Modelling power output at nuclear power plant by neural networks

Abstract

In this paper, we propose two different neural network (NN) approaches for industrial process signal forecasting. Real data is available for this research from boiling water reactor type nuclear power reactors. NNs are widely used for time series prediction, but it isn't utilized for Olkiluoto nuclear power plant (NPP), Finland. Preprocessing, suitable input signals and delay analysis are important phases in modelling. Optimized number of delayed input signals and neurons in hidden-layer are found to make possible prediction of idle power process signal. It is mainly concentrated on algorithms on input signal selection and finding the optimal model for one-step ahead prediction. © 2010 Springer-Verlag Berlin Heidelberg.