Monitoramento multidimensional da qualidade de energia elétrica baseado no conceito de detecção de novidade

• Autor(a) principal: Mendes, Thais Martins
• Data de Publicação: 2017
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UFLA
• Texto Completo: http://repositorio.ufla.br/jspui/handle/1/12754
• topic: Energia elétrica - Qualidade; Detecção de novidade; Detecção de distúrbios elétricos; Classificação de distúrbios elétricos; Power quality; Novelty detection; Detection of electrical disturbances; Classification of electrical disturbances; Engenharias

Power Quality (PQ) has emerged as an important research field. This fact is explained by increasing use of high power converters and the increase of nonlinear loads with high power that cause unwanted changes in the electrical signals. These changes are called electrical disturbances. This work proposes a multidimensional approach for detecting and classifying PQ disturbances. The innovation of this work is the development of methods that apply the concept of novelty detection to PQ not yet proposed in the literature. A simple method for disturbance detection is proposed, where a general index of PQ is provided. This approach has the advantage of using the calculation of only a distance between two points in a feature space to obtain the result of the disturbance detection and to provide a general PQ index. As results we obtained performances greater than 90% for simulated data and of 100% for real data considering a the real time acquisition system. An innovative approach was proposed regarding the use of unsupervised classifier Support Vector Machine (SVM) to construct a multidimensional envelope to cover samples of each disturbances class, and to detect and classify them through the novelty detection concept. An accuracy of 100% was achieved by the detection system and the efficiency achieved by the classification system was above 99%. A the main advantage of the proposed method concerns is its ability to include new disturbance classes without the need of redesigning the classifier from scratch.