Title Page

Contents

Abstract 15

1. Introduction 18

1.1. Background 18

1.2. Literature Review 19

1.3. Problem Statement and Research Objectives 24

1.4. Thesis Contribution 25

2. Information Gap Decision Theory: Uncertainty Modeling 27

2.1. Introduction 27

2.2. Uncertainty Models of IGDT 27

2.3. Functions of IGDT 29

2.4. Classification of RES and Loads 33

3. Sensitivity Analysis Techniques: Uncertainty Quantification 37

3.1. Local SA 37

3.2. Screening SA 38

3.3. Global SA (GSA) 40

3.3.1. Multi-Attribute Decision Making (MADM) 41

3.3.2. Pearson Correlation Coefcient 42

3.3.3. Borgonovo Method (BM) 43

3.3.4. Output Variance-based GSA 44

3.3.5. Strategy Sequence of GSA 51

3.4. Sparse Polynomial Chaos Expansion (SPCE) 53

3.4.1.1. A posteriori error estimation 54

3.4.1.2. Sparse PCE: Least Angle Regression 55

3.4.1.3. Bootstrap SPCE 56

3.4.2. Rosenblatt Based GSA 56

4. Artificial Intelligence techniques: Application Choice 60

4.1. Application of RNN, LSTM, and CNN 65

4.2. Online Load Monitoring and Classification 69

5. The Proposed Research Methodology 74

5.1. Skeleton of the Proposed Research Methodology 74

5.1.1. IGDT based Uncertainty classes 75

5.1.2. IGDT based Uncertainty classes 76

5.1.3. Artificial Neural Network Model: LSTM 77

5.1.4. Decision matrices: Data base for SDM 78

5.1.5. IGDT based Sympathetic function and decision model 80

6. Sympathetic Decision Model Applications: Simulation Results 82

6.1. IEEE 30 bus system: Test System Description 82

6.2. Steps to Model SDM and the Results 83

6.2.1.1. Input variables sampling, distribution, and correlation 84

6.2.1.2. Rosenblatt Transform and SPCE 86

6.2.1.3. RMV-GSA Model and FSIs Values 88

6.2.2. Uncertainty Modeling and Classification: Application of IGDT 93

6.2.3. Decision Sets Formation of SDM 93

6.2.4. Deployment of SDM on Power System (IEEE 30 bus system) 95

7. Unit Commitment and Economic Dispatch: Application of Sympathetic Decision Model using Reinforcement Learning 114

7.1. Unit commitment and economic dispatch 115

7.1.1. Modeling of unit commitment and economic dispatch 117

7.2. Reinforcement learning and agent selection 118

7.2.1. Reinforcement Learning Agents 119

7.3. RL based modeling of unit commitment and economic dispatch 121

7.3.1. RL agent modeling 124

7.4. Deployment of SDM-RL agent and simulation results 129

7.4.1. IEEE 30 bus test system 129

7.4.2. IEEE 123 bus test system 137

8. Conclusion and Future Work 142

8.1. Conclusion 142

8.2. Future Work 143

References 144

논문 요약 151

Table 2-1. Classification of RES and load demands RES or Load 35

Table 2-2. Comparison of proposed uncertainty modeling with previous literature 36

Table 5-1. General formation of the decision matrix 79

Table 6-1. Pre-requisite methods of RMV-GSA 84

Table 6-2. SPCE design parameters 89

Table 6-3. RMV-GSA based FSI values at load variations 90

Table 6-4. RMV-GSA based FSI values at RES variations 91

Table 6-5. Uncertainty classes in numerical values 93

Table 6-6. Decision Set for PV 94

Table 6-7. Decision Set for WT 94

Table 6-8. Decision Set for Loads 94

Table 7-1. Thermal units data IEEE 30 bus test system 129

Table 7-2. Thermal units data IEEE 123 bus test system 137

Figure 2-1. Sympathetic immunities of IGDT 30

Figure 2-2. Sympathetic impact of IGDT: α values 33

Figure 4-1. General applications of AI 61

Figure 4-2. Conventional method for loading mrgin 62

Figure 4-3. Gram-Chmidt technique for inputs reduction 64

Figure 4-4. MLP of ANN 65

Figure 4-5. LSTM basic structure 66

Figure 4-6. CNN basic structure 68

Figure 4-7. ANN based power system load monitoring and classification 70

Figure 4-8. BT-based ELM model of PI estimation for VSM 72

Figure 5-1. IGDT based Uncertainty Classification 75

Figure 5-2. RMV-GSA algorithm structure 76

Figure 5-3. LSTM structure for classification 78

Figure 5-4. Final proposed decision framework 80

Figure 6-1. IEEE 30 bus test system 84

Figure 6-2. Sampling methods response 85

Figure 6-3. Sampling data distribution and correlation (C-Vine Copula) 86

Figure 6-4. Sampling data before and after Rosenblatt transform (RT) 87

Figure 6-5. SPCE effective mean coefficients 88

Figure 6-6. Estimation errors 88

Figure 6-7. Voltage at the buses: Base case 96

Figure 6-8. Net active power at the buses: Base case 97

Figure 6-9. Net active power at the buses after SDM application 98

Figure 6-10. Net active power at the buses after modified SDM application 100

Figure 6-11. Uncertainty Control factor 101

Figure 6-12. ANN Outputs comparison: (a) regression model, (b) proposed classification model 102

Figure 6-13. LSTM classification model training and validation 103

Figure 6-14. LSTM classification model performance evaluation factors 104

Figure 6-15. LSTM classification model confusion matrix 104

Figure 6-16. Comparison of results of LSTM-SDM and SDM 105

Figure 6-17. Proposed model application to IEEE 123 bus test system 107

Figure 6-18. Loads active power (at forecasted data) 108

Figure 6-19. Active power at RES buses (at forecasted data) 109

Figure 6-20. Loads active power at RES buses and respective change after SDM application 109

Figure 6-21. Active power at RES buses and respective change after SDM application 109

Figure 6-22. Predicted classes by LSTM trained network for IEEE 123 bus test system 111

Figure 6-22,6-23. Testing result of LSTM trained network for IEEE 123 bus test system 112

Figure 7-1. RL agent model 126

Figure 7-2. RL agent Simulink environment 127

Figure 7-3. Voltage at each bus: base case 130

Figure 7-4. Active power dispatch thermal units: base case 131

Figure 7-5. RES active power penetration: base case 131

Figure 7-6. BES charging and discharging: base case 132

Figure 7-7. Unit commitment status: base case 132

Figure 7-8. Voltage at each bus: SDM-RL case 134

Figure 7-9. Active power dispatch thermal units: SDM-RL case 135

Figure 7-10. RES active power penetration: SDM-RL case 135

Figure 7-11. Unit commitment status: SDM-RL case 136

Figure 7-12. BES charging and discharging: SDM-RL case 136

Figure 7-13. Unit commitment status 3 DG: base case 138

Figure 7-14. Active power contribution of DGs: base case 138

Figure 7-15. RES active power and voltage status: base case 139

Figure 7-16. Loads active power: base case 139

Figure 7-17. Unit commitment status 3 DG: proposed method 140

Figure 7-18. Active power contribution of DGs: proposed method 140

Figure 7-19. RES active power and voltage status: proposed method 141

Figure 7-20. Loads active power: proposed method 141