Title Page
Contents
LIST OF ABBREVIATIONS AND SYMBOLS 15
ABSTRACT 19
CHAPTER ONE. INTRODUCTION 21
1.1. Suitability of MEBN for PSAW 23
1.2. Problem Statement 30
1.3. Thesis Statement and Scope 37
1.4. Contributions 38
1.5. Organization of the Dissertation 39
CHAPTER TWO. BACKGROUND 40
2.1. From Data Fusion to Situation Awareness 40
2.1.1. Data Fusion 40
2.1.2. Situation Awareness 43
2.2. Bayesian Network 46
2.3. Multi-Entity Bayesian Network 47
2.3.1. Definition of MEBN 47
2.3.2. A Script for MEBN 55
2.4. Data for MEBN Learning from RM 56
2.4.1. Relational Model 57
2.5. Uncertainty Modeling Process for Semantic Technology (UMP-ST) 63
2.6. Conclusion 65
CHAPTER THREE. MEBN-RM MAPPING MODEL 66
3.1. MEBN-RM 67
3.2. Entity Mapping 72
3.3. Resident Node Mapping 72
3.3.1. Predicate 73
3.3.2. Function 75
3.4. Relation Schema and MFrag Mapping 76
3.5. Relational Database Schema and MTheory Mapping 78
3.6. MEBN-RM Mapping Algorithm 79
3.7. Conclusion 82
CHAPTER FOUR. PSAW-MEBN REFERENCE MODEL 84
4.1. Predictive Situation Awareness 86
4.1.1. Definition of PSAW 86
4.1.2. Properties of a PSAW Model 91
4.1.3. Physical and Mental Situations in PSAW 96
4.1.4. Properties of OODA 98
4.1.5. PSAW in terms of OODA 102
4.2. PSAW-MEBN Reference Model 105
4.2.1. Entities for the Reference Model 109
4.2.2. Random Variables for the Reference Model 110
4.2.3. The Five MFrag Groups 115
4.3. Conclusion 119
CHAPTER FIVE. HUMAN-AIDED MEBN LEARNING FOR PSAW 120
5.1. Introduction 121
5.1.1. Illustrative Running Example of Relational Data for MEBN Learning 121
5.1.2. Elements of MEBN Learning from Relational Data 123
5.2. A Process for Human-Aided MEBN Learning for PSAW 124
5.2.1. Analyze Requirements 125
5.2.2. Define World Model 128
5.2.3. Construct Reasoning Model 136
5.2.4. Test Reasoning Model 164
5.3. Conclusion 165
CHAPTER SIX. APPLYING HMLP-USE CASES AND EXPERIMENT 171
6.1. Use Case 1 - HERALD 171
6.1.1. Critical Infrastructure Defense Situation 172
6.1.2. HERALD System 175
6.1.3. Discussion of HMLP Performance 178
6.2. Use Case 2 - PROGNOS 181
6.2.1. Introduction 182
6.2.2. PROGNOS PO via HMLP 183
6.2.3. Comparing UMP-ST and HMLP 190
6.3. Use Case 3 - Smart Manufacturing 192
6.3.1. Introduction 193
6.3.2. MSAW-MEBN Model for Predictive Situation Awareness 195
6.3.3. Discussion 199
6.4. Experimental Comparison between UMP-ST and HMLP 199
6.4.1. Preparation Process 201
6.4.2. Execution Process 202
6.4.3. Evaluation Process 206
6.4.4. Comparison Results 208
CHAPTER SEVEN. CONCLUSION AND FUTURE WORK 213
7.1. MEBN-RM Mapping Model 213
7.2. The PSAW-MEBN Reference Model 214
7.3. Human-aided MEBN Learning for PSAW 214
7.4. Future Research Questions 215
7.4.1. Aggregating Influence Problem 215
7.4.2. Dynamic Model Learning 216
7.4.3. Continuous Random Variable Learning 216
7.4.4. Learning from Incomplete Data 217
7.4.5. Learning for Hidden Variables 218
7.4.6. Incremental MEBN Learning 219
7.6. Conclusion 219
APPENDIX 221
Appendix A. Bayesian Network Learning 223
A.1. BN Parameter Learning 223
Appendix B. PSAW Questions 231
Appendix C. Metrics for Prediction Accuracy 233
Appendix D. Local Probability Description Language (LPDL) 235
Appendix E. Conditional Probability Script Language (CPSL) 237
Appendix F. An HMLP Tool 244
Appendix G. Use Case 1: Human-Aided MEBN Learning for HERALD and Test Results 246
G.1. Analyze Requirements 246
G.2. Define World Model 248
G.3. Construct Reasoning Model 258
G.4. Test Reasoning Model 261
Appendix H. Use Case 1: HERALD Scenario Simulator 276
H.1. Environmental Factors, Spatial objects, Way Points, and Time 276
H.2. Simulation Entities 278
H.3. Sensor Systems in the Simulation 280
H.4. Ground Truth Dataset 283
Appendix I. Use Case 1: HERALD MTheory 286
I.1. SensorOf MFrag 288
I.2. Predecessor MFrag 289
I.3. SensorTemporalProperty MFrag 289
I.4. ReportedTarget_MTIRPT MFrag 289
I.5. ReportedTarget_IMINTSRPT MFrag 290
I.6. ReportedTarget_GEOINTRPT MFrag 290
I.7. GEOINTS_Report MFrag 290
I.8. MTI_Report MFrag 291
I.9. IMINTS_Report MFrag 291
I.10. Target MFrag 292
I.11. TargetTemporalProperty MFrag 292
I.12. Situation MFrag 293
Appendix J. Use Case 1: Confusion Matrix for Categorical Variables of HERALD MTheory 294
J.1. SSBN_g1_t2 294
J.2. SSBN_g1_t3 295
J.3. SSBN_g1_t4 295
Appendix K. Use Case 1: SSBN_g1_t4 297
Appendix L. Use Case 2: PROGNOS PO via UMP-ST 321
L.1. Requirements 321
L.2. Analysis & Design 322
L.3. Implementation 324
L.4. Test 328
Appendix M. Use Case 3: Predictive Situation Awareness Model for Smart Manufacturing 329
M.1. Smart Manufacturing 329
M.2. Representation for Uncertainty in Manufacturing 332
M.3. Use Case 339
M.4. Conclusion 345
Appendix N. Experimental Comparison of HMLP with UMP-ST 346
N.1. Slides for the Lectures 346
N.2. Data given to Participants 354
N.3. MEBN Model Results from Participants 355
Appendix O. Hybrid Message Passing with Gaussian Mixture Reduction with Optimal Settings 361
O.1. Introduction 362
O.2. Preliminaries 368
O.3. Extended Hybrid Message Passing Algorithm 378
O.4. Optimizing the Settings of HMP-GMR 383
O.5. Experiment 388
O.6. Conclusion 406
Appendix P. Test HBN for HMP-GMR 407
CLG BN 1 (n=5) 407
CLG BN 2 (n=5) 407
CLG BN 3 (n=5) 408
CLG BN 4 (n=5) 409
CNG BN 1 (n=5) 410
CNG BN 2 (n=5) 410
CNG BN 3 (n=5) 411
CNG BN 4 (n=5) 412
REFERENCES 414
BIOGRAPHY 434
Table 3.1. Resident node types on MEBN-RM 73
Table 3.2. Predicate mapping in MEBN-RM 74
Table 3.3. Function mapping in MEBN-RM 76
Table 3.4. Mapping types on MEBN-RM 82
Table 4.1. Possible inputs/outputs for the four steps in OODA 101
Table 5.1. Parts of the Situation Identification Relational Database 123
Table 5.2. Normalized relational dataset from Table 5.1 138
Table 5.3. Joined dataset 143
Table 5.4. CSD dataset 154
Table 5.5. Communicate Relation and Meet Relation 161
Table 5.6. Converted Relations from Communicate Relation and Meet Relation 162
Table 5.7. Processing Method for Steps in HMLP 169
Table 6.1. Comparison between UMP-ST and HMLP in terms of the development periods 181
Table 6.2. Comparison between the original PROGNOS probabilistic ontology... 191
Table 6.3. Preparation Process 202
Table 6.4. Execution Process 206
Table 6.5. Evaluation Process 208
Table 6.6. Comparison results for total development times and average accuracies 209
Table 6.7. Comparison results for time-consuming tasks 210
Table 7.1. Dataset A represents the BN data case. Dataset B represents the MEBN data case 218
Figure 1.1. Danger Assessment Problem 23
Figure 1.2. Bayesian Network of the Danger Assessment Problem 24
Figure 1.3. Danger Assessment MTheory 27
Figure 1.4. An example SSBN with instance local distributions derived from the Vehicle Object MFrag given... 28
Figure 1.5. Generating Several SSBNs 29
Figure 1.6. Concept of Research 31
Figure 2.1. Danger MFrag 48
Figure 2.2. SSBN from Danger MFrag (given v1, v2, and v3 as vehicle, and region1_1 as region) 50
Figure 2.3. Hybrid SSBN from Danger MFrag 52
Figure 2.4. Example of a Vehicle Identification RDB 58
Figure 4.1. An illustrative example of situation awareness 87
Figure 4.2. OODA Loop 99
Figure 4.3. What an observer sees 101
Figure 4.4. A situation of an observer in terms of Interpreted/Actual OODA 103
Figure 4.5. Illustrative example of a PSAW model 105
Figure 4.6. An example of an SSBN generated by a PSAW-MTheory 107
Figure 4.7. Four core kinds of RVs and examples of each 114
Figure 4.8. Illustrative example of PSAW-MFrags 116
Figure 5.1. Schema of a Situation Identification Relational Database 121
Figure 5.2. Learning Subjects for MEBN Learning 124
Figure 5.3. Process for Human-Aided MEBN Learning 125
Figure 5.4. Analyze Requirements 126
Figure 5.5. Define World Model 130
Figure 5.6. Construct Reasoning Model 136
Figure 5.7. Test Reasoning Model 164
Figure 6.1. Ranges of a target from a critical infrastructure 174
Figure 6.2. HERALD System in a C4I system 177
Figure 6.3. Comparison between UMP-ST and HMLP 179
Figure 6.4. Part of EER Model for a PROGNOS world model 186
Figure 6.5. Extended PROGNOS probabilistic ontology 188
Figure 6.6. Extended MSAW-MEBN Model derived using the PSAW-MEBN reference model (Chapter 4) 196
Figure 6.7. Situation for the simple heating machinery 203
Figure 6.8. Each of training and test data has sensed data and actual data for the simple heating machinery 205
Figure 6.9. Ideal causal relationships between random variables for the simple heating machinery 210