The analysis of waveforms related to transient events is an important task in power system maintenance. Currently, electric power systems are monitored by several event recorders called phasor measurement units (PMUs) which generate a large amount of data. The number of records is so high that it makes human analysis infeasible. An alternative way of solving this problem is to group events in similar classes so that it is no longer necessary to analyze all the events, but only the most representative of each class. Several automatic clustering algorithms have been proposed in the literature. Most of these algorithms use validation indexes to rank the partitioning quality and, consequently, find the optimal number of clusters. However, this issue remains open, as each index has its own performance highly dependent on the data spatial distribution. The main contribution of this paper is the development of a methodology that optimizes the results of any clustering algorithm, regardless of data spatial distribution. The proposal is to evaluate the internal correlation of each cluster to proceed or not in a new partitioning round. In summary, the traditional validation indexes will continue to be used in the cluster’s partition process, but it is the internal correlation measure of each one that will define the stopping splitting criteria. This approach was tested in a real waveforms database using the K-means algorithm with the Silhouette and also the Davies–Bouldin validation indexes. The results were compared with a specific methodology for that database and were shown to be totally consistent.
1 August 2024 in Journal of Control, Automation and Electrical Systems
Os sistemas de teleproteção tem sido modernizados com a inserção de novas tecnologias de telecomunicações. Sistemas mais confiáveis com menor tempo de resposta estão sendo testados e aplicados às linhas de transmissão. Este trabalho apresenta uma revisão tema destacado, abordando aspectos do sistema de transmissão, equipamentos, funções de proteção aplicáveis e sistemas de telecomunicações, detalhando aspectos técnicos e apresentando uma avaliação comparativa entre tecnologias.
This work presents a case study of the application of supervised and unsupervised machine learning techniques in the study of three of Petrobras FPSO units in regards to its equipments power demand. Specifically, it delves into the outcomes of the clustering and equipment modelling modules of a computational solution developed in a partnership between Universidade Federal Fluminense (UFF) and Petrobras, FPSO Power Demand Analytics (FPDA). The presented results were found satisfactory by UFF and Petrobras development and engineering teams. For example the equipment modelling methodology resulted in a library of models from which the median absolute error rarely exceeds the 3% mark. The median of the median absolute errors observed across platforms and test scenarios is often less than 1%.
As the complexity of the power grid increases, so does the need for monitoring to ensure a high degree of service availability. However, the extensive instrumentation of the network generates such a large volume of data that it becomes impractical for specialists to analyze most of the recorded events. The literature reports that unsupervised machine learning techniques, particularly clustering methods, have been widely employed to address this problem. Another issue is that the spatial distribution of the data directly influences the performance of these tools. This work proposes using novel heuristics based on an infrastructureless (ad hoc) radio network, enabling the development of a clustering algorithm with greater immunity to database geometry. In an initial attempt, the algorithm exhibited high memory consumption and was limited to processing smaller databases. A new approach successfully addressed this limitation by discretizing the sample space through constructing a grid. Both algorithms were validated against synthetically generated databases, and experts classified a reduced set of real events. Finally, the proposed algorithm was applied to a real database composed of $\mathbf{5 2, 2 9 9}$ oscillographs recorded at a substation of the Brazilian Electric System. The consistency of the results was demonstrated through comparisons with other well-established algorithms in the scientific community, paving a new path for clustering data with unknown distributions.