Abstract:

As to the limitation of traditional starting criteria for fault recorder algorithm, this article proposes a algorithm based on SOM neural network. An example of A-phase current limit is done in this research of algorithm. The construction of SOM neural network, network training and cluster prediction are completed, then inputting the input vector normalized to the trained SOM model, the output will be shown in dimensional plane array, and blue neurons in network topology represent phase A over limit, at this point, wave recording should be started. In order to verify the accuracy of the model, two input vector of different dimensionality are inputted into network model, then the outcome shows that the starting criteria for fault recorder algorithm based on SOM neural network has strong adaptive ability and can effectively complete the recording start with minor error.

Key words: smart substation, fault recorder, starting criteria, SOM neural network, cluster analysis