請用此 Handle URI 來引用此文件：
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/98940完整後設資料紀錄
| DC 欄位 | 值 | 語言 |
|---|---|---|
| dc.contributor.advisor | 呂良正 | zh_TW |
| dc.contributor.advisor | Liang-Jenq Leu | en |
| dc.contributor.author | 林承翰 | zh_TW |
| dc.contributor.author | Cheng-Han Lin | en |
| dc.date.accessioned | 2025-08-20T16:22:07Z | - |
| dc.date.available | 2025-08-21 | - |
| dc.date.copyright | 2025-08-20 | - |
| dc.date.issued | 2025 | - |
| dc.date.submitted | 2025-08-15 | - |
| dc.identifier.citation | [1] 混凝土工程設計規範與解說（土木401–112）. 2023. In chinese.
[2] R. Caruana. Multitask learning. Machine Learning, 28(1):41–75, 1997. [3] Y.-C. Hsieh. Application of deep learning to assist in database creation and beam section design in rc building. Master’s thesis, National Taiwan University, 2024. In Chinese. [4] M.-H. Li. Application of machine learning in predicting compressive strength of concrete and design of beam and column for rc building. Master’s thesis, National Taiwan University, 2021. In Chinese. [5] M.-H. Li. Applying deep learning to assist the design of column section in rc building. Master’s thesis, National Taiwan University, 2022. In Chinese. [6] Q.-E. Li. Applying deep learning to predict the column section design of rc building. Master’s thesis, National Taiwan University, 2023. In Chinese. [7] S. M. Lundberg and S.-I. Lee. A unified approach to interpreting model predictions. In Proceedings of the 31st International Conference on Neural Information Processing Systems, NIPS’17, page 4768–4777, Red Hook, NY, USA, 2017. Curran Associates Inc. [8] D. E. Rumelhart, G. E. Hinton, and R. J. Williams. Learning representations by back-propagating errors. Nature, 323(6088):533–536, 1986. [9] L. S. Shapley. A Value for N-Person Games. RAND Corporation, Santa Monica, CA, 1952. [10] A. M. TURING. I.—computing machinery and intelligence. Mind, LIX(236):433–460, 10 1950. | - |
| dc.identifier.uri | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/98940 | - |
| dc.description.abstract | 本研究開發一套以深度學習為基礎的預測系統，用於結構設計中梁柱斷面尺寸的預測。研究中採用多任務學習（MTL）模型同時預測結構元件在兩個方向上的尺寸，並與傳統單任務模型進行比較。實驗結果顯示，MTL 模型能有效捕捉兩個方向間的相依性，並提升整體預測表現。此外，本研究針對結構元件的特性，設計了多組兼具物理意義與工程背景的輸入特徵，並提出客製化的損失函數以強化模型對尺度差異的適應能力。而本研究設計加入的特徵多符合預期表現，影響模型表現。
為增進模型可解釋性，本研究引入 SHAP（SHapley Additive exPlanations）方法進行特徵重要度分析，探討各特徵在不同預測階段對模型判斷的貢獻。透過實例驗證顯示，SHAP 分析所得的重要特徵與結構設計的工程直覺具有一致性，具實務應用價值。最後，研究亦實作出一套友善的使用者介面，整合模型預測與資料輸入流程，降低應用的門檻。 | zh_TW |
| dc.description.abstract | This study develops a deep learning–based prediction system for estimating the cross-sectional dimensions of structural beams and columns. A multi-task learning (MTL) model is adopted to simultaneously predict the dimensions of structural components in two orthogonal directions, and its performance is compared with that of traditional single-task models. Experimental results indicate that the MTL model effectively captures the interdependence between the two directions, leading to improved overall predictive accuracy.
To reflect the physical and engineering characteristics of structural components, this study designs multiple input features grounded in domain knowledge and proposes a customized loss function to enhance adaptability to scale variations. Most of the engineered features demonstrate expected performance to the model's predictions. To improve interpretability, SHAP is employed for feature importance analysis, showing whether features are consistent with engineering intuition. Finally, a user-friendly interface is implemented to integrate data input and model prediction. | en |
| dc.description.provenance | Submitted by admin ntu (admin@lib.ntu.edu.tw) on 2025-08-20T16:22:07Z No. of bitstreams: 0 | en |
| dc.description.provenance | Made available in DSpace on 2025-08-20T16:22:07Z (GMT). No. of bitstreams: 0 | en |
| dc.description.tableofcontents | Acknowledgements i
摘要iii Abstract v Contents vii List of Figures xiii List of Tables xix Chapter 1 Introduction 1 1.1 Research Purpose and Motivation 1 1.2 Literature Review 4 1.3 Thesis Organization 7 Chapter 2 Methodology 9 2.1 Introduction 9 2.2 Artificial Intelligence and Machine Learning 9 2.2.1 Artificial Intelligence(AI) 9 2.2.2 Machinie Learning(ML) 11 2.3 ANN 13 2.4 Multi-task Learning(MTL) 14 2.4.1 Concept and benefits of multi-task learning 14 2.4.2 Parameters sharing strategies in MTL 15 2.5 Activation function 16 2.6 Loss function 17 2.6.1 Customized loss function based on structural code 18 2.7 Evaluation methods 21 2.7.1 Permutaion feature importance 21 2.7.2 SHAP 22 2.7.2.1 Shapley values 22 2.7.2.2 SHAP values 25 2.7.3 Comparison of the two methods 29 2.8 Summary 29 Chapter 3 Dataset and Features 31 3.1 Introduction 31 3.2 Database Assumptions 32 3.3 Generate Dataset 34 3.3.1 Dataset Structure 34 3.3.2 Derive Data from Etabs 36 3.4 Database Features 39 3.4.1 Feature Engineering 44 3.4.2 Structure-Related Features 46 3.4.3 Floor-Related Features 49 3.4.3.1 Height Related 49 3.4.3.2 Weight related 50 3.4.3.3 Shear related 52 3.4.3.4 Concrete Related 53 3.4.3.5 Plan Related 54 3.4.4 Column-Related Features 55 3.4.5 Girder-Related Features 64 3.5 Dataset Partitioning 67 Chapter 4 Prediction of Column Cross-Sectional Dimensions 71 4.1 Introduction 71 4.2 Prediction Targets 72 4.3 Data Preprocessing 75 4.3.1 Column dimension combinations 75 4.3.2 Elimination of outliers 78 4.3.3 Data standardization 80 4.4 Model Evaluation Metrics 81 4.5 Model Introduction 82 4.6 Loss function 83 4.7 Training Model 84 4.7.1 Training Process 84 4.7.2 Initial Used Features 86 4.7.3 Training Model 88 4.7.4 Selecting Features 91 4.7.5 Training Model After Selecting Features 92 4.8 Feature Importance discussion 97 4.9 Discussion About STL and MTL 101 4.10 Summary 102 4.10.1 Challenges and solutions in training models with clustered datasets 102 Chapter 5 Prediction of Girder Cross-Sectional Dimensions 107 5.1 Introduction 107 5.2 Prediction Targets 107 5.3 Data Preprocessing 109 5.3.1 Beam dimension combinations 109 5.3.2 Elimination of outliers 110 5.3.3 Data standardization 111 5.4 Model Evaluation Metrics 111 5.5 Model Introduction 111 5.6 Loss function 112 5.7 Training Model 112 5.7.1 Training process 112 5.7.2 Initial Used Features 112 5.7.3 Training Model 114 5.7.4 Selecting Features 116 5.7.5 Training Model After Selecting Features 117 5.8 Feature Importance discussion 119 5.9 Summary 122 Chapter 6 Evaluate Test Structural 123 6.1 Introduction 123 6.2 Test Structure : 10010C 124 6.2.1 In column view 128 6.2.2 In girder view 129 6.3 Discussion of all test structures 138 6.4 Summary 139 Chapter 7 Software Development and Application 145 7.1 Introduction 145 7.2 Manual 146 7.2.1 User input interface 146 7.2.2 Verification and feature calculation 153 7.2.3 Prediction and evaluation 156 7.2.4 Conclusion 157 Chapter 8 Conclusions 159 8.1 Future Work 161 References 165 Appendix A — Illustration of gradient explosion using one-hot encoded feature 167 A.1 A.1 Model structure and background 167 A.2 A.2 One-hot encoded feature setup 168 A.3 A.3 Forward and backward computation 170 A.4 A.4 Observation and explanation 173 | - |
| dc.language.iso | en | - |
| dc.subject | 結構斷面設計 | zh_TW |
| dc.subject | 多任務學習 | zh_TW |
| dc.subject | 軟體開發 | zh_TW |
| dc.subject | 自訂損失函數 | zh_TW |
| dc.subject | structural cross-section design | en |
| dc.subject | multi-task learning | en |
| dc.subject | software interface development | en |
| dc.subject | customized loss function | en |
| dc.title | 應用多任務機器學習於結構物梁柱斷面預測與應用 | zh_TW |
| dc.title | Application of Multi-Task Machine Learning for Structural Beam and Column Section Prediction and Applications | en |
| dc.type | Thesis | - |
| dc.date.schoolyear | 113-2 | - |
| dc.description.degree | 碩士 | - |
| dc.contributor.oralexamcommittee | 黃仲偉;郭世榮;柯俊宇 | zh_TW |
| dc.contributor.oralexamcommittee | Chang-Wei Huang;Shyh-Rong Kuo;Chun-Yu Ke | en |
| dc.subject.keyword | 多任務學習,結構斷面設計,自訂損失函數,軟體開發, | zh_TW |
| dc.subject.keyword | multi-task learning,structural cross-section design,customized loss function,software interface development, | en |
| dc.relation.page | 173 | - |
| dc.identifier.doi | 10.6342/NTU202504218 | - |
| dc.rights.note | 同意授權(全球公開) | - |
| dc.date.accepted | 2025-08-15 | - |
| dc.contributor.author-college | 工學院 | - |
| dc.contributor.author-dept | 土木工程學系 | - |
| dc.date.embargo-lift | 2025-08-21 | - |
| 顯示於系所單位： | 土木工程學系 | |
文件中的檔案：
| 檔案 | 大小 | 格式 | |
|---|---|---|---|
| ntu-113-2.pdf | 36.23 MB | Adobe PDF | 檢視/開啟 |
系統中的文件，除了特別指名其著作權條款之外，均受到著作權保護，並且保留所有的權利。