Optimization and Simulation of Evaluating the Impact of Reactivity Changes in a Typical Pressurized Water Reactor Core with Artificial Neural Networks

Abstract

Pressurized Water Reactors (PWRs) are integral to the nuclear energy sector, necessitating precise modeling of their dynamic behavior for improved safety and performance. This study introduces a novel approach to simulating PWR dynamics using Artificial Neural Networks (ANNs) optimized by three methods: Levenberg-Marquardt (trainlm), Gradient Descent with Momentum (traingdm), and the metaheuristic Whale Optimization Algorithm (WOA). The ANN model is trained on differential equations characterizing PWR dynamics and is evaluated under various reactivity scenarios to analyze key parameters, including thermal and hydrodynamic parameters. The results demonstrate that the optimization method significantly affects the ANN's performance. WOA outperforms other techniques, achieving the lowest mean squared error (0.0018), highest prediction accuracy (99.1%), and faster convergence for complex reactor scenarios. Furthermore, the innovative integration of WOA provides robust predictions for reactivity-induced variations, emphasizing its superiority in optimizing ANN models for real-time applications. This research uniquely combines reactor physics and chemical engineering principles, offering a comprehensive analysis of how reactivity changes influence dynamic and chemical behavior in PWRs. By bridging these domains, the study highlights the potential of advanced machine learning methods in enhancing reactor safety and efficiency under diverse operational conditions.