Predictive modeling of pressurized water reactor transients using nonlinear autoregressive with exogenous input neural network

Abstract

A pressurized water reactor (PWR) is a multivariable system that consists of several subsystems such as a core, steam generator, pipings, and plenums that show highly nonlinear behavior. These components are prone to critical parameters that cause potential accidents and propagate to the entire system. Therefore, the PWR needs continuous and fast pre-accident assessment through modeling and predicting by leveraging powerful intelligence technologies. In this regard, this study demonstrates the potential of the nonlinear autoregressive with exogenous inputs (NARX) neural network modeling approach. The established NARX model is evaluated using PWR transients under perturbations in control rod velocity and core inlet coolant temperature. The trained network was then applied for state estimation, future prediction, and fault detection. The simulation results verified that the NARX model predicts the target value with good accuracy as compared with the nonlinear input-output (NIO) neural network strategy. Further, it is used to forecast the step-ahead values of the targets in the defined confidence interval and fault detection. Overall, the present study gives the benefit of the NARX approach for estimation and fault detection applications in other nuclear engineering fields.