Table of Contents
Preface xi
Acknowledgments xv
About the Authors xvii
1 Introduction 1
1.1 Learning Issues in Feedback Control 1
1.1.1 Adaptive and Learning Control 1
1.1.2 Intelligent Control and Neural Network Control 4
1.2 Learning Issues in Temporal Pattern Recognition 6
1.2.1 Pattern Recognition in Feedback Control 6
1.2.2 Representation, Similarity, and Rapid Recognition 7
1.3 Preview of the Main Topics 9
1.3.1 RBF Networks and the PE Condition 9
1.3.2 The Deterministic Learning Mechanism 10
1.3.3 Learning from Adaptive Neural Network Control 11
1.3.4 Dynamical Pattern Recognition 12
1.3.5 Pattern-Based Learning Control 13
1.3.6 Deterministic Learning Using Output Measurements 14
1.3.7 Nature of Deterministic Learning 15
2 RBF Network Approximation and Persistence of Excitation 17
2.1 RBF Approximation and RBF Networks 18
2.1.1 RBF Approximation 18
2.1.2 RBF Networks 20
2.2 Persistence of Excitation and Exponential Stability 23
2.3 PE Property for RBF Networks 27
3 The Deterministic Learning Mechanism 37
3.1 Problem Formulation 38
3.2 Locally Accurate Identification of Systems Dynamics 39
3.2.1 Identification with σ-Modification 40
3.2.2 Identification without Robustification 44
3.3 Comparison with System Identification 46
3.4 Numerical Experiments 49
3.5 Summary 58
4 Deterministic Learning from Closed-Loop Control 61
4.1 Introduction 61
4.2 Learning from Adaptive NN Control 62
4.2.1 Problem Formulation 62
4.2.2 Learning from Closed-Loop Control 63
4.2.3 Simulation Studies 70
4.3 Learning from Direct Adaptive NN Control of Strict-Feedback Systems 75
4.3.1 Problem Formulation 76
4.3.2 Direct ANC Design 77
4.3.3 Learning from Direct ANC 79
4.4 Learning from Direct ANC of Nonlinear Systems in Brunovsky Form 82
4.4.1 Stability of a Class of Linear Time-Varying Systems 83
4.4.2 Learning from Direct ANC 86
4.4.3 Simulation Studies 92
4.5 Summary 95
5 Dynamical Pattern Recognition 97
5.1 Introduction 97
5.2 Time-Invariant Representation 99
5.2.1 Static Representation 99
5.2.2 Dynamic Representation 100
5.2.3 Simulations 101
5.3 A Fundamental Similarity Measure 104
5.4 Rapid Recognition of Dynamical Patterns 107
5.4.1 Problem Formulation 108
5.4.2 Rapid Recognition via Synchronization 109
5.4.3 Simulations 112
5.5 Dynamical Pattern Classification 117
5.5.1 Nearest-Neighbor Decision 117
5.5.2 Qualitative Analysis of Dynamical Patterns 118
5.5.3 A Hierarchical Structure 119
5.6 Summary 121
6 Pattern-Based Intelligent Control 123
6.1 Introduction 123
6.2 Pattern-Based Control 124
6.2.1 Definitions and Problem Formulation 124
6.2.2 Control Based on Reference Dynamical Patterns 126
6.2.3 Control Based on Closed-Loop Dynamical Patterns 127
6.3 Learning Control Using Experiences 128
6.3.1 Problem Formulation 128
6.3.2 Neural Network Learning Control 129
6.3.3 Improved Control Performance 132
6.4 Simulation Studies 133
6.5 Summary 137
7 Deterministic Learning with Output Measurements 139
7.1 Introduction 139
7.2 Learning from State Observation 141
7.3 Non-High-Gain Observer Design 146
7.4 Rapid Recognition of Single-Variable Dynamical Patterns 149
7.4.1 Representation Using Estimated States 149
7.4.2 Similarity Definition 151
7.4.3 Rapid Recognition via Non-High-Gain State Observation 152
7.5 Simulation Studies 156
7.6 Summary 165
8 Toward Human-Like Learning and Control 167
8.1 Knowledge Acquisition 167
8.2 Representation and Similarity 169
8.3 Knowledge Utilization 169
8.4 Toward Human-Like Learning and Control 170
8.5 Cognition and Computation 171
8.6 Comparison with Statistical Learning 172
8.7 Applications of the Deterministic Learning Theory 172
References 175
Index 189