Gradient descent-particle swarm optimization based deep neural network predictive control of pressurized water reactor power

Abstract

A pressurized water reactor (PWR) is an integrated system of various interdependent subsystems that show highly nonlinear behavior, and each component is prone to uncertainties and malfunctions that initiate potential accidents. In contrast, PWR needs to be continuously monitored for stable and safe operation over a service time. In this study, a gradient descent-particle swarm optimization hybrid algorithm-based deep neural network (GD-PSO-based DNN) approach is proposed to monitor the PWR core power and outlet temperature. The Gaussian noise is introduced to the input signal, and the PWR core stability conditions are examined by using Lyapunov analysis. The simulation results verified under different conditions, in which the proposed control approach tracks the reference inputs successfully and enhances the stability as compared with the sliding mode control, linear quadratic regulator, and proportional-integral-derivative methods. The controller is validated by using the data from RELAP5 system codes, and the performance is examined using the mean square error, integral square error, integral absolute error, integral time square error, and integral time absolute error criterion functions. This study gives the benefit to apply the GD-PSO-based DNN control method for other nuclear engineering fields for system identification, prediction, and control.