DC 欄位 值 語言 DC.contributor 土木工程學系 zh_TW DC.creator 邱鈺智 zh_TW DC.creator Yu-Zhi Qiu en_US dc.date.accessioned 2022-9-19T07:39:07Z dc.date.available 2022-9-19T07:39:07Z dc.date.issued 2022 dc.identifier.uri http://ir.lib.ncu.edu.tw:444/thesis/view_etd.asp?URN=109322082 dc.contributor.department 土木工程學系 zh_TW DC.description 國立中央大學 zh_TW DC.description National Central University en_US dc.description.abstract 隨著氣候的劇烈變遷，山崩破壞也越加嚴重，因此如果能預測山崩就能減少人員傷亡以及財產損失，而現今山崩預測研究中又以深度學習（Deep Learning）的成長最為顯著，但利用深度學習預測山崩上有一困難點為山崩監測資料較為缺乏，如地下水位、邊坡位移量獲取相對於其他水文或氣象資料更難以取得。而在深度學習模型的訓練過程中，是需要有大量的資料才能有良好的預測結果，故本研究同步提出使用生成對抗網路（Generative Adversarial Networks, GAN），以2007-2008年廬山監測資料為範例進行資料擴增，並建置支持向量機（Support Vector Machine, SVM）、長短期記憶網路（Long Short-Term Memory Network, LSTM）和門閥限循環單元網路（Gated Recurrent Unit Network, GRU）三種預測模型，除探討利用原始資料進行訓練預測外，本研究並利用資料擴增後之資料集探討預測結果。實驗顯示使用原始資料且預測成效較差的水位(孔位B01、B04、B07)，其GRU預測R-Square值分別為 0.690、0.347、0.759，再使用GAN資料擴增後，B01、B04、B07 GRU預測R-Square值提升至0.943、0.901、0.760。最後使用文獻兩組山崩實際案例進行比較可得知，在資料擴增後預測山崩位移皆有更佳預測結果。本研究所提出深度學習應用可應用對於缺乏山崩資料而想使用深度學習預測之參考依據。 zh_TW dc.description.abstract With the dramatic climate changes, landslides damage has become more serious. Therefore, if landslides can be successfully predicted, casualties and property damage can be reduced. For various landslide predictions, the growth of Deep Learning is the most significant. However, a major difficulty in landslide prediction is obtaining landslide monitoring data, such as groundwater level and slope displacement. They are more challenging to obtain than other hydrological or meteorological data. In the deep learning model training process, a large amount of data is necessary for good prediction results. Therefore, this study proposes to combine Generative Adversarial Networks, using data augmentation with 2007-2008 Lushan monitoring data as an example, and build three prediction models: Support Vector Machine (SVM), Long Short-Term Memory(LSTM), and Gated Recurrent Unit (GRU), to compare whether the augmented data has more predictive results than using the original data. Experiment results show that GRU with original data can predict water levels and R-Square values of B01, B04, and B07 are 0.690, 0.347, and 0.759. Using GAN data augmentation for water levels has the predicted R-Square values of B01, B04, and B07 of 0.943, 0.901, and 0.760. The latter one has significantly improved. This study further evaluates deep learning applications that can be applied in other cases. Consequently, the proposed GRU with the GAN method is a feasible approach for landslide prediction. en_US DC.subject 深度學習 zh_TW DC.subject 山崩預測 zh_TW DC.subject 資料擴增 zh_TW DC.subject 生成對抗網路 zh_TW DC.subject 支持向量機 zh_TW DC.subject 長短期記憶神經網路 zh_TW DC.subject 門閥循環單元網路 zh_TW DC.subject Deep Learning en_US DC.subject Landslide Prediction en_US DC.subject Data Augmentation en_US DC.subject GAN en_US DC.subject SVM en_US DC.subject LSTM en_US DC.subject GRU en_US DC.title 深度學習與資料擴增於山崩監測預測之可行性評估 zh_TW dc.language.iso zh-TW zh-TW DC.title Development of Deep learning and GAN applied to landslide prediction en_US DC.type 博碩士論文 zh_TW DC.type thesis en_US DC.publisher National Central University en_US