Power system topology verification using fuzzy neural networks

The paper deals with power system topology verification. The correct model of a power system topology is essential for different power system applications. There are different approaches to power system topology verification. One of these approaches assumes utilization of fuzzy neural networks (FNNs). In the paper, this approach is considered using the raw measurement data of active and reactive power flows on the one end of the branches and data of voltage magnitudes at nodes, which are used as input variables to the fuzzy neural network. FNN proposed for topology verification is Takagi-Sugeno-Kang’s fuzzy neural network.To verify the correctness of the choice of input variables Kohonen’s selforganising map (SOM) is applied. A projection based on Kohonen’s SOM is used to map variables from a d-dimensional input space to a two-dimensional output space. Kohonen’s SOM has a very desirable property of topology preserving, which means that clustering tendencies of different classes in the input space are preserved in the projected output space, providing good visualisation capability.An important factor influencing the accuracy of topology verification is normalization of input data. Data normalization is necessary for adequate application of mathematical models and computer-aided computations during calculations connected with big and small quantities for their equal distribution, for the values to be represented in the range [0, 1]. The calculations showed that topology verification errors depend tangibly on the method of input data normalization. In this regard, studies have been conducted to select normalization formula.In order to gauge the effectiveness of this method, the method was tested using the IEEE 14-bus test system. Operating conditions was calculated in Matlab software. The research results allowed making a conclusion that the topology verification approach using fuzzy neural networks showed a good accuracy.