Radial basis function neural networks for fault diagnosis in controllable series compensated transmission lines

Since the complex variation of line impedance is controlled by thyristors and is accentuated as the capacitor's own protection equipment operates randomly under fault conditions in controllable series compensated transmission systems, conventional distance protection schemes are limited to certain applications. The authors have extensively addressed the development of new adaptive protection techniques for such power systems using multilayer perceptrons. The basic idea of the method is to design a protection scheme using a neural network approach by catching the feature signals in a certain frequency range under fault conditions. This is different from conventional schemes that are based on deriving implicit mathematical equations based on the information obtained by complex filtering techniques. This paper presents some recent results of employing different types of neural networks for this particular application. The performances of two neural networks have been analyzed and compared, including the: (i) backpropagation network (BP); and (ii) radial basis function network (RBFN). The study shows that the RBFN has better performance than the commonly used BP network. As fault identification is only part of the protection scheme, further work is focusing towards the development of a completed neural network-based protection technique.