Diploma and Master Theses (authored and supervised):
"An Autonomic Approach for Adaptive Monitoring in a Smart Grid Context";
Supervisor: S. Dustdar, B. Satzger, C. Inzinger;
Institut für Informationssysteme, AB Verteilte Systeme,
final examination: 2013-03-05.
Today´s power grids are mainly based on a relatively small number of power plants converting fuel-energy into electricity. The outdated infrastructure and the resulting inefficient use of our resources requires steps towards a smarter grid. The aim of a smart grid is to make extensive use of renewable energy resources, like wind and solar,
and include more power producers like solar panels of single households. The basis for these additional features are the increasing number and capabilities of today´s smart meters. They are widely spread within a power grid and expected to deliver more and
more detailed information about the power consumption in the future.
In order to determine the balance of supply and demand within the power grid the highly distributed meters have to be monitored. But this will exert a huge load on the network, thus more sophisticated monitoring approaches are required. We present an autonomic approach for adapting the amount of monitoring in a smart grid context. This
approach is based on the existence of correlated sensors. Wind wheels, for example, are placed in a high density in a close neighbourhood, thus it is very likely that the power
generation of these power plants is somehow related. We are using machine learning algorithms for detecting and grouping correlated sensors, thus the measurement of one sensor can be inferred by the measurement of another. The number of sensors suitable for preforming such a prediction is influenced by a required accuracy. We assume that
it is not necessary to have a 100% accurate view onto the power generation, when an administrator is at least able to interpret if there is enough supply for the current or upcoming demand. The required accuracy has to be defined by an expert familiar with the
domain and its properties. We manage two properties that directly influence the amount of monitoring. Firstly, the number of requests sent over the network and secondly the frequency in which these requests are performed. We present different strategies for
adapting these properties and evaluate them on a prototype implementation. The comparison of our approach against a reference scenario with adaptation showed retrenchment potentials in sensor requests up to 95%, while providing a required accuracy.
Created from the Publication Database of the Vienna University of Technology.