| DC 欄位 |
值 |
語言 |
| DC.contributor | 太空科學與工程學系 | zh_TW |
| DC.creator | 蔡孟融 | zh_TW |
| DC.creator | Meng-Jung Tsai | en_US |
| dc.date.accessioned | 2025-1-13T07:39:07Z | |
| dc.date.available | 2025-1-13T07:39:07Z | |
| dc.date.issued | 2025 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:444/thesis/view_etd.asp?URN=111623003 | |
| dc.contributor.department | 太空科學與工程學系 | zh_TW |
| DC.description | 國立中央大學 | zh_TW |
| DC.description | National Central University | en_US |
| dc.description.abstract | 極光電噴流(AE)指數通常用於監測極區磁場的擾動程度，可以藉助其發
展程度來判斷磁副暴發生過程，也可以透過AE指數的數值來了解太陽風、磁層以及電離層耦合狀況。本論文使用了1996年至2019年期間，解析度為一分鐘和一小時的太陽風和地磁指數資料來預測可能的地磁變化為提升預測準確度，本研究引入了多種太陽風磁層耦合函數作為輸入參數，包含太陽風能量參數、磁層頂參數、磁重聯率(Rcs)以及極冠指數(PC)，這些參數提供了磁層受到太陽風擾動後的額外資訊，比起只使用太陽風基本參數的多數先前研究，更能有效地預測AE指數?然而加入時間參數，包含年積日還有當地時間更能示AE年日變化。為解決先前研究時間延遲問題，本研究利用LightGBM，一種基於決策樹的梯度提升架構，建立了兩種AE預測模型：ModelL利用滑動窗口建立延遲特徵作為輸入；ModelN 則直接使用耦合與時間參數作為輸入。並透過ModelL 的延遲特徵重要性分析來改進ModelN 參數的輸入時間。結果顯示，在小時尺度下，ModelN 表現最佳(C.C.=0.93)；在分鐘尺度下，ModelL表現最佳(C.C.=0.89)。證實LightGBM模型能有效預測AE指數。此外，本研究發現PC和Rcs在AE指數預測中扮演重要角色，而Rcs具有替代PC的潛力。 | zh_TW |
| dc.description.abstract | The Auroral Electrojet (AE) indices were created for monitoring geomagnetic field disturbances in the auroral zone. Their variations can be used to determine the development stages of substorms. We can use AE to understand the solar wind-magnetosphere-ionosphere (S-M-I) coupling driven by solar wind forcing. This study uses 1-min and 1-hr solar wind parameters and geomagnetic indices from 1996 to 2019 to construct two AE-predicting models. To improve their prediction accuracy, extra solar wind coupling functions introducedinto the models used as input parameters, including the solar wind energy in put functions, the magnetopause shape parameters, magnetic reconnection rate
(Rcs), and the polar cap index (PC). These functions consider the effects of solar wind in the magnetosphere, enabling them more efficient in predicting AE than the previous models that used only the solar wind parameters as inputs. In addition, incorporating time parameters such as day of year (DOY) and hour allows us to adapt the annual and diurnal variations of AE. Two models were developed in this study: ModelL, which incorporates ”Lagged features”? and ModelN, whichdirectly inputs the coupling and time parameters without includeing Lagged features. Meanwhile, the importance analysis of ModelL’s lagged features is used to determine the lag time for ModelN’s parameters. The results show that ModelN performs best (with a correlation coefficient of 0.93) at 1-hr resolution, while ModelL performs best (with a correlation coefficient of 0.89) at the 1-min resolution, showing that LightGBM is suitable for predicting AE. Furthermore, PC and Rcs are found to play an important roles in predicting AE, and Rcs has the potential to replace PC. | en_US |
| DC.subject | AE指數 | zh_TW |
| DC.subject | 太陽風-磁層-電離層耦合 | zh_TW |
| DC.subject | 機器學習 | zh_TW |
| DC.subject | 磁層頂 | zh_TW |
| DC.subject | 磁重聯 | zh_TW |
| DC.subject | AE indices | en_US |
| DC.subject | solar wind-magnetosphere-ionosphere coupling | en_US |
| DC.subject | Machine Learning | en_US |
| DC.subject | Magnetopause | en_US |
| DC.subject | Magnetic Reconnection | en_US |
| DC.title | 基於機器學習LightGBM架構預測AE地磁指數 | zh_TW |
| dc.language.iso | zh-TW | zh-TW |
| DC.title | Forecasting the AE Geomagnetic Indices with the Machine Learning LightGBM Framework | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |