Title Details

Frédéric Magoulès is Professor at the Ecole Centrale Paris in France and Honorary Professor at the University of Pècs in Hungary. His research focuses on parallel computing, numerical linear algebra and machine learning.

Hai-Xiang Zhao is Senior Researcher at Amadeus in France. His research focuses on parallel computing, data mining and machine learning.

Preface ix

Introduction  xi

Chapter 1. Overview of Building Energy Analysis 1

1.1. Introduction 1

1.2. Physical models 3

1.3. Gray models 6

1.4. Statistical models 6

1.5. Artificial intelligence models 8

1.5.1. Neural networks  8

1.5.2. Support vector machines 13

1.6. Comparison of existing models  14

1.7. Concluding remarks . 16

Chapter 2. Data Acquisition for Building Energy Analysis 17

2.1. Introduction  17

2.2. Surveys or questionnaires 18

2.3. Measurements 21

2.4. Simulation 25

2.4.1. Simulation software 26

2.4.2. Simulation process  28

2.5. Data uncertainty  34

2.6. Calibration 35

2.7. Concluding remarks  37

Chapter 3. Artificial Intelligence Models 39

3.1. Introduction  39

3.2. Artificial neural networks 40

3.2.1. Single-layer perceptron 41

3.2.2. Feed forward neural network 43

3.2.3. Radial basis functions network 44

3.2.4. Recurrent neural network 47

3.2.5. Recursive deterministic perceptron 49

3.2.6. Applications of neural networks 51

3.3. Support vector machines 53

3.3.1. Support vector classification 54

3.3.2. ε-support vector regression 59

3.3.3. One-class support vector machines 62

3.3.4. Multiclass support vector machines 63

3.3.5. v-support vector machines 64

3.3.6. Transductive support vector machines 65

3.3.7. Quadratic problem solvers . 67

3.3.8. Applications of support vector machines 75

3.4. Concluding remarks  76

Chapter 4. Artificial Intelligence for Building Energy Analysis 79

4.1. Introduction  79

4.2. Support vector machines for building energy prediction  80

4.2.1. Energy prediction definition 80

4.2.2. Practical issues 81

4.2.3. Support vector machines for prediction 85

4.3. Neural networks for fault detection and diagnosis 91

4.3.1. Description of faults  94

4.3.2. RDP in fault detection 95

4.3.3. RDP in fault diagnosis 100

4.4. Concluding remarks 102

Chapter 5. Model Reduction for Support Vector Machines 103

5.1. Introduction  103

5.2. Overview of model reduction 104

5.2.1. Wrapper methods 105

5.2.2. Filter methods 106

5.2.3. Embedded methods 107

5.3. Model reduction for energy consumption 108

5.3.1. Introduction 108

5.3.2. Algorithm 109

5.3.3. Feature set description 111

5.4. Model reduction for single building energy 112

5.4.1. Feature set selection  112

5.4.2. Evaluation in experiments  114

5.5. Model reduction for multiple buildings energy 116

5.6. Concluding remarks  119

Chapter 6. Parallel Computing for Support Vector Machines 121

6.1. Introduction  121

6.2. Overview of parallel support vector machines 122

6.3. Parallel quadratic problem solver  123

6.4. MPI-based parallel support vector machines  127

6.4.1. Message passing interface programming model  127

6.4.2. Pisvm  129

6.4.3. Psvm  130

6.5. MapReduce-based parallel support vector machines  130

6.5.1. MapReduce programming model  131

6.5.2. Caching technique 133

6.5.3. Sparse data representation 133

6.5.4. Comparison of MRPsvm with Pisvm  134

6.6. MapReduce-based parallel ε-support vector regression 138

6.6.1. Implementation aspects  138

6.6.2. Energy consumption datasets 139

6.6.3. Evaluation for building energy prediction  140

6.7. Concluding remarks  142

Summary and Future of Building Energy Analysis  145

Bibliography 149

Index 163