turkmath.org

Pi-sigma neural network was first introduced by Shin and Ghosh in 1991. Pi-sigma is a higher order neural network, which employs additive and multiplicative aggregation functions. It is well known that pi-sigma neural network has a good function approximation performance. Pi-sigma neural network has been successfully applied in forecasting. When it is used as a forecasting technique, lagged values of the time series are used as inputs to a neural network. In time series analysis, moving average models are frequently used stochastic models where lagged values of the errors are employed as inputs. In this study, a new recurrent Pi-sigma neural network (RPiS-ANN) is proposed with inputs being the lagged values of both the time series and errors. A learning algorithm based on particle swarm optimization for the new network is introduced. The performance of RPiS-ANN is compared with well known neural networks for illustration.