딥러닝 알고리즘을 이용한 케이블지지교량의 구조응답 패턴 분석 = Pattern analysis of structural responses of cable supported bridges based on deep learning algorithm

다국어 초록 (Multilingual Abstract)

In order to maintain the cable supported bridges in a reasonable way, long-term measurement data through the measurement system has been accumulated, but it is not utilized except to determine the abnormal signal measurement for the specific physical quantity of the main members. In addition, the long-term measurement data of cable supported bridges with 30~50 years design life is very low in terms of cost utilization.
Therefore, in this study, the deep learning algorithm DNN (Deep Neural Network), which is based on artificial neural network theory, and Long Shot-Term Memory) and Bi-LSTM(Bidirectional Long Shot-Term Memory deep learning algorithm, which are specialized for time series data analysis In order to improve the usability based on long-term measurement data, and to evaluate the condition and performance of cable supported bridges and to effectively use the change characteristics analysis, the utility of the analysis method and the predicted structural response are analyzed through a pattern analysis model using measurement data. The effectiveness of this analytical method was examined by direct comparison with measurement data. Deep learning-based pattern analysis model was constructed by using the hourly average data of (GNSS) (2016.01.26-2016.08.01) of the ○○ bridge's temperature and horizontal displacement (GNSS) at the top of the pylon. Through the deep learning based algorithm, various models were constructed to evaluate the predictive performance through RMSE(Root Mean Square Error).
The predictive performance of the deep learning based pattern analysis model was DNN (T5-HL3) = 2.675, LSTM (T5-HL1-SL7) = 1.578, and Bi-LSTM (T5-HL3-SL14) = 1.552. In case of the DNN model, the predictive performance index is lower than that of the LSTM and Bi-LSTM models, which are specialized for time series data analysis. In the case of the LSTM model, when the model is complex, the performance decreases due to overfitting. The LSTM model also showed a similar trend as the LSTM model. In this study, we developed a pattern analysis model based on the measurement data of cable supported bridges through various deep learning algorithms and directly compared with the actual measurement data to examine the effectiveness of the method and predictive patterns and quantitative figures very similar to the actual measurement data. Proved its effectiveness. Through deep learning-based pattern analysis model, the utilization of long-term measurement data is expected to improve the utilization of correction and recovery of missing section or measurement data, recognition and prediction of state change, abnormal signal, and soundness evaluation.

목차 (Table of Contents)

• 목 차 ⅰ
• 표 목 차 ⅳ
• 그림목차 ⅴ
• Ⅰ. 서 론 1
• 1. 연구배경 및 목적 1
• 2. 연구내용 및 범위 3
• Ⅱ. 이론적 배경 4
• 1. 케이블지지교량의 모니터링시스템 4
• 2. 케이블지지교량의 모니터링시스템 장기계측데이터 5
• 3. Deep Learning Algorithm 6
• 1) DNN(Deep Neural Network) 6
• 2) LSTM(Long Short-Term Memory) 8
• 3) Bi-LSTM(Bidirectional Long Short-Term Memory) 12
• Ⅲ. 케이블지지교량의 계측시스템 현황 및 기존 문헌고찰 13
• 1. 케이블지지교량의 계측시스템 운영 현황 13
• 2. 케이블지지교량의 계측시스템 기반 기존 문헌고찰 18
• 1) 장기계측데이터 분석 연구 18
• 2) Deep Learning Algorithm 기반 장기계측데이터 분석 연구 19
• Ⅳ. 딥러닝 알고리즘을 활용한 구조응답 패턴분석 20
• 1. 케이블지지교량 구조응답 패턴분석 20
• 2. 딥러닝 알고리즘 기반 케이블지지교량 구조응답 패턴분석 기법제안 21
• 3. 딥러닝 알고리즘 기반 케이블지지교량 구조응답 패턴분석 22
• 1) 케이블지지교량의 패턴분석 데이터 22
• (1) DNN MODEL 학습데이터 29
• (2) LSTM, Bi-LSTM MODEL 학습데이터 30
• (3) Deep Learning MODEL 학습데이터 전처리 32
• 2) Deep Learning Model 33
• (1) DNN Model 36
• (2) LSTM, Bi-LSTM Model 37
• Ⅴ. 딥러닝 알고리즘을 활용한 구조응답 패턴분석 결과 39
• 1. DNN Model 학습 및 예측결과 39
• 2. LSTM Model(SL7) 학습 및 예측결과 43
• 3. LSTM Model(SL14) 학습 및 예측결과 47
• 4. Bi-LSTM Mode(SL7) 학습 및 예측결과 51
• 5. Bi-LSTM Model(SL14) 학습 및 예측결과 55
• Ⅵ. 딥러닝 알고리즘을 활용한 구조응답 패턴분석 성능평가 59
• 1. DNN Model 성능평가 59
• 2. LSTM Model 성능평가 61
• 3. Bi-LSTM Model 성능평가 64
• Ⅶ. 고 찰 67
• Ⅷ. 결 론 69
• 참 고 문 헌 70
• Abstract 71
• 감 사 의 글 73