Survey on the Use of Artificial Intelligence in Nuclear Power Plants

Noh, Hyun Seok; Kim, Jung Soo; Jung, Woo Sik

Permalink

https://hdl.handle.net/2142/121797 Copy

Description

Title

Survey on the Use of Artificial Intelligence in Nuclear Power Plants

Author(s)

• Noh, Hyun Seok
• Kim, Jung Soo
• Jung, Woo Sik

Issue Date

2023

Keyword(s)

• Nuclear power plant
• Artificial Intelligence (AI)
• Supervised learning
• Unsupervised learning
• Reinforced learning

Abstract

According to the development of artificial intelligence technology, various fields are introducing artificial intelligence technology. In addition, AI research is also being actively conducted in the field of nuclear power generation. However, there are no papers published to date that analyze and summarize the cases of applying artificial intelligence technology to nuclear power generation, making it difficult for researchers to apply artificial intelligence technology. This paper introduces the studies that have applied artificial intelligence technology to the field of nuclear power generation to date, and summarizes the achievements and limitations of these studies. In addition, for researchers who want to apply artificial intelligence technology to, the existing studies are classified into three categories: nuclear power generation, training data, and learning algorithms. The first classification is based on the field of nuclear power generation. Nuclear power generation are categorized into diagnosis, prediction, response, and process. Diagnosis is applied to the detection of abnormalities in nuclear power plant devices. Prediction is used to help prevent accidents by predicting transient conditions or severe accidents in nuclear power plants. Response is applied to real-time risk assessment and emergency response in the event of a severe accident in a nuclear power plant. Process is the optimization of the design and operation of nuclear power plants. The second classification is based on the training data. The types of data are generally divided into structured, unstructured, and time series data. The structured data of nuclear power plants are values such as reactor temperature, pressure, density, flow rate, concentration, etc. Unstructured data of nuclear power plants are in the form of images, videos, audio, etc. such as non-destructive testing data and radiation measurement data. Time series data is data listed in chronological order. The trend of electricity production and equipment operation status of nuclear power plants are examples of time series data. The third is classification based on learning algorithms. Classification by learning algorithm is divided into supervised learning, unsupervised learning, and reinforcement learning. A representative example of supervised learning is learning vibration data to detect abnormal condition of primary coolant pump and determine whether the pump is abnormal or not. A typical example of unsupervised learning is learning key variables of transient conditions in nuclear power plants to predict the trend of subsequent transient conditions or severe accidents. A typical example of reinforcement learning is the autonomous calibration of key parameters of a nuclear power plant using a computerized digital twin that simulates a 100% real nuclear power plant. This paper analyzed and classified the existing studies and compared the advantages and disadvantages of previous studies for researchers who want to apply new AI techniques to the nuclear power generation. We hope to help the future AI researcher in the field of nuclear power generation.

Type of Resource

text

Language

eng

Handle URL

https://hdl.handle.net/2142/121797

Owning Collections