Abstract: This study is intended to propose new criteria to decide appropriate hidden layer neuron numbers in Recursive Radial Basis Function Networks (RRBFN) and successfully applied to the wind speed forecasting application in renewable energy system. Purpose of the proposed methodology eliminate both either over fitting or under fitting issues. The proper hidden layer neuron numbers is evolved through the presented 150 various criteria. Exact modeling of recursive radial basis function networks possess with three input variables using the proposed new determining criteria are validated by means of the convergence theorem. In order to verify effectiveness and generalization capability of the proposed methodology, computer simulation is carried out on two real-time data sets and selection of data influence on the results are analyzed with various training and testing data. Experiment results confirmed that the proposed criteria result better framework for recursive radial basis function networks with reduced statistical errors compared with other previous methodologies.
M. Madhiarasan and S.N. Deepa, 2016. New Criteria for Estimating the Hidden Layer Neuron Numbers for Recursive Radial Basis Function Networks and its Application in Wind Speed Forecasting. Asian Journal of Information Technology, 15: 4377-4391.