應用 PCTRAN核電廠事故模擬警報系統和深度學習進行警報洪水分類與預診斷和健康管理;Using PCTRAN Nuclear Power Plant Accident Simulation Alarm System with Deep Learning for Alarm Flood Classification and Prognostics and Health Management

NCUIR > School of Management at National Central University > Graduate Institute of Industrial Management > Electronic Thesis & Dissertation > Item 987654321/97258

Please use this identifier to cite or link to this item: https://ir.lib.ncu.edu.tw/handle/987654321/97258

Title:	應用 PCTRAN核電廠事故模擬警報系統和深度學習進行警報洪水分類與預診斷和健康管理;Using PCTRAN Nuclear Power Plant Accident Simulation Alarm System with Deep Learning for Alarm Flood Classification and Prognostics and Health Management
Authors:	李侑芳;Li, Yu-Fang
Contributors:	工業管理研究所
Keywords:	核能電廠;PCTRAN軟體;預後與健康管理;警報洪水分類;深度學習;RNN;LSTM;GRU;Nuclear Power Plant;PCTRAN Software;Prognostic and Health Management;Alarm Flood Classification;Deep Learning;RNN;LSTM;GRU
Date:	2025-06-23
Issue Date:	2025-10-17 11:03:30 (UTC+8)
Publisher:	國立中央大學
Abstract:	由於全球暖化日益嚴重，以及台灣對高效率與高可靠度能源的需求日漸增加，核能被視為一種潛在的解決方案，可有效減少碳排放，並對抗氣候變遷。然而，核能技術在應用上的安全性仍是高度關切的議題，特別是在面對異常狀況與不可預測事故風險時。因此，針對核能電廠進行故障監測與預後健康管理（PHM）成為確保其安全運轉的必要措施。本論文旨在透過開發與評估先進的警報洪水分類（Alarm Flood Classification, AFC）方法，提升核能電廠（Nuclear Power Plants, NPP）之安全性與預測效能。為了解決現有研究中對於警報資料集缺乏的問題，本研究採用PCTRAN模擬軟體所提供之IEEE開放資料集，建立一套涵蓋多種異常情境的綜合性資料庫。此資料集能夠深入分析警報行為，並為AFC方法的效能評估提供堅實的基礎。研究中以系統化方式產生警報資料，並依據誤報率（False Alarm Rate, FAR）與漏報率（Missed Alarm Rate, MAR）之間的平衡，設置alpha參數範圍從0.00至1.00的不同警報觸發門檻值。此外，研究進一步探討深度學習技術於此類時序資料的應用，包括循環神經網路（Recurrent Neural Network, RNN）、長短期記憶網路（Long Short-Term Memory, LSTM）、門控循環單元（Gated Recurrent Unit, GRU），以及結合滑動視窗技術之LSTM與GRU模型。透過這些模型對特徵進行時序分析與擷取，以提升分類模型的準確性與穩定性。實驗結果顯示，深度學習技術中以GRU模型搭配滑動視窗方法進行警報分類效果最佳，當滑動視窗大小（window size）設為290，步長（step size）為1，並於最佳alpha值0.10下運作時，能達到99%的分類準確率，其加權平均的精確率（Precision）、召回率（Recall）與F1分數亦皆為99%。此結果顯示該方法具備即時反應能力與高分類效能，有助於提升核能電廠之整體運作可靠性，並在事故應對與設施營運成本方面達到最小化之效益。 ;Due to global warming and Taiwan’s growing need for efficient and reliable energy sources, nuclear energy is considered a potential solution for reducing carbon emissions and combating climate change. However, safety remains a critical concern due to the risks of abnormal situations and uncertain accidents. This makes monitoring and prognostic and health management (PHM) of faults essential. The goal of this thesis is to provide the safety and efficient prognostic of nuclear power plants (NPP) by developing and evaluating advanced methods for alarm flood classification (AFC). To address a gap in existing research regarding the availability of alarm datasets, a comprehensive dataset derived from simulated data of a nuclear power plant using the IEEE open dataset from PCTRAN software was utilized. This dataset enables detailed analysis of alarm behavior and provides a foundation for evaluating AFC methods. A systematic approach was used to generate alarm data, setting thresholds based on balancing false alarm rates and miss alarm rates from alpha 0.00 to 1.00. Additionally, the research investigates the application of deep learning techniques, such as Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), LSTM and GRU with sliding window function to analyze the dataset and features that improve classification model performance. In this thesis, I find that deep learning deals with alarm classification in GRU with a sliding window function with window size of 290 and a step size of 1 operating with an optimal alpha value of 0.10 achieving 99% accuracy, with weighted average precision, recall, and F1-score also at 99%. These contribute to enhancing overall NPP reliable operation through on-time handling, high classification accuracy, and minimized overall facility costs.
Appears in Collections:	[Graduate Institute of Industrial Management] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	5	View/Open

社群 sharing

Loading...