Model-Based Fault Diagnosis of an Electrical Low-Voltage Grid

Abstract: Reliable access to power is essential for modern society’s fundamental functions. It is necessary that faults and abnormalities in low-voltage grids can easily be detected and diagnosed. A fault in a grid can be a consequence of several causes, including internal damage, technical errors, malfunctioning electricity meters and electricity theft. This thesis investigates the possibilities with model-based approaches for fault diagnosis and monitoring of low-voltage grids. Data and properties from a real low-voltage grid is utilised in this thesis. Furthermore, hypothetical sensors are introduced in the model equations and simulations. To delimit and make the results easier to understand, three different clusters, made up from a few households and intermediate cables, from the grid are analysed. The methods used includes modelling of the real low-voltage grid, structural analysis of the isolability performance of faults, and simulation of faulty scenarios in MATLAB. The investigation of the isolability performance uses isolability matrices and Dulmage Mendelsohn decomposition. The results from the structural analysis show that it is hard to design residuals that are only sensible to certain faults and draw conclusions on where a particular fault takes place. Many times, it shows that several faults cannot be isolable from each other. The redundancy has to be increased in the equations with even more sensors. By using the simulated model directly for residual generation, clearer and more determined results can be seen. Because it does take quantities into account, it is easier to analyse and look after changes in the grid. A few introduced sensors can often tell where a current through a cable must have been increased. An injected fault will more or less always affect residuals indirect proximity. However, because faults can occur in both sensors and cables, itis often hard to specify exactly why and where the fault takes place.