표제지
요약
목차
1. 서론 12
1.1. 연구 배경 및 필요성 12
1.2. 선행연구 동향 15
2. 이론적 고찰 18
2.1. 투수성 반응벽체 (PRB; Permeable Reactive Barriers) 18
2.1.1. 투수성 반응벽체 개요 18
2.1.2. 설계 및 적용 19
2.1.3. 반응 매체 24
2.2. 인공지능(AI;Artificial Intelligence) 30
2.2.1. 기계학습 30
2.2.2. 하이퍼파라미터 최적화 35
2.2.3. 기계학습 모델의 해석 36
3. 연구 내용 및 방법 37
3.1. 데이터 수집 38
3.1.1. 데이터 수집 기준 및 방법 38
3.1.2. 모델 입력변수 39
3.2. 기계학습 모델 개발 41
3.2.1. 데이터 전처리 41
3.2.2. 모델 최적화 44
3.3. 모델 성능평가 및 해석 49
3.3.1. 오차행렬(Confusion matrix) 49
3.3.2. SHAP Values 51
4. 연구 결과 및 고찰 52
4.1. 수집 데이터 분석 52
4.1.1. 데이터 구축 결과 52
4.1.2. 입력 변수간 상관관계 분석 결과 59
4.2. 기계학습을 이용한 모델 결과 60
4.2.1. 모델별 최적의 하이퍼파라미터 설정 60
4.2.2. 변수 기여도 기반 모델의 해석 결과 63
4.3. 모델 적용가능성 분석 67
4.3.1. 적용가능성 평가를 위한 부지데이터 구축 67
4.3.2. 모델 결과 해석을 통한 적용가능성 평가 69
5. 결론 71
References 73
Abstract 80
Table 1.1. Classification of remediation technologies 12
Table 2.1. Difference Characteristics of parameters and hyperparameters 35
Table 3.1. Type of Reactive media 39
Table 3.2. Independent Variable used in this thesis 40
Table 3.3. Model version used 44
Table 3.4. Hyperparameter range of Random Forest 45
Table 3.5. Hyperparameter range of XGB Model 46
Table 3.6. Hyperparameter range of SVM Model 47
Table 3.7. Hyperparameter range of ANN Model 48
Table 4.1. Site information of usinf for the model 53
Table 4.2. Random Forest Hyperparameter 60
Table 4.3. Evaluation metrics for Random Forest model 60
Table 4.4. XGB Hyperparameter 61
Table 4.5. Evaluation metrics for XGB model 61
Table 4.8. ANN Hyperparameter 62
Table 4.9. Evaluation metrics for ANN model 62
Table 4.10. Site information used to evaluate ANN Model for applicability 68
Table 4.11. Assessment applicability of reaction media in PRB according to site(Table 4.10) by ANN model 70
Figure 2.1. Diagarm of PRB 18
Figure 2.2. Standard design protocols for the PRBs, prior to installation at groundwater contaminated site 19
Figure 2.3. Schematic of the arrangement of a PRB structure in the ground 23
Figure 2.4. Factors influencing the reactive media selection and related effects 26
Figure 2.5. conceptual model for metals removal from water using NZVI (Permeable Reactive Barrier: Technology Update, 2011) 27
Figure 2.6. Maximum adsorption/sorption capacities of AC from reported studies used in PRBs for few contaminants based on the Langmuir model 29
Figure 2.7. Conceptual diagram of Artifitial intelligence 30
Figure 2.8. Random Forest Schematic Diagram 31
Figure 2.9. Ensemble algorithm Schematic Diagram 32
Figure 2.10. Support Vector Machine Schematic Diagram 33
Figure 2.11. ANN Schematic Diagram 34
Figure 3.1. Flow chart of this thesis 37
Figure 3.2. Data relabeling process 41
Figure 3.3. Method of data distribution verification 42
Figure 3.4. Data division schematic diagram 43
Figure 3.5. Conceptual diagram of Confusion Matrix 49
Figure 3.6. Schematic diagram of SHAP 51
Figure 4.1. Ca Box-plot 56
Figure 4.2. pH Box-plot 56
Figure 4.3. ORP Box-plot 56
Figure 4.4. hydraulic Conductivity Box-plot 56
Figure 4.5. Alkalinity Box-plot 56
Figure 4.6. DO Box-plot 56
Figure 4.7. Bar graph of reaction media 57
Figure 4.8. Bar graph of pollutant characteristics 58
Figure 4.9. Correlation heatmap of Independent variable 59
Figure 4.10. Plots of Random Forest Moddel SHAP dependence 64
Figure 4.11. Plots of XGB Moddel SHAP dependence 65
Figure 4.12. Plots of ANN Moddel SHAP dependence 66