Application of neural networks to analyze load-follow operation in a pressurized water reactor

A new analytic model based on hidden-layer neural networks is designed to a nalyze load-follow operation in a pressurized wafer reactor (PWR). The new model is mainly made up of four error backpropagation neural networks and p rocedures to calculate core parameters such as k(infinity) and xenon distri butions ill a transient core. The first two neural networks are designed to retrieve the power distribution, the third is for axial offset, and the fo urth is for reactivity corresponding to a given core condition. The trainin g data sets are generated by three-dimensional nodal code and the measured data of the first-day load-follow operation. The simulation results of the 5-day lend-follow test in a PWR using the new analytic model show that it i s an attractive tool for plant simulations in terms of accuracy, computing time, cost, and adaptability to measurements.