Our first investigations concern multi-contingency emergency state detection of a synthetic ``academic'' 7-bus system. System and data base descriptions are given in [9][25]; to save space we do not reproduce them here.
The data base concerns 500 normal pre-contingency operating states and
5 disturbances yielding 2500 JAD states, which are randomly split into
1250 learning and 1250 test states. The pre-contingency states are
generated by random sampling and loadflow computation, whereas the
contingencies are simulated by numerical integration. 28 candidate
attributes are used to characterize the JAD states which correspond to
snapshots at 
after the disturbance. To classify the states,
the simulation proceeds until an equilibrium state is reached (e.g.
some minutes) or a voltage collapse is diagnosed. Accordingly, 36%
of the 2500 JAD states are found to be critical.