Title Page
Abstract
Contents
Chapter 1. Introduction 14
1.1. Background of Research 15
1.2. Objectives and Organization of the Dissertation 20
1.2.1. Research Objectives 20
1.2.2. Organization of the Dissertation 22
Chapter 2. Development of Revised Multi-layer Perceptron Model 23
2.1. Introduction 24
2.1.1. Structure of MLP Model 25
2.1.2. Learning Algorithms 27
2.2. Revised MLP Model 31
2.2.1. Objectives and Methodologies 31
2.2.2. Revised MLP Learning Procedure 33
2.3. Application of RMLP Model 37
2.3.1. Estimation of Inflow for Dam Operation 38
2.3.2. Prediction of Water Level in Urban Drainage Basin 47
2.4. Summary of Improvement of Learning Algorithm 67
Chapter 3. Data Pre-processing in Consideration of Basin Characteristics 69
3.1. Introduction 70
3.2. Data Cleaning for Deep Learning 72
3.2.1. Interpolation of Missing Data 72
3.2.2. Removal of Noise and Outliers 74
3.3. Data Pre-processing in Consideration of Hydrological Characteristics of Basin 78
3.3.1. Normalization of Data 78
3.3.2. Seasonal Division Based on Water Level 79
3.3.3. Time Offset of Data Considering Travel Time 81
3.4. Pre-processing Application Results 82
3.4.1. Improved Dam Inflow Prediction Accuracy 82
3.4.2. Improvement of Water Level Prediction Accuracy for Urban Basin 88
3.4.3. Data Pre-processing for RMLP Model 91
3.5. Summary of Results 94
3.6. Effects of Input Data Variation 96
3.7. Comparison with Prediction Results of SWMM 99
3.7.1. SWMM Calibration Results (SIT, 2020) 99
3.7.2. Error Comparison with RMLP Model 101
Chapter 4. Structure and Parameter Optimization for RMLP Models 106
4.1. Introduction 107
4.2. Methodology for Model and Parameter Optimization 108
4.2.1. Harmony Search (HS) 108
4.2.2. Parameter-Setting-Free Harmony Search (PSF-HS) 110
4.2.3. Parameter Adaptive Harmony Search (PAHS) 114
4.3. Results for Optimal Learning Models 115
4.3.1. Deriving Optimal Learning Model 115
4.3.2. Optimal Model Application Results (PSF-HS) 117
4.3.3. Structural Optimization Results for PAHS 134
Chapter 5. Summary and Discussions 139
5.1. Summary and Conclusions 140
5.2. Limitations and Future Research 141
References 144
Table 2.1. Model composition 35
Table 2.2. Data analysis for Soyanggang Dam catchment 40
Table 2.3. Sensitivity analysis of MLP model parameters for Soyanggang Dam catchment 41
Table 2.4. Sensitivity analysis of RMLP model parameters for Soyanggang Dam catchment 42
Table 2.5. Model parameter determination for Soyanggang Dam catchment 42
Table 2.6. Results of RMLP model for Soyanggang Dam catchment 43
Table 2.7. Extraction of dates with 1 inch of rainfall in Sinwol basin 55
Table 2.8. Model input and output data for Sinwol basin 56
Table 2.9. Sensitivity analysis of MLP model parameters for Sinwol basin 58
Table 2.10. Sensitivity analysis of RMLP model parameters for Sinwol basin 59
Table 2.11. Model parameter determination for Sinwol basin 60
Table 2.12. Results of RMLP model for Soyanggang Dam catchment 60
Table 3.1. Seasonal division data analysis for Soyanggang Dam catchment 84
Table 3.2. Data pre-processing cases for Soyanggang Dam catchment 85
Table 3.3. Results of pre-processing for Soyanggang Dam catchment 87
Table 3.4. Lead times and correlation coefficients for Sinwol basin 88
Table 3.5. Time offset for 5-min water level prediction 89
Table 3.6. Data pre-processing cases for Sinwol basin 89
Table 3.7. Results of pre-processing for Sinwol basin 90
Table 3.8. Pre-processing results of MLP and RMLP models for Soyanggang Dam catchment 91
Table 3.9. Pre-processing results of MLP and RMLP models for Sinwol basin 93
Table 3.10. Additional data pre-processing cases for Sinwol basin 96
Table 3.11. Results of additional pre-processing for Sinwol basin 97
Table 4.1. Boundary conditions for determining variables 117
Table 4.2. Final HM for Soyanggang Dam catchment (peak season) 119
Table 4.3. Final HM for Soyanggang Dam catchment (off season) 119
Table 4.4. Results of increasing prediction accuracy for Soyanggang Dam catchment 120
Table 4.5. Final HM for Sinwol basin (high inlet) 125
Table 4.6. Final HM for Sinwol basin (low inlet 1) 126
Table 4.7. Final HM for Sinwol basin (low inlet 2) 126
Table 4.8. Results of increasing prediction accuracy for Soyanggang Dam catchment 128
Table 4.9. PAHS Final HM for Sinwol basin (high inlet) 136
Table 4.10. PAHS Final HM for Sinwol basin (low inlet 1) 137
Table 4.11. PAHS Final HM for Sinwol basin (low inlet 2) 137
Figure 1.1. Sequence of this dissertation 22
Figure 2.1. Structure of MLP model 27
Figure 2.2. Activation functions 29
Figure 2.3. Gradient decent method 30
Figure 2.4. Probabilities of BR and PR 33
Figure 2.5. Learning process of model 34
Figure 2.6. Pseudo code for extracting model weights 36
Figure 2.7. Soyanggang Dam catchment 38
Figure 2.8. Dataset of the Soyanggang Dam catchment 39
Figure 2.9. Comparison of model losses during training for the Soyanggang Dam catchment 45
Figure 2.10. Comparison of inflow prediction results of revised model for the Soyanggang Dam catchment 46
Figure 2.11. Flood area and tunnel location in Yangcheon-gu 47
Figure 2.12. Design of the Sinwol deep underground rainwater tunnel 49
Figure 2.13. Field survey of inlet gates 50
Figure 2.14. Weir opening process 51
Figure 2.15. Locations of sewer pipes and water level gauges 52
Figure 2.16. Locations of KMA rain gauges for the Sinwol basin 53
Figure 2.17. Locations of tunnel inlets and subareas for the Sinwol basin 57
Figure 2.18. Comparison of model losses during training for the Sinwol basin 62
Figure 2.19. Water depth prediction results for the Sinwol basin 64
Figure 3.1. Procedure and classification of data pre-processing 71
Figure 3.2. Interpolation process for missing data 73
Figure 3.3. Water gauge field survey in the Sinwol basin 75
Figure 3.4. Vegetation growth in rivers 76
Figure 3.5. Noise removal for water gauges in the Sinwol basin 77
Figure 3.6. Data normalization of scaled input data 78
Figure 3.7. Seasonal division process 80
Figure 3.8. TLCC analysis of time-series data 82
Figure 3.9. TLCC analysis for the Soyanggang Dam catchment 83
Figure 3.10. Results of pre-processing to raw data 85
Figure 3.11. Data separation based on type of pre-processing 86
Figure 3.12. Basic SWMM(left) and detailed model(right) for the Sinwol basin 100
Figure 3.13. SWMM calibration results for the Sinwol basin 101
Figure 3.14. Comparison of revised MLP and SWMM calibration results (2018.8.28.) 102
Figure 3.15. Comparison of revised MLP and SWMM calibration results (2020.7.19.) 103
Figure 3.16. Comparison of revised MLP and SWMM calibration results (2020.8.10.) 103
Figure 4.1. Harmony search algorithm flowchart 109
Figure 4.2. Model structure and parameter optimization flowchart 115
Figure 4.3. PSF-HS results for the Soyanggang Dam catchment (peak season) 118
Figure 4.4. PSF-HS results for the Soyanggang Dam catchment (off season) 118
Figure 4.5. Histograms of prediction error for the Soyanggang Dam catchment 121
Figure 4.6. Comparison of inflow prediction results of PSF-HS for the Soyanggang Dam catchment 122
Figure 4.7. PSF-HS results for the Sinwol basin (high inlet) 124
Figure 4.8. PSF-HS results for the Sinwol basin (low inlet 1) 124
Figure 4.9. PSF-HS results for the Sinwol basin (low inlet 2) 125
Figure 4.10. Histograms of prediction error for the Sinwol basin (high inlet) 129
Figure 4.11. Histograms of prediction error for the Sinwol basin (low inlet 1) 129
Figure 4.12. Histograms of prediction error for the Sinwol basin (low inlet 2) 130
Figure 4.13. Comparison of inflow prediction results of PSF-HS for the Sinwol basin 131
Figure 4.14. PAHS results for the Sinwol basin (high inlet) 135
Figure 4.15. PAHS results for the Sinwol basin (low inlet 1) 135
Figure 4.16. PAHS results for the Sinwol basin (low inlet 2) 136