A hybrid "k-means, VSS LMS" learning method for RBF network in short-term load forecasting

In this paper we investigate the performance of a hybrid learning algorithm for RBF network in the application of short-term load forecasting. In this method the algorithm for finding radial basis function centers of hidden layer is k-means and the algorithm for training the weights of output layer is adaptive variable step-size algorithm. We proved this method is both accurate and fast in comparison with other presented schemes. Also we demonstrated that this method requires less computational processing and can perform well when amount of the input data is large. Our simulation results show there is up to 30 percent improvement in processing time and 37% improvement in prediction accuracy when compared with previously improved k-means learning.