TABLE OF CONTENTS The classification of power system states with the self-organising feature map can be summarised as follows:
The neural net is trained off-line with one or several base cases and the related n ?1 or n ?2 contingencies. These contingencies can equally be regarded as different topologies of the power system.
An unknown operating state is presented in real time to the neural net. The neural net considers this state as a base case and will classify this case by a prototype state, e.g. weight vector 33.
Knowing that e.g. all contingencies of line N-S of the vectors of class 33 will lead to an overload in line N-L the neural net draws similar conclusions for all n ?1 or n ?2 contingencies of the unknown operating state. If the unknown operating state corresponds already to a different topology obtained from the base case by taking one line out, then conclusions can be drawn for all single contingencies of the case provided all n ?2 contingencies have been trained.
If during power system operation the classification of the operating state moves from e.g. neuron 33 over to neuron 25, the neural net further indicates that the trajectory of the operating point moves towards an overload situation concerning line N-S or line S-E.
There are several advantages to this entirely new approach:
There is no need to run a contingency analysis in real time and therefore the problem of combinatorial explosion can be avoided during...
