Modified Sigma-Pi Neural Network for Dynamical System Identification

This study introduces a newly developed Modified Sigma-Pi Neural Network (MSPNN) architecture designed to identify the dynamics of the plant. The proposed structure is an extension of the classical sigma-pi neural network (SPNN) and includes a self-feedback loop at the output node of the output layer. Since sigma-pi artificial neural networks contain additive and multiplicative units in their structure. We apply the back-propagation optimization algorithm to train the proposed MSPNN model. The results of the proposed neural network are compared with preexisting models, such as the Jordan neural network, the feedforward neural network, and the sigma-pi neural network. An analysis of the simulation data indicates that the proposed architecture of MSPNN is more effective than the other models.