PSAM 16 Conference Paper Overview
Welcome to the PSAM 16 Conference paper and speaker overview page.
Lead Author: Gulcin Sarici Turkmen Co-author(s): Alper Yilmaz yilmaz.15@osu.edu
Tunc Aldemir aldemir.1@osu.edu
| USE OF MACHINE LEARNING TECHNIQUES TO REDUCE THE COMPUTATIONAL EFFORT FOR DYNAMIC PROBABILISTIC RISK ASSESSMENT |
|---|
| Dynamic probabilistic risk assessment (DPRA) is an important approach for assessing safety of nuclear power plant (NPP) operation. Since NPPs are highly complex systems, it is necessary to produce large amounts of data that represent different possible situations during NPP evolution following an accident in order to carry out the DPRA comprehensively. In addition to the fact that it may take months to produce such data with NPP accident analysis codes (e.g, RELAP5, MELCOR) and DPRA software developed for such a task (e.g., ADAPT), the task also requires the use of significant amount of computer and human resources. The effectiveness of models that have been developed using Machine Learning (ML) techniques in recent years, especially those that can represent time-dependent data and make predictions for the consequences of possible initiating events, have proven to be useful in reducing the computational effort for such a task. Recurrent Neural Network (RNN) approach represent some efficient ML methods that can be used in modeling the development of accidents and predicting potential consequences. This study is aimed at using DPRA data sets for two different NPPs to train RNN model to make predictions of possible NPP behavior under accident conditions as the accident evolves. The data sets to be used have been obtained from previous studies. The first data set consists of about 10,000 scenarios generated for a 4-loop pressurized water reactor (PWR) with station blackout (SBO) as the initiating event using RELAP5-3D/RAVEN, and the second data set consists of 2,656 scenarios generated for a 3-loop PWR under SBO using MELCOR/ADAPT. After developing deep learning (DL) model trained with the data for one power plant, the same DL model will be retrained with the data of the second power plant to apply the Transfer learning (TL) approach. An important advantage of TL is that it would save time and resources for a more comprehensive DPRA, as well as making make more accurate predictions under accident conditions. |
Paper SA202 Preview
Author and Presentation Info
Presentation only, a full paper is not available.
Lead Author Name: Gulcin Sarici Turkmen (sariciturkmen.1@osu.edu)
Bio: Gulcin Sarici Turkmen got her Bachelor of Science and Master of Science degrees from the Department of Nuclear Engineering at Hacettepe University in Turkey. In Master of Science, she focused on modelling “Effect of Thermal-Neutronic Coupling on the Cross-Sections of Nuclear Fuel”. She also worked as a project manager in a private company in Turkey and worked the system modelling of VVER-1200. She was awarded a doctoral scholarship by the Turkish government and now she is a PhD student at the Ohio State University and she is doing her research on multi-unit dynamic probabilistic risk assessment for small modular reactors with Prof. Tunc Aldemir.
Country: United States of America
Company: The Ohio State University
Job Title: Graduate Research Associate
Bio: Gulcin Sarici Turkmen got her Bachelor of Science and Master of Science degrees from the Department of Nuclear Engineering at Hacettepe University in Turkey. In Master of Science, she focused on modelling “Effect of Thermal-Neutronic Coupling on the Cross-Sections of Nuclear Fuel”. She also worked as a project manager in a private company in Turkey and worked the system modelling of VVER-1200. She was awarded a doctoral scholarship by the Turkish government and now she is a PhD student at the Ohio State University and she is doing her research on multi-unit dynamic probabilistic risk assessment for small modular reactors with Prof. Tunc Aldemir.
Country: United States of America
Company: The Ohio State University
Job Title: Graduate Research Associate