Home // DATA ANALYTICS 2018, The Seventh International Conference on Data Analytics // View article

Algorithms for Electrical Power Time Series Classification and Clustering

Authors:
Gaia Ceresa
Andrea Pitto
Diego Cirio
Emanuele Ciapessoni
Nicolas Omont

Keywords: Multimodality; Gaussian Mixture; Cluster; Time series

Abstract:
The EU-FP7 project iTesla developed a toolbox aimed to assess dynamic security of large electrical power systems, taking into account the forecast uncertainties due to renewable energy sources and load. Important inputs to the toolbox consist in the forecasts and the realizations of thousands of active power injections from renewable generators, and of active and reactive power absorption of the load, in the high voltage French transmission grid, collected into hourly historical time series. Data show a deep variety of distribution functions and profiles in the time domain. In this context, the statistical analysis of historical dataset is very important in order to characterise and manage such a large variability of distributions. In particular, the potential multimodality of the variables has to be identified in order to adapt the sampling technique developed in the iTesla toolbox, thus assuring accurate results also for this subset of variables. Moreover, clustering some variables can help reduce the dimensionality of the problem, which represents an important advantage while analysing security on very large power systems. The paper describes four algorithms: one looks for the number of distribution’s peaks and classifies the variables into unimodal or multimodal; the second and the third cluster and combine multimodal variables to obtain unimodal ones, because they are more suitable for the subsequent computation. All of them are part of an advanced tool for automatic data description, that pre-processes the raw data and produces descriptive statistics on them. The Separation Algorithm is the last one, it back-projects the sum of two series into the original components.

Pages: 79 to 84

Copyright: Copyright (c) IARIA, 2018

Publication date: November 18, 2018

Published in: conference

ISSN: 2308-4464

ISBN: 978-1-61208-681-1

Location: Athens, Greece

Dates: from November 18, 2018 to November 22, 2018