Recurrent Sigmoid-Wavelet Neurons for Forecasting of Dynamic Systems

In this paper, recurrent neuron models used in feed-forward network are proposed. Each neuron in this model is composed of the sigmoidal activation function (SAF) and wavelet activation function (WAF). The output of the proposed neuron is the product of output from SAF and WAF. In recurrent neuron models delayed output of the sigmoidal and the wavelet activation function is feedback to each other. Performance of the recurrent models is evaluated on two different kind of benchmark problem of dynamical systems and compared with earlier proposed models.