Growing-type WASD for power-activation neuronet to model and forecast monthly time series

In this paper, a novel WASD (weights and structure determination) algorithm is presented for the power-activation feed-forward neuronet (PFN) to solve monthly time series modeling and forecasting problems. Besides, a simple and effective data preprocessing approach is employed. Based on the WDD (weights direct determination) method and the relationship between the structure and the performance of PFN, the WASD algorithm can determine the weights and the optimal structure (i.e., the optimal numbers of input-layer and hidden-layer neurons) of the PFN. Numerical experiment results further substantiate the superiority of the PFN equipped with the WASD algorithm to model and forecast monthly time series from M forecasting competition.