Power system transient stability analysis by using modified Kohonen network

To maintain transient stability is one of the crucial conditions for designing and operating a modern interconnected power system. Using either time integration or transient energy functions, the critical clearing time (CCT) in a power system may be accurately obtained and can be regarded as an index for stability analysis. This method is not yet suitable for on line use. Many publications have shown that by using a feedforward preceptron with a backpropagation learning rule (BP network), the CCT can be fairly accurately determined. However, the long training procedures and occasionally serious incorrect predictions are associated. In this paper, an alternative to ANN has been developed for improving the performance of the CCT determination. The proposed technique combines both the characteristics of the Kohonen self-organised feature map (SOFM) and the BP network to form a multilayer neural network. In this paper, the transient energy function is employed to generate the desired output for the network learning process and the test results compared. The effectiveness of the proposed method is demonstrated in a sample power system