We denote by 
the pre-contingency topology and electrical
state and 
the pre-contingency margin.
Further, we denote by 
the post-contingency
load power margin when contingency 
is applied to the state , and we thus define by
the severity (in state ) of the contingency.
In the same fashion, denoting by 
the topology and electrical state of a
system following 
outages, the severity of a sequence of 
outages is derived by
The latter formula expresses the fact that if we are able to predict the security margin for a single outage in terms of parameters whose changes in the post-contingency state are easy to evaluate then we are also able to predict the effect of any sequence of outages.
To construct automatically an approximation of the term 
, we propose to use computer based learning
techniques in the form of non-parametric statistical regression
methods, using representative learning sets of randomly generated
power system operating states, pre-characterized with respect to
voltage security by their pre- and post-disturbance LPM.
