Empirical Steam Generator Water-Level Modeling Using Neural Networks

Abstract

Neural networks such as the radial basis function network, adaptive neuro-fuzzy inference systems, and the multilayer feedforward neural network were adopted to model the steam generator water level, which was intended to be the analytic redundancy in the signal validation system. The training data were the simulation results of the small-demand turbine power variations around the steady state. The test data were from two small-load maneuvers: the load reduction from 100 to 50% of the rated power, and one feedwater pump trip event. The network training required only a short time, and the simulation results show that the neural networks are suitable for the modeling of steam generator water level.