Application of the machine learning methods for development of zero-power reactor simulator

点击翻译按钮获取中文摘要

This research program explores the integration of Physics Informed Neural Networks (PINNs) into reactor simulations, aiming to revolutionize traditional modeling and simulation approaches. The project seeks to leverage machine learning methods to construct reactor models that can be tested against conventional simulation tools. By employing PINNs, the reliance on vast amounts of data for pure Neural Networks is mitigated, as PINNs incorporate well-established physical equations to enhance predictive capabilities. This innovative approach not only reduces the burden of data collection but also presents opportunities for creating virtual reality environments for reactor training and operation. The primary objective of this PhD project is to demonstrate the efficacy of machine learning techniques, specifically PINNs, in reactor modeling, comparing their performance against traditional simulation methods. Success in this, promises to simplify and expedite the development of reactor simulators by integrating physical models and experimental data into the learning process. Moreover, the proposed approach aims to define a unified framework for the development of zero-power simulators, containing both current and future applications. Through this research, the potential of machine learning in advancing reactor simulation technology is explored, with implications for enhancing safety, efficiency, and more importantly accessibility in nuclear engineering.