探討特徵選取法應用於飛機油耗效能預測之研究;Exploring Feature Selection Techniques for Predicting Aircraft Fuel Efficiency Performance

NCU Institutional Repository > 管理學院 > 資訊管理學系碩士在職專班 > 博碩士論文 > Item 987654321/95491

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/95491

题名:	探討特徵選取法應用於飛機油耗效能預測之研究;Exploring Feature Selection Techniques for Predicting Aircraft Fuel Efficiency Performance
作者:	王柔蘋;Wang, Jou-Ping
贡献者:	資訊管理學系在職專班
关键词:	特徵選取;回歸預測;飛機性能監控;引擎衰退;Feature Selection;Regression Prediction;Aircraft Performance Monitoring;Engine Degradation
日期:	2024-07-10
上传时间:	2024-10-09 16:54:13 (UTC+8)
出版者:	國立中央大學
摘要:	在本研究中探討了航空公司面對高燃油成本的挑戰及其在全球經濟中的關鍵角色，如何透過精準的燃油管理和飛機性能監控來優化操作效率。引入預測性健康管理(PHM) 和飛機性能監控計劃(APM)觀念，並透過監測設備實時收集飛機性能數據並比較實際性能與標準性能，識別性能差異，提供對飛機當前狀況的評估。本文主要研究動機源於現有飛機性能監控計畫收集的大量數據，這些數據指出了飛機性能的退化，但未能明確指出關鍵影響因素。透過資料探勘技術，我們旨在識別出影響飛機性能衰退的關鍵特徵，從而提供更精確的維護建議和操作優化策略。為達上述目的，本研究進行三個不同主題的對照實驗，飛機機齡、飛行航程長度和季節，來探究其對飛機性能的影響。研究中採用過濾器、包裝器和內嵌法這三種特徵選取模型進行比較，並搭配隨機森林回歸與支持向量回歸預測方法，來確定關鍵特徵和評估模型表現。研究結果展示了不同特徵選取方法與預測模型組合對改善航機燃油效率預測模型的影響。本研究證實上述方法能減少模型的複雜性，同時維持或提升預測的精確度。高維度的 APM 資料保持較高的預測精度，但使用較少的關鍵特徵同樣能夠達到接近完整數據集的預測效果，降低資料處理的複雜度並提升分析效率。本研究提供了一個全面的方法來分析和優化飛機性能，並有效識別出影響飛機性能的關鍵特徵，旨在幫助航空公司更有效地管理燃油成本和提高運營效率，同時確保飛行的安全性。;This study investigates the impact of high fuel costs on airlines and their significant role in the global economy, focusing on enhancing operational efficiency through precise fuel management and aircraft performance monitoring. We introduce Predictive Health Management (PHM) and Aircraft Performance Monitoring (APM), utilizing real-time monitoring tools to gather data and identify discrepancies by comparing actual aircraft performance against benchmarks. Motivated by the extensive data from aircraft performance monitoring, which highlights performance degradation without clarifying the causes, we employ data mining techniques to pinpoint key factors affecting aircraft performance. This enables more precise maintenance recommendations and optimization of operations. Our research involves controlled experiments related to aircraft age, flight distance, and seasonality to assess their effects on performance. We analyze the efficacy of three feature selection models—filter, wrapper, and embedded methods—combined with Random Forest and Support Vector Regression to identify critical performance features. The research results show that these methods reduce model complexity while maintaining or enhancing accuracy. High-dimensional APM data provides high prediction accuracy, but using fewer key features achieves similar results, reducing data processing complexity and improving analysis efficiency. This study aims to help airlines manage fuel costs effectively and enhance operational efficiency while prioritizing safety, providing a thorough approach to aircraft performance optimization.
显示于类别:	[資訊管理學系碩士在職專班 ] 博碩士論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	42	检视/开启

在NCUIR中所有的数据项都受到原著作权保护.

社群 sharing

数据加载中.....