This thesis presents an approach for identifying a maximal number of dangerous contingencies in large scale power system security assessment problems with bounded computational resources.
The method developed in this work relies on the definition of an objective function associating to each contingency a real value that quantifies its severity for the security of the system, this value being greater than or equal to a given threshold only for dangerous contingencies. The value of this function for a given contingency is computed from the result of a security analysis executed on the post-contingency configuration.
The framework we propose for identifying dangerous contingencies is derived from
an algorithm from the optimization literature so as to find, with a given number of evaluations of the objective function, a maximal number of contingencies whose value of this function exceeds the adopted threshold. This approach performs successive
samplings of the space gathering all the contingencies, and exploits the information contained in each of these samples in order to direct the subsequent sampling process towards contingencies with high values of the objective function. Our algorithm is first introduced in the case where the search space is a Euclidean space. Then we propose an extension of this approach to the more common case where the search space is discrete, thanks to a procedure allowing to embed a discrete contingency space in a Euclidean space, over which a metric is defined.
The efficiency of the developed method is evaluated on several case studies: an
N − 3 analysis of a benchmark test system, the IEEE 118 bus test system, and N − 1 and N − 2 studies of a real system, the Belgian transmission network.
Afterwards, we consider the case where several of these iterative sampling algorithms are available. Assuming that these algorithms are executed sequentially, we propose two different strategies for selecting on-line which of them to execute at the next step in order to identify as many dangerous contingencies as possible, while still
respecting the given computational budget.
We finally provide an adapted version of the developed iterative sampling algorithm allowing to estimate the probability of occurrence of a dangerous contingency and the number of dangerous contingencies in a discrete search space.