Dynamic Modeling of Complex Industrial Processes: Data-driven Methods and Application Research (eBook)

Supervisor’s Foreword 6
Preface 8
Part of this thesis has been published in the following journal articles: 9
Acknowledgements 10
Contents 11
Acronyms 14
1 Introduction 16
1.1 Background 16
1.2 Literature Review on Multivariate Statistical Process Monitoring 17
1.2.1 Principal Component Analysis 18
1.2.2 Current Research Status of Process Monitoring 20
1.3 Literature Review on Data-driven Soft Sensor Modeling 22
1.3.1 Partial Least Squares 23
1.3.2 Current Research Status of Soft Sensing 25
1.4 Challenges and Opportunities 27
1.5 Contents and Outline 29
1.6 Conclusions 31
References 31
2 Monitoring of Operating Condition and Process Dynamics with Slow Feature Analysis 35
2.1 Introduction 35
2.2 Slow Feature Analysis---Revisit 37
2.2.1 Definition of Slowness and Some Notations 37
2.2.2 Optimization Problem 38
2.2.3 Solution Algorithm 39
2.2.4 Geometric and Statistical Properties of SFA 41
2.2.5 Comparison with Classic Latent Variable Models 42
2.3 Process Monitoring with SFA 43
2.3.1 Dimension Reduction of SFA 43
2.3.2 Monitoring Statistics Design with Slow Features 45
2.4 Case Studies 46
2.4.1 Simulated CSTR 46
2.4.2 TE Benchmark Process 52
2.5 Conclusions 59
References 61
3 Control Performance Monitoring and Diagnosis Based on SFA and Contribution Plot 63
3.1 Introduction 63
3.2 Control Performance Monitoring with SFA 64
3.3 Control Performance Diagnosis with Contribution Analysis 65
3.3.1 Revisiting Contribution Analysis 65
3.3.2 Control Performance Diagnosis 66
3.3.3 Comparison with the Covariance-Based Approach 67
3.4 Case Studies 70
3.4.1 A Simple Simulated Multivariate Process 70
3.4.2 TE Benchmark Process 72
3.5 Conclusions 77
References 77
4 Recursive SFA Algorithm and Adaptive Monitoring System Design 1
4.1 Introduction 79
4.2 Updating Law for Covariance Matrices 80
4.3 The Recursive SFA Algorithm 83
4.3.1 Improved Monitoring Statistics 83
4.3.2 Rank-One Modification for the First Decomposition 83
4.3.3 OIP Algorithm for the Second Decomposition 84
4.3.4 Complexities of the RSFA Algorithm 85
4.4 Adaptive Monitoring System Design 86
4.4.1 Control Limits Updating 86
4.4.2 An Improved Stopping Criterion for Model Updating 87
4.5 Case Studies 88
4.5.1 Simulated CSTR 88
4.5.2 An Industrial Heating Furnace System 90
4.6 Conclusions 95
References 95
5 Probabilistic Slow Feature Regression for Dynamic Soft Sensing 96
5.1 Introduction 96
5.2 Probabilistic SFA 98
5.2.1 Mathematical Definition 99
5.2.2 Parameter Estimation Using the EM Algorithm 101
5.3 Soft Sensor Modeling Based on Probabilistic Slow Feature Regression 106
5.3.1 Probabilistic Slow Feature Regression 106
5.3.2 Online Implementation 107
5.4 Case Studies 109
5.4.1 A Hybrid Tank System 109
5.4.2 SRU Process 116
5.5 Conclusions 119
References 119
6 Enhanced Dynamic PLS with Temporal Smoothness for Soft Sensing 121
6.1 Introduction 121
6.2 Dynamic PLS with Temporal Smoothness Regularization 124
6.2.1 Problem Description 124
6.2.2 Optimization Problem 126
6.2.3 Implementation Procedure in Soft Sensing 127
6.3 Case Studies 127
6.3.1 Numerical Simulations 127
6.3.2 TE Benchmark Process 132
6.4 Conclusions 133
References 134
7 Nonlinear Dynamic Soft Sensing Based on Bayesian Inference 136
7.1 Introduction 136
7.2 Support Vector Machine and Its Probabilistic Interpretation 137
7.2.1 Problem Description 137
7.2.2 Bayesian Interpretation 138
7.3 Nonlinear Dynamic Soft Sensor Model Based on FIR and SVM 139
7.4 Parameter Optimization Based on Bayesian Inference 140
7.4.1 Optimizing Model Parameters of SVM 141
7.4.2 Optimizing Regularization Parameters 141
7.4.3 Optimizing Kernel Parameter 143
7.4.4 Optimizing Dynamic Parameters in FIR 144
7.5 Case Studies 145
7.5.1 A Simulated Example 145
7.5.2 Online Prediction of Propylene Melt Index 148
7.6 Conclusions 150
References 150
8 Conclusions and Recommendations 152
8.1 Concluding Remarks 152
8.2 Recommendations for Future Work 153