A novel type of trigonometric neural network trained by Extended Kalman Filter

In this study, a new type of trigonometric neural network is presented by adding frequency and phase to trigonometric activation functions. The proposed trigonometric neural network has more flexibility in comparison with conventional trigonometric neural networks and even other types of neural networks. Due to the low convergence rate and high posibility of trapping in a local minimum of backpropagation algorithm, Extended Kalman Filter algorithm is used to train the neural network´s parameters which they appear in a nonlinear form. The Simulation of the suggested neural network based on the prediction of Mackey-Glass time series and identification of a nonlinear dynamic ystem reveals the efficiency of the proposed network. To show the efficiency of this method, the results are compared with the results of the others.