Figure 1 highlights the overall voltage security framework.
Figure 1: Voltage security framework
Figure 2: Typical EHV PV transmission characteristics
For a given power system and voltage security problem, operating states are first generated via random sampling, in a sufficiently broad and diverse range so as to screen all situations deemed relevant. Each case is characterized by pre-contingency attributes, i.e. parameters describing its electrical and topological normal state. Further, it is pre-analyzed in terms of voltage security by computing its pre- and post-contingency LPMs, for various contingencies. Typically, several thousand operating states are drawn and several tens of contingencies are considered in a given study. Massive parallelism may be exploited to speed up this otherwise lengthy off-line simulation phase.
At the second step, the non-parametric regression techniques are applied, in a contingency by contingency fashion, to extract synthetic security information in the form of approximate models expressing the relationship between the attributes and the severity of a contingency, which is defined by the drop in LPM from pre-contingency to the post-contingency state (see Fig. 2). In this context, regression trees are used to identify salient attributes which influence the severity of a contingency and to provide a first guess transparent model, while multilayer perceptrons are used to refine this model.
The third step consists of using this synthetic security information on-line for security assessment.