GMDH 알고리즘을 이용한 수직벽과 경사식 호안에 대한 월파량 공식 유도

학위논문 (석사)-- 서울대학교 대학원 : 공과대학 건설환경공학부, 2018. 2. 황진환.Since wave overtopping is a complex phenomenon and it is sensitive to hydraulic and structural parameters, hydraulic model tests have been used to estimate the wave overtopping rates at coastal structures. Efforts have also been made toward deriving empirical formulas based on a large amount of accumulated data, e.g. the EurOtop formulas based on the CLASH database. To complement the EurOtop formulas, artificial neural network (ANN) models were presented, which is one of the data models. Other data models have also been developed. On the other hand, Goda developed unified formulas for vertical and inclined seawalls with smooth, impermeable surfaces. In this study, new data-based formulas are developed by using the group method of data handling (GMDH) algorithm with the CLASH and new EurOtop datasets. The GMDH formulas are shown to be more accurate than the Godas formulas and comparable in accuracy to other formulas developed separately for vertical and inclined seawalls. The estimation errors of 95% confidence interval and the range of 95% prediction error are also given. Sensitivity analysis of the derived formulas is also carried out.CHAPTER 1. INTRODUCTION 1 1.1 Background 1 1.2 Objectives 3 CHAPTER 2. THEORETICAL BACKGROUDS 5 2.1 Mean wave overtopping rate 5 2.2 Overtopping rate prediction formula 6 2.2.1 EurOtop formula (2007, 2016) 6 2.2.2 Formula by Goda (2009) 11 2.2.3 Formula by Etemad-Shahidi et al. (2014, 2016) 12 2.2.4 ANN model (Van Gent et al., 2007, Zanuttigh et al., 2016) 13 2.3 Group Method of Data Handling 14 2.3.1 Basic concept of GMDH algorithm 14 2.3.2 Procedure of GMDH algorithm 16 2.3.3 External criteria for self-organizing 17 2.4 Data sampling technique - bootstrap sampling 20 CHAPTER 3. METHODOLOGY 23 3.1 CLASH data for model construction 23 3.1.1 Variables applying Froude similitude law 23 3.1.2 Selected data for GMDH model 26 3.2 Training data extraction using bootstrap sampling 30 CHAPTER 4. RESULTS 34 4.1 Unified formulas for mean overtopping rate derived by GMDH algorithm 34 4.1.1 Unified formulas by GMDH algorithm 34 4.1.2 Confidence interval of estimation error 35 4.1.3 Performance of derived formulas 38 4.1.4 Comparison of unified GMDH formulas with other overtopping formulas 44 4.1.5 Sensitivity analysis for unified GMDH formulas 51 4.2 Separate formulas for overtopping rate derived by GMDH algorithm 61 4.2.1 GMDH-based overtopping rate prediction formulas for smooth-slope, normal-incidence wave 61 4.2.2 GMDH-based overtopping rate prediction formulas for inclined seawalls considering oblique wave condition and roughness factor 67 4.2.3 GMDH-based overtopping rate prediction formulas for vertical walls considering oblique wave condition 75 CHAPTER 5. CONCLUSIONS 80 REFERENCES 82 APPENDIX A Results of bootstrap sampling for separate formulas by GMDH 85 APPENDIX B Distribution of estimation error 89 APPENDIX C Performance graph of formulas 91 국문초록 93Maste