Design and analysis of genetic hybrid learning basis function neural network for transient stability of power systems

In this paper, the design and analysis of a genetic hybrid learning radial basis function neural network is done for the transient stability problems of power systems. In the paper, a Gaussian function along with its Euclidean summation is used as the activation function of the pattern layer node of the proposed neural network. A hybrid learning algorithm was developed where the unsupervised sub-algorithm was used to heuristically select the centers and widths of the activation function, and the supervised learning to adaptively adjust the weights. A genetic preprocessor was added to the proposed neural network as an input variable selector. The combined 'genetic and hybrid learning radial basis function' (GHBF) neural network was applied as an intelligent assessor for power system stability analysis. From simulation results, this intelligent assessor was proven to be capable of approximating the dynamic behavior of the power systems under study, with an acceptable level of accuracy. The GHBFNN assessor was simulated for two different power systems. The training of this network is divided into two learning phases. In the unsupervised learning (USL), an adaptive clustering method was used to combine the adaptive k-means algorithm with its linear learning rate. In the supervised learning (SL), the generalized bar-delta (GBD) rule was applied which was proven to excel the generalized delta rule (GDR). Despite the fact that fewer input variables were to train this network, its performance is quite similar to its counterpart, the hybrid basis function neural network (HBFNN).