Neural network adjustment of characteristics calculated for a power reactor with a daily load schedule

Abstract

The present paper considers the possibility of improving the neural network forecast for a power reactor operating in a daily load schedule. We have prepared two simplified one-dimensional models of a VVER reactor: one as the reactor itself and another as its calculation model including several types of deviations from the reactor model for simulating the calculation error. A simple single-layer neural network is trained by comparing data obtained from the calculation and reactor models. The trained neural network effectively refines the results of the calculated forecast for the reactor model beyond the training time interval.