Relais de perte de synchronisme par apprentissage profond

Abstract

The objective of this exercise is to explore the possibility of using Deep Learning to develop an out of step relay. With the transient electromagnetic simulator EMTP, a series of simulations are produced and then classified with a machine learning algorithm for multiple time series called MALSTM-FCN (Multivariate Attentional Long Short Term Memory Fully Convolutional Network). Using EMTP, a test network is modelled to represent a region to study. In this network, 4 places will be chosen to take measures for the loss of step relay. The measures include three phase voltages, three-phase currents and active and reactive threephase powers. These measures are than imported in a matrix treatment tool called MATLAB well suited to handle big matrices. Once imported in MATLAB, the data are separated per generator and catalogued in two classes of stable or unstable simulation. Following this, for each generator, a training database and a test database are constructed from the set of stable and unstable simulations randomly so that they are evenly represented in those two databases. Then the Deep Learning algorithm MALSTM-FCN is used on the training database to allow the algorithm to learn to recognize stable and unstable network situations. Finally, the performance of the algorithm is evaluated in the training database to determine its generalization capability on an unlearned set of simulations. The novelty of this approach resides in the use of unfiltered raw data from a small number of inputs such as three phase voltage and currents and tree-phase active and reactive power. The approach is always markedly different from other machine learning methods in that there is no feature extraction phase used to extract additional information from the inputs such as additional information concerning the state of the generators.

Date	7 Aug 2020
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Ambrish Chandra (Supervisor) & Innocent Kamwa (Co-supervisor)