Description
Date depot: 1 janvier 1900
Titre: Modeling and optimizing a distributed power network: A complex system approach of the 'prosumer' management in the smart grid
Encadrant :
Vincent GAUTHIER (SAMOVAR)
Directeur de thèse:
Michel MAROT (LTCI (EDMH))
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Non defini
Resumé:
Introduction
Our current way of powering the grid based on fossil energies is becoming delicate as these are becoming scarcer and scarcer while the demand keeps progressing over the years [1]. A very nice way of resolving this problem as well as providing a cleaner power system is often assumed to result from the integration of numerous renewable plants inside the grid. These “multi-behavioral micro-generators”, whose purpose is to be as close as possible to the end users, will populate the low voltage grid and deliver energy depending on external parameters such as meteorological conditions for instance. This is a very delicate point as these generators, in addition of being numerous, exhibit also uncertain generation profiles and can hardly be scheduled over time. This means that current power engineering philosophy of estimating the demand and scheduling the production in accordance may not provide a proper control for such a system. Furthermore, assuming also the development of electrical cars and storage devices, the structure of this futuristic system seems to be really different from our current hierarchical and centralized power network. How can we then determine the system’s characteristics (topology, control…) that will provide the necessary resilience for the smart grid to become reasonably responsible for our energy alimentation ?
The recent catastrophic blackout events (Italy 2003, North America 2003, etc…) [2] [3] remind us indeed that, no matter how robust the power grid seems to be, small events, like failures of components, can potentially trigger large wide-spread cascades [4]. Failure events are hardly predictable as they may result from extremely various events (storm, heat, dysfunctions…), and the classical way of dealing with them is to introduce some redundancy in the system so that failure of a component does not trigger cascades.
Nevertheless, even without failure, the power grid remains a dynamical and non linear system, which means that small perturbations like local loss of synchrony can be amplified and cause the desynchronization of the whole system. In other words, the power grid has to be driven continually within its stability basin (in terms of respect to the grid’s main frequency for example). This task requires some synchronization among the grid’s devices as well as the possibility to drive them without deviating too much from this synchronous state [5]. Given the demand forecasts, one of the operator’s main concerns lies then on the power production allocation among the generators, in order to meet the demand while staying as close as possible to the synchronous stable state [6].
These centralized control operations are well established for the current power networks [6]. However, this approach has to be reconsidered in a scenario with decentralized power production system. Especially with a high penetration of renewable sources that generate power according to meteorological conditions that can only be vaguely estimated [7] [8], the stability questions as well as the insurance to meet demand become critical.
Issues
Since a few years, it is common to use tools from the complex system theory to study large interacting systems [9] [10]. Power grids make no exception, and scientific literature provide interesting investigations on power grid’s topologies, aiming at finding a relation between the resilience of these dynamical systems and their underlying topologies [2] [11] [12]. This is often done by abstracting the grids into mathematical graphs and performing then statistical studies on them. A question of interest for instance lies in determining power grids degree distributions, which can lead to further conclusions. Mostly, these studies conclude exponential or degenerated scale-free distributions as well as low clustering coefficients, indicating that, unlike other networks like the internet or the World Wide Web, the power grids do not exhibit high degree hubs. This approach provides interesting results but doesn’t capture the dynamical essence of the power grid. It can be seen as topological studies of the underlying electrical circuits, but without taking into account the nature of what is flowing in them [13].
Actually, electricity cannot be handled very easily in the sense that, unlike data for instance, it cannot be routed inside the network as it only obeys Kirchhoff’s current law. This means that, when injected inside the grid, electricity splits along different paths according to complex impedances relations. Thus, considering the grid as a simple unweighted undirected graph with a shortest path distance metric may not provide insightful results in terms of a dynamical oriented study.
Other publications tackle this topology question by introducing edge weights corresponding to impedance values and by thinking in terms of betweeness centrality rather than degrees [14]. The idea behind this being that betweeness measure is m
Doctorant.e: Gensollen Nicolas