The analysis of unbalanced power system problems using the method of phase co-ordinates by which the phase quantities are preserved has obvious practical advantages over the transformation methods requiring the phase quantities to be substituted by, for example, symmetrical component and d-q quantities. Since the physical identity of the system is maintained in the phase frame of reference, the matematical models of the system can be subjected to actual operating constraints enabling a unified approach to be adopted in the study of both symmetrical and unsymmetrical conditions. In this thesis the bond graph techniques are used to model power system components in terms of their phase co-ordinates. The bond graph structure, which is based on energy continuity and power balance, classifies system and sub-systems, with respect to the number of energy ports through which energy or power is exchanged with the environment as well as in terms of the particular internal energy transformations involved. The use of bond graphs for the analysis of non-linear electromagnetic systems has resulted in the evolution of a uniform diagramatic structure which, represents in a single diagram the basic field properties such as the magneto-motive force and the flux linkages together with the conjugate circuit power variables, voltage and current. As an extension of this ideal the general n-port electromagnetic and electrostatic field systems are derived in bond graph notations. A very desirable feature of this method of presentation is that there exists a one-to-one-correspondence between the bond graph and a computation structure which may be used for the purpose of simulation by either analogue or digital computers. The general n-port electromagnetic and electrostatic field systems form the basis for the effective modelling of power system components such as rotating machines, transformers and transmission lines. In the investigations of earth faults, these models are simplified in accordance with the need of the particular study but are sufficiently accurate representations to allow predictions on the overall system behaviours. The bond graph approach is ideally suited to expedite the modelling of dynamic interacting energy systems and when used in conjunction with the computation structure can provide the simulation technique required to model accurately the non-linear behaviours of multiport electromechanical energy converters. The effects of magnetic saturation in synchronous generators are presented in detail.

Cette thèse porte sur la surveillance des perturbations de la qualité de l’énergie d’un réseau électrique via des techniques paramétriques de traitement du signal. Pour élaborer nos algorithmes de traitement du signal, nous avons traité les problèmes d’estimation des différentes grandeurs du réseau électrique triphasé et de classification des perturbations de la qualité d'énergie. Pour ce qui est du problème d’estimation, nous avons développé une technique statistique basée sur le maximum de vraisemblance. La technique proposée exploite la nature multidimensionnelle des signaux électriques. Elle utilise un algorithme d’optimisation pour minimiser la fonction de vraisemblance. L’algorithme utilisé permet d’améliorer les performances d’estimation tout en étant d’une faible complexité calculatoire en comparaison aux algorithmes classiques. Une analyse plus poussée de l’estimateur proposé a été effectuée. Plus précisément, ses performances sont évaluées sous un environnement incluant entre autres la pollution harmonique et interharmonique et le bruit. Les performances sont également comparées aux exigences de la norme IEEE C37.118.2011. La problématique de classification dans les réseaux électriques triphasés a plus particulièrement concerné les perturbations que sont les creux de tension et les surtensions. La technique de classification proposée consiste globalement en deux étapes : 1) une pré-classification du signal dans l’une des 4 préclasses établis et en 2) une classification du type de perturbation à l’aide de l’estimation des composants symétriques.Les performances du classificateur proposé ont été évaluées, entre autres, pour différentes nombre de cycles, de SNR et de THD. L’estimateur et le classificateur proposés ont été validés en simulation et en utilisant les données d’un réseau électrique réel du DOE/EPRI National Database of Power System Events. Les résultats obtenus illustrent clairement l’efficacité des algorithmes proposés quand à leur utilisation comme outil de surveillance de la qualité d’énergie. / This thesis deals with electric grid monitoring of power quality (PQ) disturbances using parametric signal processing techniques. The first contribution is devoted to the parametric spectral estimation approach for signal parameter extraction. The proposed approach exploits the multidimensional nature of the electrical signals.For spectral estimation, it uses an optimization algorithm to minimize the likelihood function. In particular, this algorithm allows to improve the estimation accuracy and has lower computational complexity than classical algorithms. An in-depth analysis of the proposed estimator has been performed. Specifically, the estimator performances are evaluated under noisy, harmonic, interharmonic, and off-nominal frequency environment. These performances are also compared with the requirements of the IEEE Standard C37.118.2011. The achieved results have shown that the proposed approach is an attractive choice for PQ measurement devices such as phasor measurement units (PMUs). The second contribution deals with the classification of power quality disturbances in three-phase power systems. Specifically, this approach focuses on voltage sag and swell signatures. The proposed classification approach is based on two main steps: 1) the signal pre-classification into one of 4 pre-classes and 2) the signature type classification using the estimate of the symmetrical components. The classifier performances have been evaluated for different data length, signal to noise ratio, interharmonic, and total harmonic distortion. The proposed estimator and classifier are validated using real power system data obtained from the DOE/EPRI National Database of Power System Events. The achieved simulations and experimental results clearly illustrate the effectiveness of the proposed techniques for PQ monitoring purpose.