Approximating transfer functions using neural network weights

Artificial neural networks are widely used in the identification and control of complex systems. However, the network model which is based on neuron nodes, activation functions, and network weights is rarely related to the system transfer function. In this paper, a clear relationship between the network weights and the transfer function parameters is established. The developed mathematical equations are based on approximating the neuron activation function using Taylor expansion and relating the results to a linear transfer function based on Auto Regressive Moving Average model. Simulation results show that the approximated transfer function behavior resembles the original system function.