimization), a metodologia IA proposta foi devidamente implementada e aplicada na aquisição e decomposição dos LMPs associados às barras constituintes de sistemas-testes submetidos a diferentes cenários operativos envolvendo centrais de geração convencionais, plantas de geração eólica e sistemas de armazenamento de energia.

Maitriser le triptyque coût-qualité-délai lors des différentes phases du Processus de Développement d’un Produit (PDP) dans un environnement de plus en plus concurrentiel est un enjeu majeur pour l’industrie. Le développement de nouvelles méthodes et de nouveaux outils pour adapter une représentation du produit à une activité du PDP est l’une des nombreuses pistes d’amélioration du processus et certainement l’une des plus prometteuses. Cela est particulièrement vrai dans le domaine du transfert de modèles de Conception Assistée par Ordinateur (CAO) vers des activités de simulations numériques. Actuellement, les méthodes et outils de préparation d’un modèle CAO original vers un modèle dédié à une activité existent. Cependant, ces processus de préparation sont des tâches complexes qui reposent souvent sur les connaissances des experts et sont peu formalisés, en particulier lorsque l’on considère des maquettes numériques riches comprenant plusieurs centaines de milliers de pièces. Pouvoir estimer a priori l’impact de la préparation de la maquette numérique sur le résultat de la simulation permettrait d’identifier dès le début le meilleur processus et assurerait une meilleure maitrise des processus et des coûts de préparation. Cette thèse a pour objectif de relever ce défi en utilisant des techniques d’intelligence artificielles capables d'imiter et de prévoir un comportement à partir d'exemples judicieusement choisis. L’idée principale est d’utiliser des exemples de préparation de maquettes numériques comme entrées d’algorithmes d’apprentissage pour configurer des estimateurs de la performance d’un processus. Lorsqu’un nouveau cas se présente, ces estimateurs pourront alors prédire a priori l’impact de la préparation sur le résultat de l’analyse sans avoir à la réaliser. Afin d'atteindre cet objectif, une méthode a été développée pour construire une base d’exemples représentatifs, identifier les variables d’entrée et de sortie déterminantes et configurer des modèles d’apprentissage. La performance d’un processus de préparation sera évaluée à l’aide de critères tels que des coûts de préparation, des coûts de simulation et des erreurs sur le résultat de l’analyse dues à la simplification des modèles CAO. Ces critères seront les données de sortie des algorithmes d’apprentissage. Le premier challenge de l’approche proposée est d’extraire les données des modèles 3D complétées par des données relatives au cas de simulation qui caractérisent au mieux un processus de préparation , puis d’identifier les variables explicatives les plus déterminantes. Un autre challenge est de configurer des modèles d’apprentissage capables d’évaluer avec une bonne précision la qualité d’un processus malgré un nombre limité d’exemples de processus de préparation et de données disponibles (seules les données relatives aux modèles CAO originaux, aux cas de simulation sont connues pour un nouveau cas). Au final, l’estimateur de la performance d’un processus aidera les analystes dans le choix d'opérations de préparation de modèles CAO. Cela ne les dispensera pas de la simulation mais permettra d'obtenir plus rapidement un modèle préparé de meilleure qualité. Les techniques d’intelligence artificielles utilisées seront des classifieurs de type réseaux de neurones ou arbres de décision. L’approche proposée sera appliquée à la préparation de modèles CAO riches pour l’analyse CFD. / Controlling the well-known triptych costs, quality and time during the different phases of the Product Development Process (PDP) is an everlasting challenge for the industry. Among the numerous issues that are to be addressed, the development of new methods and tools to adapt to the various needs the models used all along the PDP is certainly one of the most challenging and promising improvement area. This is particularly true for the adaptation of CAD (Computer-Aided Design) models to CAE (Computer-Aided Engineering) applications. Today, even if methods and tools exist, such a preparation phase still requires a deep knowledge and a huge amount of time when considering Digital Mock-Up (DMU) composed of several hundreds of thousands of parts. Thus, being able to estimate a priori the impact of DMU preparation process on the simulation results would help identifying the best process right from the beginning, and this will ensure a better control of processes and preparation costs. This thesis addresses such a difficult problem and uses Artificial Intelligence (AI) techniques to learn and accurately predict behaviors from carefully selected examples. The main idea is to identify rules from these examples used as inputs of learning algorithms. Once those rules obtained, they can be used as estimators to be applied a priori on new cases for which the impact of a preparation process can be estimated without having to perform it. To reach this objective, a method to build a representative database of examples has been developed, the right input and output variables have been identified, then the learning model and its associated control parameters have been tuned. The performance of a preparation process is assessed by criteria like preparation costs, analysis costs and the errors induced by the simplifications on the analysis results. The first challenge of the proposed approach is to extract and select most relevant input variables from the original and 3D prepared models, which are completed with data characterizing the preparation processes. Another challenge is to configure learning models able to assess with good accuracy the quality of a process, despite a limited number of examples of preparation processes and data available (the only data known to a new case are the data that characterize the original CAD models and simulation case). In the end, the estimator of the process’ performance will help analysts in the selection of CAD model preparation operations. This does not exempt the analysts to make the numerical simulation. However, this will get faster a simplified model of best quality. The rules linking the output variables to the input ones are obtained using AI techniques such as well-known neural networks and decision trees. The proposed approach is illustrated and validated on industrial examples in the context of CFD simulations.